title: Cognition in the Wild

author: Hutchins, Edwin.

publisher: MIT Press
isbn10 | asin: 0262581469
print isbn13: 9780262581462
ebook isbn13: 9780585021867
language: English
 2/1540

Cognition--Social aspects--
Case studies, Cognition and
subject culture--Case studies,
Navigation--Psychological as-
pects, Psychology, Naval.
publication date: 1995
lcc: BF311.H88 1995eb
ddc: 153
Cognition--Social aspects--
Case studies, Cognition and
subject: culture--Case studies,
Navigation--Psychological as-
pects, Psychology, Naval.
Cognition in the Wild
Edwin Hutchins

The MIT Press

Cambridge, Massachusetts
London, England
Second printing, 1996

© 1995 Massachusetts Institute.of Technology

All rights reserved. No part of this book may be

reproduced in any form by any electronic or
mechanical means (including photocopying, re-
cording, or information storage and retrieval)
without permisson in writing from the publisher.

Set in Helvetica and Melior by Asco Trade

Typesetting Ltd., Hong Kong.
Printed and bound in the United States of
America.

Library of Congress Cataloging-in-Publication

Data
 5/1540

Hutchins, Edwin.
Cognition in the wild / Edwin Hutchins.
p. cm.
"A Bradford book."
Includes bibliographical references and index.
ISBN 0-262-08231-4 (HB), 0-262-58146-9 (PB)
1. Cognition-Social aspects-Case studies. 2.
Cognition and culture-Case studies.
3. Navigation-Psychological aspects. 4. Psycho-
logy, Naval. I. Title.
BF311.H88 1994
153dc20
82866990447203f57400372e62d765b5.gif
82866990447203f57400372e62d765b5.gif
94-21562
82866990447203f57400372e62d765b5.gif
82866990447203f57400372e62d765b5.gif
82866990447203f57400372e62d765b5.gifCIP
for Dona
Contents
Acknowledgements

Introduction

1 Welcome Aboard

2 Navigation as Computation

3 The Implementation of Contemporary Pilotage

4 The Organization of Team Performances

5 Communication

6 Navigation as a Context for Learning

7 Learning in Context

8 Organizational Learning
 8/1540

9 Cultural Cognition

References

Index
 Page ix

Acknowledgements
This book has been a long time in the making.
Its creation has been a widely distributed cognit-
ive process. I wish to thank first those who
provided me the opportunity to make the obser-
vations on which this work is based. I am grate-
ful to the crews of the Palau (a pseudonym) and
all the other ships I sailed upon. The command-
ing officer and the navigator of the Palau merit
special recognition for allowing me to work
aboard their ship. I am especially grateful to the
quartermaster chief and the men of the Palau's
Navigation Department for working with me
and sharing their working lives so generously.
Although I will not name them here or in the
text, they know who they are and I am grateful
to them.
 10/1540

James Tweedale, then Technical Director of the

Navy Personnel Research and Development
Center, generously supported the early phases of
the research as an independent research project.
Additional support was provided by the Office
of Naval Research's Division of Psychology and
Personnel Training under the guidance of Susan
Chipman and Michael Shafto. My supervisor
and colleague at NPRDC, James Hollan,
provided a great working environment for me
and helped me to organize my thinking in the
early stages. Barbara Morris and Michael
Goeller helped with the transcriptions and cod-
ing of the data. Colleen Siefert worked with me
as a postdoc, made observations on another ship,
and co-authored portions of the discussion of
learning from error.
 11/1540

I thank the John D. and Catherine T. MacArthur

Foundation for a five-year foundation fellow-
ship that permitted me to work on this material
when no suitable institutional setting existed.
Perhaps more important, the fellowship gave me
the courage to follow ideas that lay outside the
mainstream.

Over the years in which this work developed, I

profited from my involvement in the cognitive
science community at the University of Califor-
nia at San Diego. I am especially grateful to
Donald Norman, who shared many ideas with
me as we ran a research laboratory and taught
courses together. I am also grateful to
 Page x

Aaron Cicourel, Roy D'Andrade, Rik Belew,

Mike Cole, and Yrjö Engeström for helping me
think through these ideas.

The preparation of the book was facilitated by

the helpful comments of Bambi Schieffelin,
Jacques Theureau, Everett Palmer, Nick Flor,
and Christine Halverson.

My greatest debt is to my wife, Dona, who

provided encouragement, support, great meals,
and editorial assistance throughout the project.
 Page xi

Introduction
 14/1540

The seed from which this book grew was

planted in November 1980, when I spent most
of a day on the navigation bridge of a U.S. Navy
ship as it worked its way in from the open North
Pacific, through the Straits of Juan de Fuca, and
down Puget Sound to Seattle. I was aboard the
ship to study what the operators of its steam
propulsion plant knew and how they went about
knowing it. I had spent most of the preceding
week down in the bowels of the ship, observing
engineering operations and talking to the boiler
technicians and machinist's mates who inhabited
that hot, wet, noisy tangle of boilers, pumps, and
pipes called the engineering spaces. I'll admit to
having felt a little claustrophobic after all that
time spent below the water line, where there is
no night or day and the only evidence of being
at sea is the rhythmic tipping of the deck plates
and sloshing of water in the bilge below one's
feet as the ship rolls in the swell. A chief boiler
technician confided to me that in 21 years on
 15/1540

Navy ships he had never yet been on deck to ex-

perience either of those two most romantic sea-
faring events, a ship's arrival at or departure
from a port.

I resolved, therefore, to take my last few hours

aboard this ship on the navigation bridge, where
I could see out the windows or even go out on
the bridge wing to get a breath of cold fresh air.
My professional rationalization for being on the
bridge was that there I would be able to observe
the process that generates the flurry of engine
commands that always taxes the engineering
crew when the ship nears the dock. And I did
make a detailed record of all engine and helm
commands given in the 75 minutes from the
time the engines were first slowed until they
were secured-there were 61 in all. But what
really captured my attention was the work of the
navigation team.
 16/1540

Three and a half years later, the project that be-

came this book began in earnest. In the summer
of 1984, I was still working for the Navy Per-
sonnel Research and Development Center in
San Diego as a civilian scientist with the title
Personnel Research Psychologist. By then I had
participated in two successful and well-known
 Page xii

projects. With these successes came the freedom

to conduct an independent research project. I
was given carte blanche to study whatever I
thought was of most interest. I chose to study
what I was then calling naturally situated cogni-
tion. Having a research position in a Navy labor-
atory made it possible for me to gain access to
naval vessels, and my longtime love of naviga-
tion and experience as a racing yacht navigator
made it easy for me to choose navigation as an
activity to study afloat. I talked my way aboard
a ship and set up shop on the navigation bridge.
At the time, I really had no notion what an ideal
subject navigation would turn out to be. When I
began, I was thinking in terms of the naturally
situated cognition of individuals. It was only
after I completed my first study period at sea
that I realized the importance of the fact that
cognition was socially distributed.
 18/1540

A little earlier, I had been asked to write a book

describing what is in cognitive anthropology for
the rest of cognitive science. I began that pro-
ject, but after I became disillusioned with my
field I lost interest in it. The choice of naturally
situated cognition as a topic came from my
sense that it is what cognitive anthropology
really should have been about but largely had
not been. Clifford Geertz (1983) called for an
"outdoor psychology," but cognitive anthropo-
logy was unable or unwilling to be that. The re-
spondents may have been exotic, but the meth-
ods of investigation were largely borrowed from
the indoor techniques of psychology and lin-
guistics. When cognitive and symbolic anthro-
pology split off from social anthropology, in the
mid 1950s, they left society and practice behind.
 19/1540

As part of the cognitive revolution, cognitive

anthropology made two crucial steps. First, it
turned away from society by looking inward to
the knowledge an individual had to have to
function as a member of the culture. The ques-
tion became "What does a person have to
know?" The locus of knowledge was assumed to
be inside the individual. The methods of re-
search then available encouraged the analysis of
language. But knowledge expressed or express-
ible in language tends to be declarative know-
ledge. It is what people can say about what they
know. Skill went out the window of the "white
room." The second turn was away from practice.
In the quest to learn what people know, anthro-
pologists lost track both of how people go about
knowing what they know and of the contribution
of the environments in which the knowing is ac-
complished. Perhaps these narrowing assump-
tions were necessary to
20/1540
 Page xiii

get the project of cognitive anthropology off the

ground. I will argue that, now that we are under-
way as a discipline, we should revoke these as-
sumptions. They have become a burden, and
they prevent us from seeing the nature of human
cognition.
 22/1540

In particular, the ideational definition of culture

prevents us seeing that systems of socially dis-
tributed cognition may have interesting cognit-
ive properties of their own. In the history of an-
thropology, there is scarcely a more important
concept than the division of labor. In terms of
the energy budget of a human group and the ef-
ficiency with which a group exploits its physical
environment, social organizational factors often
produce group properties that differ consider-
ably from the properties of individuals. Clearly,
the same sorts of phenomena occur in the cog-
nitive domain. Depending on their organization,
groups must have cognitive properties that are
not predictable from a knowledge of the proper-
ties of the individuals in the group. The emphas-
is on finding and describing "knowledge struc-
tures" that are somewhere "inside" the individu-
al encourages us to overlook the fact that human
cognition is always situated in a complex so-
ciocultural world and cannot be unaffected by it.
 23/1540

Similar developments in the other behavioral

sciences during the cognitive revolution of the
late 1950s and the 1960s left a troubled legacy
in cognitive science. It is notoriously difficult to
generalize laboratory findings to real-world situ-
ations. The relationship between cognition seen
as a solitary mental activity and cognition seen
as an activity undertaken in social settings using
various kinds of tools is not at all clear.
 24/1540

This book is about softening some boundaries

that have been made rigid by previous ap-
proaches. It is about locating cognitive activity
in context, where context is not a fixed set of
surrounding conditions but a wider dynamical
process of which the cognition of an individual
is only a part. The boundaries to be softened or
dissolved have been erected, primarily for ana-
lytic convenience, in social space, in physical
space, and in time. Just as the construction of
these boundaries was driven by a particular the-
oretical perspective, their dissolution or soften-
ing is driven by a different perspectiveone that
arose of necessity when cognition was confron-
ted in the wild.

The phrase "cognition in the wild" refers to hu-

man cognition in its natural habitatthat is, to nat-
urally occurring culturally constituted human
activity. I do not intend "cognition in the wild"
to
25/1540
 Page xiv

be read as similar to Levi-Strauss's ''pensée

sauvage," nor do I intend it to contrast with Jack
Goody's (1977) notion of domesticated mind.
Instead, I have in mind the distinction between
the laboratory, where cognition is studied in
captivity, and the everyday world, where human
cognition adapts to its natural surroundings. I
hope to evoke with this metaphor a sense of an
ecology of thinking in which human cognition
interacts with an environment rich in organizing
resources.
 27/1540

The attempt is cultural in nature, giving recogni-

tion to the fact that human cognition differs
from the cognition of all other animals primarily
because it is intrinsically a cultural phenomen-
on. My aim is to provide better answers to ques-
tions like these: What do people use their cog-
nitive abilities for? What kinds of tasks do they
confront in the everyday world? Where shall we
look for explanations of human cognitive
accomplishment?
 28/1540

There is a common misconception among cog-

nitive scientists, especially those who do their
work in laboratory settings, that research con-
ducted outside the laboratory is necessarily "ap-
plied" work. I will argue in what follows that
there are many excellent reasons to look at the
"real world" that are not concerned with hoped-
for applications of the research findings (al-
though funding sponsors often like to think in
those terms). Pure research on the nature of real
cognitive practices is needed. In this book, I em-
phasize practice not in order to support a utilit-
arian or functionalist perspective but because it
is in real practice that culture is produced and
reproduced. In practice we see the connection
between history and the future and between cul-
tural structure and social structure. One of my
goals in writing this book is to make clear that
the findings of pure research on cognition in the
wild should change our ideas about the nature of
human cognition in general. This is not news to
 29/1540

anthropologists, who have been doing pure re-

search in the form of ethnography for decades.

This book is an attempt to put cognition back in-

to the social and cultural world. In doing this I
hope to show that human cognition is not just
influenced by culture and society, but that it is
in a very fundamental sense a cultural and social
process. To do this I will move the boundaries
of the cognitive unit of analysis out beyond the
skin of the individual person and treat the navig-
ation team as a cognitive and computational
system.

Chapter 1, "Welcome Aboard," attempts to loc-

ate the activity of ship navigation in the larger
world of modern life. It weaves to-
Page xv
 31/1540

gether three journeys: a movement through

physical space from the "street" to the ship, a
movement through social space from civilian to
military life, and a movement through conceptu-
al space from everyday notions of wayfinding to
the technical domain of navigation. Both the re-
searcher and the reader must make these jour-
neys to arrive at the activity of navigation as
practiced on the bridge of a Navy ship. Military
ranks and the ways in which military identities
are formed are presented here because these
things affect individual's relationships to their
work. An important aspect of the larger unit is
that it contains computational elements
(persons) who cannot be described entirely in
computational terms. Who they talk to and how
they talk to one another depend on these social
organizational factors. This chapter also con-
tains a discussion of the relationship of the re-
searcher to the activity under study. (The name
of the ship and the names of all the individuals
 32/1540

mentioned in the book are pseudonyms. All the

discourses reported, whether standing alone in
transcript form or embedded in narrative pas-
sages were transcribed directly from audio re-
cordings of actual events.)
 33/1540

Having taken navigation as it is performed by a

team on the bridge of a ship as the unit of cog-
nitive analysis, I attempt in chapter 2, "Naviga-
tion as Computation," to apply the principal
metaphor of cognitive sciencecognition as com-
putationto the operation of this system. I should
note here that in doing so, I do not make any
special commitment to the nature of the compu-
tations that are going on inside individuals ex-
cept to say that whatever happens there is part of
a larger computational system. This chapter de-
scribes the application of David Marr's notions
of levels of analysis of cognitive systems to the
navigation task and shows that, at the computa-
tional level, it is possible to give a single de-
scription of the computational constraints of all
known technical forms of human navigation. A
comparison of modern Western navigation with
navigation as practiced in Micronesia shows that
considerable differences between these tradi-
tions lie at the representational/algorithmic level
 34/1540

and at the implementational level. A brief his-

torical review of the development of modern
navigation shows that the representational and
implementational details of contemporary prac-
tice are contingent on complex historical pro-
cesses and that the accumulation of structure in
the tools of the trade is itself a cognitive
process.

Chapters 3-5 explore the computational and

cognitive properties of systems that are larger
than an individual. The issues addressed
 Page xvi

in these chapters concern how these larger sys-

tems operate and how their cognitive properties
are produced by interactions among their parts.
 36/1540

Chapter 3, "The Implementation of Contempor-

ary Pilotage," describes the physical structures
in which the navigation computations are imple-
mented. This chapter elaborates a conception of
computation as the propagation of representa-
tional state across a variety of media. This view
of computation permits the use of a single lan-
guage of description to cover cognitive and
computational processes that lie inside and out-
side the heads of the practitioners of navigation.
The first section of this chapter describes the
"fix cycle" as a cognitive process. The second
section describes how navigation tools are used
and how local functional systems composed of a
person in interaction with a tool have cognitive
properties that are radically different from the
cognitive properties of the person alone. The
third section discusses the ways in which the
computational activity can be distributed
through time by precomputing not only partial
results but also the means of computation. I
 37/1540

show here how the environments of human

thinking are not "natural" environments. They
are artificial through and through. Humans cre-
ate their cognitive powers by creating the envir-
onments in which they exercise those powers.
This chapter concludes with a discussion of the
relationship between the cognitive properties of
the individuals performing a task and the cognit-
ive properties of the system in which they
participate.
 38/1540

Chapter 4, "The Organization of Team Perform-

ances," moves the boundaries of the unit of ana-
lysis even further out to consider the cognitive
properties of the team as a whole. Here I note
some of the problems that are encountered when
cognitive activities are distributed across the
members of a group. It is not the case that two
or more heads are always better than one. This
chapter describes the structures and processes
involved in the group performance of the navig-
ation task. The first section follows through on
the application of Marr's concepts of computa-
tion to the navigation activity and discusses the
properties of the activity as an explicitly compu-
tational system. The second section presents a
problem in work organization encountered by
the navigation team and shows why it is often
difficult to apply the concepts that organize indi-
vidual action to the organization of group ac-
tion. The final section shows how the members
of the navigation team form a flexible
 39/1540

connective tissue that maintains the propagation

of representational state in the face of a range of
potentially disruptive events.
 Page xvii

Chapter 5, "Communication," continues the

theme of chapter 4 but looks at communication
in more detail. It asks: How is it that patterns of
communication could produce particular cognit-
ive properties in a group? The chapter begins
with a discussion of features of communication
observed in the navigation team and their effects
on the Team's computational properties. These
observations lead to some simple hypotheses
about the ways in which patterns of communica-
tion might affect the computational properties of
a group. These hypotheses are explored using a
computer simulation of communities of connec-
tionist networks. The simulations lead to the
surprising conclusion that more communication
is not always better.

Chapters 6-8 concern learning or change in the

organization of cognitive systems at several
scales.
 41/1540

Chapter 6, "The Context of Learning," is a

bridge between the descriptions of ongoing op-
erations provided by the previous chapters and
the descriptions of changes in the nature of on-
going operations provided by the following
chapters. It describes the context in which
novice navigators become experts. This chapter
is an attempt to examine both the work that the
system does in order to scaffold learning by
practitioners and the opportunities for the devel-
opment of new knowledge in the context of
practice.
 42/1540

Whereas in chapter 6 I deal with the observable

contexts surrounding learning, in chapter 7,
"Learning in Context," I try to dissolve the
boundaries of the skin and present navigation
work as a system of interactions among media
both inside and outside the individual. I look at
learning or conceptual change as a kind of ad-
aptation in a larger dynamical system. This
chapter presents a functional notation and a
framework for thinking about learning as local
adaptation in a dynamic system of coordinations
of representational media.
 43/1540

Chapter 8, "Organizational Learning," returns

the focus to the larger unit of analysis: the team
as a whole. It presents a case study of an incid-
ent in which the navigation team was forced to
adapt to changes in its information environment.
The analysis presented here examines a particu-
lar incident in which the microstructure of the
development of the navigation practice can be
seen clearly. It is an attempt to show the details
of the kinds of processes that must be the en-
gines of cultural change.

Chapter 9, "Cultural Cognition," attempts to pull

the preceding chapters together into a coherent
argument about the relationships of culture and
cognition as they occur in the wild. I attempt
first
 Page xviii

to illustrate the costs of ignoring the cultural

nature of cognition. I argue that a new frame-
work is needed to understand what is most char-
acteristically human about human cognition. In
order to construct a new framework, the old one
must be deconstructed. I therefore provide two
readings of the history of cognitive science: a
history as seen by the proponents of the cur-
rently dominant paradigm and a rereading of the
history of cognitive science from a sociocultural
perspective. The differences between these two
readings highlight a number of problems in con-
temporary cognitive science and give new
meanings to some of the familiar events in its
history.
Cognition in the Wild
 Page 1

1 Welcome Aboard

Narrative: A Crisis
 47/1540

After several days at sea, the U.S.S. Palau was

returning to port, making approximately 10
knots in the narrow channel between Ballast
Point and North Island at the entrance to San
Diego Harbor. In the pilothouse or navigation
bridge, two decks above the flight deck, a junior
officer had the conn (i.e., was directing the
steering of the ship), under the supervision of
the navigator. The captain sat quietly in his chair
on the port side of the pilothouse watching the
work of the bridge team. Morale in the pilot-
house had sagged during two frustrating hours
of engineering drills conducted just outside the
mouth of the harbor but was on the rise now that
the ship was headed toward the pier. Some of
the crew talked about where they should go for
dinner ashore and joked about going all the way
to the pier at 15 knots so they could get off the
ship before nightfall.
 48/1540

The bearing recorder had just given the com-

mand "Stand by to mark time 3 8" and the fatho-
meter operator was reporting the depth of the
water under the ship when the intercom erupted
with the voice of the engineer of the watch:
"Bridge, Main Control. I am losing steam drum
pressure. No apparent cause. I'm shutting my
throttles." Moving quickly to the intercom, the
conning officer acknowledged: "Shutting
throttles, aye." The navigator moved to the
captain's chair, repeating: "Captain, the engineer
is losing steam on the boiler for no apparent
cause." Possibly because he realized that the
loss of steam might affect the steering of the
ship, the conning officer ordered the rudder
amidships. As the helmsman spun the wheel to
bring the rudder angle indicator to the center-
line, he answered the conning officer: "Rudder
amidships, aye sir." The captain began to speak,
saying "Notify," but the engineer was back on
the intercom, alarm in his voice this time,
 49/1540

speaking rapidly, almost shouting: "Bridge,

Main Control, I'm going to secure number two
boiler at this time. Recommend you drop the an-
chor!" The captain had been stopped in mid-sen-
tence by the blaring intercom, but before the en-
gineer could finish speaking the captain said, in
a loud but cool voice, "Notify the bosun." It is
standard procedure on
 Page 2

large ships to have an anchor prepared to drop in

case the ship loses its ability to maneuver while
in restricted waters. With the propulsion plant
out, the bosun, who was standing by with a crew
forward ready to drop the anchor, was notified
that he might be called into action. The falling
intonation of the captain's command gave it a
cast of resignation or perhaps boredom and
made it sound entirely routine.

In fact, the situation was anything but routine.

The occasional cracking voice, a muttered curse,
or a perspiration-soaked shirt on this cool spring
afternoon told the real story: the Palau was not
fully under control, and careers and possibly
lives were in jeopardy.
 51/1540

The immediate consequences of this event were

potentially grave. Despite the crew's correct re-
sponses, the loss of main steam put the ship in
danger. Without steam, it could not reverse its
propellerthe only way to slow a large ship effi-
ciently. The friction of the water on the ship's
hull will eventually reduce its speed, but the
Palau would coast for several miles before com-
ing to a stop. The engineering officer's recom-
mendation that the anchor be dropped was not
appropriate. Since the ship was still traveling at
a high rate of speed, the only viable option was
to attempt to keep the ship in the deep water of
the channel and coast until it had lost enough
speed to safely drop anchor.
 52/1540

Within 40 seconds of the report of loss of steam

pressure, the steam drum was exhausted. All
steam-turbine-operated machinery came to a
halt, including the turbine generators that pro-
duce the ship's electrical power. All electrical
power was lost throughout the ship, and all elec-
trical devices without emergency power backup
ceased to operate. In the pilothouse a high-
pitched alarm sounded for a few seconds, sig-
naling an under-voltage condition for one piece
of equipment. Then the pilothouse fell eerily si-
lent as the electric motors in the radars and other
devices spun down and stopped. Just outside the
navigation bridge, the port wing pelorus operat-
or watched the gyrocompass card in his pelorus
swing wildly and then return to its original
heading. He called in to the bearing recorder
standing at the chart table: "John, this gyro just
went nuts." The bearing recorder acknowledged
the comment and told the pelorus operator that a
 53/1540

breakdown was in progress: "Yeah, I know, I

know, we're havin' a casualty."

Because the main steering gear is operated with

electric motors, the ship now not only had no
way to arrest its still-considerable
 Page 3

forward motion; it also had no way to quickly

change the angle of its rudder. The helm does
have a manual backup system, located in a com-
partment called aftersteering in the stern of the
ship: a wormgear mechanism powered by two
men on bicycle cranks. However, even strong
men working hard with this mechanism can
change the angle of the massive rudder only
very slowly.
 55/1540

Shortly after the loss of power, the captain said

to the navigator, who was the most experienced
conning officer on board, "OK, Gator, I'd like
you to take the conn." The navigator answered
"Aye, sir" and, turning away from the captain,
announced: "Attention in the pilothouse. This is
the navigator. I have the conn." As required, the
quartermaster of the watch acknowledged
(''Quartermaster, aye") and the helmsman repor-
ted "Sir, my rudder is amidships." The navigator
had been looking out over the bow of the ship,
trying to detect any turning motion. He
answered the helmsman: "Very well. Right 5 de-
grees rudder." Before the helmsman could reply,
the navigator increased the ordered angle: "In-
crease your rudder right 10 degrees." (The rud-
der angle indicator on the helm station has two
parts; one shows the rudder angle that is ordered
and the other the actual angle of the rudder.)
The helmsman spun the wheel, causing the in-
dicator of the desired rudder angle to move to
 56/1540

the right 10 degrees, but the indicator of the ac-

tual rudder angle seemed not to move at all.
"Sir, I have no helm sir!" he reported.
 57/1540

Meanwhile, the men on the cranks in aftersteer-

ing were straining to move the rudder to the de-
sired angle. Without direct helm control, the
conning officer acknowledged the helmsman's
report and sought to make contact with after-
steering by way of one of the phone talkers on
the bridge: "Very well. Aftersteering, Bridge."
The navigator then turned to the helmsman and
said "Let me know if you get it back." Before he
could finish his sentence, the helmsman respon-
ded, "I have it back, sir." When the navigator ac-
knowledged the report, the ship was on the right
side of the channel but heading far to the left of
the desired course. "Very well, increase your
rudder to right 15." "Aye sir. My rudder is right
15 degrees. No new course given." The navigat-
or acknowledged"Very well"and then, looking
out over the bow, whispered "Come on, damn it,
swing!" Just then, the starboard wing pelorus
operator spoke on the phone circuit: "John, it
looks like we're gonna hit this buoy over here."
 58/1540

The bearing recorder had been concentrating on

the chart and hadn't quite heard. "Say again" he
requested. The starboard wing pelorus operator
leaned over the railing of his platform to
Page 4
 60/1540

watch the buoy pass beneath him. It moved

quickly down the side of the ship, staying just a
few feet from the hull. When it appeared that the
Palau would not hit the buoy, the starboard
wing pelorus operator said "Nothin' "; that
ended the conversation. The men inside never
knew how close they had come. Several sub-
sequent helm commands were answered with
"Sir, I have no helm." When asked by the cap-
tain how he was doing, the navigator, referring
to their common background as helicopter pi-
lots, quipped "First time I ever dead-sticked a
ship, captain.'' (To "dead-stick" an aircraft is to
fly it after the engine has died.) Steering a ship
requires fine judgements of the ship's angular
velocity. Even if helm response was instantan-
eous, there would still be a considerable lag
between the time a helm command was given
and the time when the ship's response to the
changed rudder angle was first detectable as the
movement of the bow with respect to objects in
 61/1540

the distance. Operating with this manual system,

the navigator did not always know what the ac-
tual rudder angle was, and could not know how
long to expect to wait to see if the ordered com-
mand was having the desired effect. Because of
the slowed response time of the rudder, the nav-
igator ordered more extreme rudder angles than
usual, causing the Palau to weave erratically
from one side of the channel to the other.
 62/1540

Within 3 minutes, the diesel-powered emer-

gency generators were brought on line and elec-
trical power was restored to vital systems
throughout the ship. Control of the rudder was
partially restored, but remained intermittent for
an additional 4 minutes. Although the ship still
could not control its speed, it could at least now
keep itself in the dredged portion of the narrow
channel. On the basis of the slowing over the
first 15 minutes after the casualty, it became
possible to estimate when and where the Palau
would be moving slowly enough to drop anchor.
The navigator conned the ship toward the
chosen spot.
 63/1540

About 500 yards short of the intended anchor-

age, a sailboat took a course that would lead it to
cross close in front of the Palau. Normally the
Palau would have sounded five blasts with its
enormous horn to indicate disagreement with
the actions taken by the other vessel. However,
the Palau's horn is a steam whistle, and without
steam pressure it will not sound. The Navigation
Department has among its equipment a small
manual foghorn, basically a bicycle pump with a
reed and a bell. The navigator remembered
Page 5
 65/1540

this piece of gear and instructed the keeper of

the deck log to leave his post, find the manual
horn, descend two levels to the flight deck, take
the horn out to the bow, and sound the five
warning blasts. The keeper of the deck log ran
from the pilothouse, carrying a walkie-talkie to
maintain communication with the bridge. The
captain grabbed the microphone for the flight
deck's public address system and asked "Can
you hear me on the flight deck?" Men below on
the deck turned and waved up at the pilothouse.
"Sailboat crossing Palau's bow be advised that I
am not ... I have no power. You cross at your
own risk. I have no power." By this time, the
hull of the sailboat had disappeared under the
bow of the ship and only its sails were visible
from the pilothouse. In the foreground, the men
on the flight deck were now running to the bow
to watch the impending collision. Meanwhile,
the keeper of the deck log had run down two
flights of stairs, emerged from the base of the
 66/1540

island, and begun sprinting across the nearly

100 yards that lay between the island and the
bow. Before he was halfway to his goal, it was
clear that by the time he would reach the bow
the signal from the horn would be meaningless.
The navigator turned to a junior officer who was
holding a walkie-talkie and exclaimed "Just tell
him to put the sucker down and hit it five
times!" The message was passed, and the five
feeble blasts were sounded from the middle of
the flight deck. There is no way to know wheth-
er the signal was heard by the sailboat, which by
then was directly ahead of the Palau and so
close that only the tip of its mast was visible
from the pilothouse. A few seconds later, the
sailboat emerged, still sailing, from under the
starboard bow. The keeper of the deck log con-
tinued to the bow to take up a position there in
case other warnings were required.
 67/1540

Twenty-five minutes after the engineering casu-

alty and more than 2 miles from where the wild
ride had begun, the Palau was brought to anchor
at the intended location in ample water just out-
side the bounds of the navigation channel.

The safe arrival of the Palau at anchor was due

in large part to the exceptional seamanship of
the bridge crew, especially the navigator. But no
single individual on the bridge acting alone-
neither the captain nor the navigator nor the
quartermaster chief supervising the navigation
team-could have kept control of the ship and
brought it safely to anchor. Many kinds of think-
ing were required to perform this task. Some of
them were happening in
 Page 6

parallel, some in coordination with others, some

inside the heads of individuals, and some quite
clearly both inside and outside the heads of the
participants.

This book is about the above event and about

the kind of system in which it took place. It is
about human cognitionespecially human cogni-
tion in settings like this one, where the problems
that individuals confront and the means of solv-
ing them are culturally structured and where no
individual acting alone is entirely responsible
for the outcomes that are meaningful to the soci-
ety at large.
 69/1540

Gaining access to this field site required me, as

an ethnographer, to make three journeys at once.
In this first chapter I will try to weave them to-
gether, for the reader will also have to make
these journeys mentally in order to understand
the world of military ship navigation. The first is
a journey through physical space from my home
and my usual workplace to the navigation bridge
of the Palau. This journey took me through
many gates, as I moved from the street to the
military base, to the ship, and within the ship to
the navigation bridge. I will try to convey the
spatial organization of the setting in which nav-
igation is performed. The second journey is a
trip through social space in which I moved from
the civilian social world past the ship's official
gatekeepers into the social organization of the
Navy, and then to the ship's Navigation Depart-
ment. This journey closely parallels the journey
through physical space because space is so often
used as an element of social organization. As the
 70/1540

spatial journey took me to regions with narrower

and narrower boundaries, so the social journey
leads us through successively narrower levels of
social organization. The third journey is a move-
ment through conceptual space, from the world
of everyday spatial cognition into the technical
world of navigation. This third journey does not
really begin until I near the end of the other two.

Through the Main Gate

A crisp salute from a young marine in dress uni-

form at the main gate's guard shack marked the
transition from the "street" to the "base"from the
civilian realm to the military. The base is a place
of close-cropped haircuts and close-cropped
lawns. Here nature and the human form are con-
trolled, arranged, disciplined, ready to make a
good impression. In boot camp inductee's credo
is: "If it
 Page 7

moves, salute it. If it doesn't move, pick it up. If

you can't pick it up, paint it white." The same
mindset imposes an orderliness and a predictab-
ility on both the physical space and the social
world of the military base.
 72/1540

As a civilian employee of the Navy, I was en-

couraged to occasionally ride a ship in order to
better understand the nature of the "operational"
world. But being encouraged by my own organ-
ization to ride a ship and being welcomed by the
crew are two different things. From the per-
spectives of the people running a ship, there
may be little to gain from permitting a civilian
on board. Civilians, who are often ignorant of
shipboard conventions, may require some tend-
ing to keep them out of trouble. They take up
living space, which on many ships is at a premi-
um, and if they do not have appropriate security
clearances they may have to be escorted at all
times.

The Ship
 73/1540

The Palau is an amphibious helicopter transport.

Its warfare mission is to transport marines
across the seas and then deliver them to the bat-
tlefields in the 25 helicopters that are carried on
board. The helicopters also bring troops back to
the ship, which has a small hospital and a com-
plete operating theater. Ships of this class are of-
ten mistaken for true aircraft carriers of the sort
that carry jet planes. As is the case with true air-
craft carriers, the hull is capped by a large flat
flight deck which creates an overhang on all
sides of the ship. But this flight deck is only 592
feet long, just over half the length of a carrier
deck and much too small to handle fixed-wing
jets. About halfway between the bow and the
stern, jutting up out of the smooth expanse of
the flight deck on the starboard rail, stands a
four-story structure called the island. The island
occupies the rightmost 20 feet of the flight deck,
which is about 100 feet wide. The ship extends
28 feet below the surface of the water and
 74/1540

weighs 17,000 tons empty. It is pushed through

the water by a single propeller driven by a
22,000-horsepower steam turbine engine.

Originally, the ships of the Palau's class were

planned to have been almost 200 feet longer and
to have two propulsion plants and two pro-
pellers. However, budget cuts in the early 1960s
led to a hasty redesign. In the original design,
the off-center weight of the steel island was to
be balanced by the second propulsion plant.
 Page 8

Unfortunately, the redesign failed to take into

account the decrease in righting moment caused
by the deletion of the second engine. When the
hull that is now the Palau was launched, it cap-
sized! It was refloated, and the steel island was
replaced with an aluminum one. The ship was
renamed and put into service. The aluminum is-
land is attached to the steel deck with steel bolts.
In a wet and salty environment, this forms an
electrolyte that causes corrosion of the attach-
ment points between the island and the deck.
There is a standing joke among those who work
in the island that someday, in a big beam swell,
the ship will roll to starboard and the island will
simply topple off the deck into the sea.
 76/1540

Two levels above the flight deck in the island is

the navigation bridge. Also in the island are the
air operations office, from which the helicopters
are controlled, and a flag bridge where an ad-
miral and his staff can work. The top of the is-
land bristles with radar antennae.

The Gator Navy and the Other Navies

When I first went aboard the Palau it was tied

up at pier 4 with several other amphibious ships.
A frigate and a destroyer were tied up to an ad-
jacent pier, but they are part of another navy
within the Navy. Membership in these navies is
an important component of naval identity.
 77/1540

Troop transport is not considered a glamorous

job in the Navy. The Palau is part of what is
called the amphibious fleet, the portion of the
fleet that delivers marines to battlegrounds on
land. The amphibious fleet is also known some-
what derogatorily as the "gator navy." The nick-
name is apparently derived from a reference to
that amphibious reptile, the alligator. While the
alligator is not a prototypical amphibian, its ag-
gressiveness may be important in Navy culture;
"salamander navy" or "frog navy" might be too
disparaging.
 78/1540

The aviation community (the "airdales") claims

to be the highest-status branch of the Navy.
Most others would say that the submarine fleet
(the "nukes") comes next, although the submar-
iners consider themselves a breed apart. (They
have a saying that there are only two kinds of
ships in the navy: submarines and targets.) Then
comes the surface fleet (the "black shoes").
Within each of these groups are subgroupings,
which are also ranked. In the sur-
 Page 9

face fleet the ranking descends from surface

combatants (cruisers, destroyers, and frigates) to
aircraft carriers, then the amphibious fleet, and
finally tenders and supply ships.

While from the civilian point of view a sailor

may be a sailor, in the Navy these distinctions
mark important subcultural identities. The per-
ceived differences are based on many factors,
including the "glamor" of the expected mission,
the sophistication of the equipment, the destruct-
ive potential, the stringency of requirements for
entry into each area, the quality and extent of
the training provided to the members of each
community, and the general sense of the quality
of the people involved. For a surface warfare of-
ficer who hopes to make a career out of the
Navy and rise to a high rank, it is not good to be
assigned to an amphibious ship for too long.
 80/1540

Ships that carry aircraft and air crewmen present

a special situation with respect to these groups.
Because they have aircraft they have members
of the aviation community aboard, but because
they are ships they must have members of the
surface community aboard. The commanding
officer of an aircraft carrier is always a member
of the air communitya measure of the notion in
the navy that the air wing is the raison d'être of
a ship that carries aircraft. The friction between
the air community and the surface community
may be manifested in subtle and not-so-subtle
ways. If members of the air community account
for the majority of the highranking positions on
a ship, junior surface warfare officers may com-
plain that junior "airdales" are given more op-
portunities for qualification and advancement.
An amphibious transport with an air wing is an
even more complicated situation. Here members
of the surface and air groups interact. And when
marines are aboard an amphibious ship, there is
 81/1540

also sometimes friction between the sailors and

the marines.

These patterns of differentiation are present at

all levels of organization in the military, from
the broadest of interservice rivalries to distinc-
tions between the occupants of adjacent spaces
on the ship. Such effects are present to some de-
gree in many social organizations, but they are
highly elaborated in the military. Much of the
establishment of identity is expressed in propos-
itions like this: "We are the fighting X's. We are
proud of what we are and what we do. We are
unlike any other group." The unspoken infer-
ence is "If you do something else, you cannot be
quite as good as we are." Identities are also
signaled by insignia and emblems of various
 Page 10

kinds. In the officer ranks, breast insignia denote

which navy one is in. Aviators wear wings, sub-
mariners wear dolphins, surface warfare officers
wear cutlasses.

Within each part of the surface fleet, there are

strong identities associated with specific ships.
Ships have stirring nationalistic or patriotic mot-
toes, which are often inscribed on plaques, base-
ball caps, t-shirts, and coffee mugs. Many ships
produce yearbooks. The bond among shipmates
is strongest when they are off ship. There is less
of an identification with the class of one's ship,
but some classes of ship are considered more
advanced (less obsolete) and more glamorous
than others.
 83/1540

The military institutionalizes competition at all

levels of organization. Individuals compete with
one another, and teams of individuals are pitted
against other teams. Ships compete in exercises,
and branches of the military compete for fund-
ing and the opportunity to participate in combat.
Aboard a ship this competitiveness manifests it-
self in a general opinion that "we in our space
know what we are doing, but the people just on
the other side of the bulkhead do not." These
sentiments can arise in situations where the suc-
cessful completion of some task relies on co-
operation between individuals in different
spaces. Sometimes the larger system may fail
for reasons having to do with the interactions of
the units rather than with any particular unit;
still, each unit needs to attach blame some-
where, and the alleged incompetence of some
other unit is the easiest and most understandable
explanation.
 84/1540

Across the Brow

A sailor standing outside a guard shack glances

at the identification badge of each person
passing onto the pier. Walking onto a pier
between two ships of the Palau's class is like
walking into a deep canyon with overhanging
gray walls and a dirty concrete floor. The
canyon is vaguely threatening. It is noisy, and
the hulls of the ships seem to box in the whine
of motors and the hiss of compressed air. There
are trucks and cranes on the pier, and cables are
strewn across the pier and suspended in space
over the narrow band of greenish water between
the pier and the hulls. Floating in the water
between each ship and the pier are several crude
rafts called "camels" and a work barge. The
camels keep the hull of the ship far enough
away from the pier so that the broad flight deck
flaring out at the top of the hull does not over-
hang the pier.
85/1540
 Page 11

To board the Palau, I climbed a sort of scaffold

up a few flights of gray metal stairs to a gang-
plank (in Navy parlance, the brow) that reached
from the top of the scaffold to a huge hole in the
side of the ship. The hole was at the level of the
hangar deck (also called the main deck), still
several levels below the flight deck. At the top
of the brow was a security desk where the of-
ficer of the deck (OOD) checked the identifica-
tion cards of sailors departing from and return-
ing to the ship. Sailors stepping aboard turned to
face the stern of the ship, came to attention and
saluted the ship's ensign (flag), which flew on a
staff over the fantail and was thus not visible
from the brow.
 87/1540

Before visiting the ship, I had been given the

NPRDC Fleet visitor's guide of basic informa-
tion, which included the following instructions
for proper performance of the boarding ritual:
"At the top of the brow or accommodation lad-
der, face aft toward the colors (national ensign)
and pause at attention. Then turn to the OOD,
pause briefly at attention, and say, 'Request per-
mission to come aboard, Sir.' State your name,
where you are from, the purpose of your visit
and the person you wish to see." This little ritual
is a symbolic pledge of allegiance to the ship be-
fore boarding. Visitors to the ship wait in limbo
at the security desk, neither ashore nor officially
aboard, while word of their arrival is sent to
their onboard host. The actual permission to go
aboard must have been arranged in advance.
 88/1540

The ship's official gatekeeper is normally the ex-

ecutive officer (abbreviated XO). The com-
manding officer, the executive officer, and the
department heads form the primary administrat-
ive structure of the ship. Every ship in the Navy
is organized into a number of departments. Each
department is supervised by an officer. In large
departments, the department head may supervise
less senior officers, who in turn supervise the
enlisted personnel who do virtually all the actual
work on the ship. Before embarking, I was re-
quired to convince the XO that I had something
to offer the navy and that I would not cause un-
due aggravation while aboard. In a brief and
somewhat discouraging interview with the XO,
it was agreed that if the navigator was willing to
tolerate my presence in his department, I could
come aboard and work with the navigation
team.
 89/1540

After getting past the XO, I made a date to have

lunch with the navigator. I met him in the
officer's dining area (the wardroom), and during
our discussion we discovered a shared past.
While a cadet at
 Page 12

the Naval Academy, the navigator had served as

racing tactician aboard a particular racing sloop
that had been donated to the academy. The sloop
was subsequently sold to a friend of mine, and I
had sailed aboard it as navigator and racing tac-
tician for 8 years. The discovery of this ex-
traordinary coincidence helped cement our
friendship and secured the navigator's permis-
sion for my work aboard the Palau. With my
prearranged permission to sail, and with the
navigator's blessing, I waited at the security
desk.
 91/1540

An escort at the security desk and led me

through the huge dark cavern of the hangar
deck. We detoured around several parked heli-
copters and skirted forklifts and pallets of mater-
ials. We ducked through a hatch in the wall of
the hangar deck and began the climb up a series
of narrow steep ladders to the navigation bridge.
(On a ship, floors are called decks, walls are
called bulkheads or partitions, corridors are
called passageways, ceilings are called over-
heads, and stairs are called ladders.)

Reconciling the Chart and the World

Navigation is a collection of techniques for an-

swering a small number of questions, perhaps
the most central of which is "Where am I?"
 92/1540

What does the word 'where' mean in this ques-

tion? When we say or understand or think where
we are, we do so in terms of some representa-
tion of possible positions. "Where am I?" is a
question about correspondences between the
surrounding world and some representation of
that world.
 93/1540

Where am I right now as I write this? I am at my

desk, in my study. The window in front of me
faces the garden; the door over there leads to the
hallway that leads to the remainder of the house.
My house is on the Pacific coast, north of the
university. I'm on the western edge of the North
American continent. I'm on the planet Earth
circling a minor star in the outer portion of an
arm of a spiral galaxy. In every one of these de-
scriptions, there is a representation of space as-
sumed. Each of these descriptions of my loca-
tion has meaning only by virtue of the relation-
ships between the location described and other
locations in the representation of space implied
by the description. This is an absolutely funda-
mental problem that must be solved by all mo-
bile organisms.

Whether the map is internal or external, whether

it is a mental image of surrounding space (on
whatever scale and in whatever
94/1540
Page 13
 96/1540

terms) or a symbolic description of the space on

a piece of paper, I must establish the corres-
pondence of map and territory in order to an-
swer the question "Where am I?" One of the
most exciting moments in navigation is making
a landfall on an unfamiliar coast. If I am making
a landfall on a high island or a mountainous
coast, as I approach the land, I first see just the
tops of mountains, then I see the lower slopes,
then the hills, and finally the features on the
shoreline itself. Now, where am I? Turning to
my chart, I see that I had hoped to meet the
coast just to the south of a major headland. Per-
haps that big hill I can see across the water on
the left is that headland. And perhaps that high
peak off in the haze, inland, is this peak shown
on the chart. Hmm, according to the chart it is
only supposed to be 500 meters high. It seems
far away and higher than that. Perhaps it is
something else, something too far inland to be
printed on the chart.
 97/1540

Through considerations like these, a navigator

attempts to establish a coherent set of corres-
pondences between what is visible in the world
and what is depicted on a chart. Some charts
even provide small profiles showing the appear-
ance of prominent landmarks from particular
sea-level vantage points. The same sort of task
confronts any of us when, for example, we walk
out of the back door of a theater onto an unfa-
miliar street. Which way am I facing? Where am
I? The question is answered by establishing cor-
respondences between the features of the envir-
onment and the features of some representation
of that environment. When the navigator is sat-
isfied that he has arrived at a coherent set of cor-
respondences, he might look to the chart and say
"Ah, yes; I am here, off this point of land." Now
the navigator knows where he is. And it is in
this sense that most of us feel we know where
we are. We feel that we have achieved a recon-
ciliation between the features we see in our
 98/1540

world and a representation of that world. Things

are not out of place. They are where we expect
them to be. But now suppose someone asks a
navigator "How far are we from the town at the
head of that bay?" To answer that question,
simply having a good sense of the correspond-
ences between what one sees and what is depic-
ted on some representation of the local space is
not enough. Now more precision is required. To
answer that question the navigator needs to have
a more exact determination of where he is. In
particular, he needs to have a sense of his loca-
tion on a representation of space in a form that
will permit him to compute the answer to the
question. This is position fixing. It is what one
does
 Page 14

when just having a sense of reconciliation

between the territory and the map is not enough.

Up the Ladder
 100/1540

From the hangar deck the escort led the way up

three steep ladders in a narrow stairwell filled
with fluorescent light, stale air, and the clang of
hard shoes on metal steps. The decks of a ship
are numbered starting with the main deck. On
most ships, the main deck is defined as the ''up-
permost deck that runs the length of the ship."
On ships that have a flight deck above a hangar
deck (this includes aircraft carriers and amphibi-
ous helicopter transports such as the Palau) the
hangar deck is the main deck. Immediately be-
low the main deck is the second deck, and be-
low that the third deck, and so on down to the
hold. Above the main deck, the decks are desig-
nated "levels" and are numbered 01, 02,..., in-
creasing in number with altitude. We stopped
periodically on deck platforms to allow sailors
going down to pass. Foot traffic on ships gener-
ally moves up and forward on the starboard side
and down and aft on the port side. However, the
layout of the hangar deck limits the number and
 101/1540

location of ladders, and in order to shorten the

route my escort was taking me against the
traffic. We climbed into a small busy foyer, and
through an open hatch I caught a breath of fresh
air and a glimpse of the flight deck in the sun.
Men in overalls were working on the hot, rough
black surface. We continued upward, now
climbing inside the narrow island. One ladder
pitch above the flight deck we came to the 04
level. The door leading to the flag bridge, where
an admiral and his staff would work, was
chained and padlocked. One more ladder
brought us to the 05 level.

Military Identities
 102/1540

The men and women in the military are divided

into two broad social classes: officer and enlis-
ted. An officer must have a college degree and
is commissioned (authorized to act in com-
mand). In the Navy, members of both classes
believe in the reality of differences between of-
ficers and enlisted personnel. The lowest-rank-
ing officer is superior in the command structure
to the highest-ranking enlisted person. The dis-
tinction between officers and enlisted is marked
by uniforms, by insignia, and by a complex set
of rituals. The simplest of these rituals is the sa-
lute, of course, but the
 Page 15

courtesies to be extended by enlisted to officers

include clearing a passageway on the approach
of an officer and refraining from overtaking an
officer on foot until permission has been
granted.

Enlisted Rates and Ratings

 104/1540

Enlisted personnel are classified according to

pay grade (called rate) and technical specializa-
tion (called rating). As Bearden and Wedertz
(1978) explain: "A rating is a Navy joba duty
calling for certain skills and attitudes. The rating
of engineman, for example, calls for persons
who are good with their hands and are mechan-
ically inclined. A paygrade (such as E-4, E-5,
E-6) within a rating is called a rate. Thus an en-
gineman third class (EN3) would have a rating
of engineman, and a rate of third class petty of-
ficer. The term petty officer (PO) applies to any-
one in paygrades E-4 through E-9. E-ls through
E-3s are called non-rated personnel."
 105/1540

The enlisted naval career begins with what is

basically a socialization period in which the re-
cruit is indoctrinated into basic military policy
and acquires the fundamental skills of a sailor.
The rates through which a recruit passes in this
phase are seaman recruit, seaman apprentice,
and able-bodied seaman. Once socialized, a sea-
man learns the skills of a particular job special-
ization or rating. An enlisted person is con-
sidered a real member of a rating when he be-
comes a petty officer (see below). The enlisted
personnel in the Navigation Department are
members of the quartermaster rating. They have
an insignia (a ship's wheel) and an identity dis-
tinct from other ratings. They are generally con-
sidered to be relatively intelligent, although not
as smart as data processing specialists. For en-
listed personnel, rating insignia denote occupa-
tional fields.
 106/1540

A petty officers is not a kind of commissioned

officer (the type of officer referred to by the un-
marked term 'officer'); the label 'petty officer'
simply designates an enlisted person who is a
practicing members of some rating. There are
two major levels of petty officer, with three
rates within each. One moves through the lowest
of these levels while learning the skills of the
speciality of the rating. One advances through
petty officer third class, petty officer second
class, and petty officer first class. A petty officer
third class is a novice in the speciality and may
perform low-level activities in concert with oth-
ers or more autonomous functions "under in-
struction." A petty officer first class is expected
to be fully competent in the rating.
Page 16
 108/1540

The next step up in rank moves one to the high-

er of the enlisted rates and is usually the most
important transition of an enlisted person's ca-
reer. This is the move to chief petty officer
(CPO). This change in status is marked by a
ritual of initiation which is shrouded in secrecy.
Just what happens at a chiefs initiation is sup-
posed to be known only by chiefs. However,
much of what happens apparently makes for
such good story telling that it cannot be kept en-
tirely in confidence. It is "common knowledge"
that these initiations frequently include hazing
of the initiate, drunkenness, and acts of special
license. Making chief means more than getting a
bigger pay packet or supervising more people.
Chiefs have their own berthing spaces (more
private that general enlisted berthing) and their
own mess (eating facility). On many ships the
chiefs mess is reputed to be better than that of
the officers. Chiefs are also important because
they are the primary interface between officers
 109/1540

and enlisted personnel. Since they typically have

from 12 to 20 years of experience in their speci-
ality, they often take part in problem-solving
sessions with the officers who are their super-
visors. Some chief petty officers have a consid-
erable amount of autonomy on account of their
expertise (or, perhaps, their expertise relative to
the supervising officer.) Chiefs frequently talk
about having to "break in" a new officer, by
which they mean getting a supervising officer
accustomed to the fact that the chief knows
more than the officer does and is actually in
charge of the space and the people in it. Officers
who directly supervise lower-level enlisted per-
sonnel risk undermining the chain of command
and incurring the resentment of a chief who
feels that his authority has been usurped. Once
one has made chief, there are still higher enlis-
ted rates to be attained. After approximately 20
years of service a competent person may make
senior chief, and after perhaps 25 years of
 110/1540

service (being now of about the same age as a

captain) one may make master chief. That is
normally the end of the line for an enlisted per-
son. There are, however, some ranks that fall
between enlisted and officer. A chief may elect
to become a chief warrant officer or a limited
duty officer (LDO). A chief who becomes an
LDO is commissioned as an ensign and may be-
gin to rise through the officer ranks. Few chiefs
take this path. As one senior chief asked rhetor-
ically, ''Why would I want to go from the top of
one career to the bottom of another?"

While an enlistee may have preferences for cer-

tain ratings, the choice of a rating is not entirely
up to the enlistee. Aptitude-test
Page 17
 112/1540

scores are also used to place people in various

specialities. The fact that people are screened
contributes to widely held stereotypes concern-
ing the intelligence of those in various ratings.
For example, boiler technicians (BTs) and
machinist's mates (MMs), who run a ship's
propulsion plant and who may go weeks without
seeing the light of day, are often the butt of
jokes about their low intelligence. Data pro-
cessing specialists, on the other hand, are gener-
ally thought to be bright. The ship, as a micro-
cosm, manifests the same patterns of competing
identities that are seen among the specialties in
the Navy as a whole. From the point of view of
the bridge personnel there may be little apparent
difference between machinist's mates and boiler
technicians, but down in the propulsion spaces
the perceived differences are many. Machinist
mates call boiler technicians "bilge divers,"
while boiler technicians call machinist's mates
"flange heads." Mostly, this is good-natured
 113/1540

teasing; name calling is a way of asserting one's

own identity.

At all levels of organization we see attempts to

establish identity by distinguishing oneself from
the other groups. This is relevant to the discus-
sion that follows because the dynamics of the
relationships among the people engaged in the
task of navigation are in part constrained by
these identities.

Officer Ranks
 114/1540

Military officers are managers of personnel and

resources. In general, their job is not to get their
hands dirty, but to ensure that those who do get
their hands dirty are doing the right things. Un-
like enlisted persons, officers do not have nar-
rowly defined specialities. An officer pursues a
career in one of the broad areas described
above: air, surface, or submarine warfare. With-
in that area, there are subspecialties such as en-
gineering and tactics.

Officers are initially commissioned as ensigns.

Ensigns have a tough lot. They are more visible
than the lowest enlisted rates, and they certainly
are given more responsibility, but often a "fresh-
caught" ensign knows little more about the
world of the ship than the seaman recruit.

Finding One's Way Around a Ship

 115/1540

A ship is a complicated warren of passages and

compartments. Every frame and compartment is
numbered with a code that
 Page 18

indicates which deck it is on, whether it is to

port or starboard of the centerline, and where it
is in the progression from stem to stern. Navig-
ating inside a ship can be quite confusing to a
newcomer. Inside the ship, the cardinal direc-
tions are forward and aft, port and starboard,
topside and below, and inboard and outboard;
north, south, east and west are irrelevant. On
large ships, orientation can be a serious prob-
lem. In the early 1980's the Navy sponsored a
research project to work on wayfinding in ships.
 117/1540

The ship is composed of a number of neighbor-

hoods. Some are workplaces, some are residen-
tial. Some are officially dedicated to recreation,
others are unofficially recreational. The fantail
on some classes of ships, for example, is a place
to hang out. Officers' accommodations and eat-
ing facilities are in a section of the ship called
"officer country." The chief petty officers have a
similar area, called "CPO country." Enlisted per-
sonnel are supposed to enter these areas only
when they are on official business. They are
supposed to remove their hats when entering
any compartment in these neighborhoods. Some
passageways inside the ship are major thorough-
fares; others are alleys or culs-de-sac. A visitor
quickly learns to search out alternative path-
ways, because corridors are frequently closed
for cleaning or maintenance.

On the 05 Level
 118/1540

As my escort and I arrived at a small platform

on the 05 level, to the right was a floor-to-ceil-
ing partition painted flat black. Behind the parti-
tion stood an exterior doorway that led out to the
starboard wing bridge. The partition forms a
"light trap" that prevents light from leaking out
at night when the ship is running dark. To the
left was a dark corridor that led to a similar
doorway on the port side of the island. Above
us, the ladder continued upward one more level
to the signal bridge. Ahead lay a narrow pas-
sageway. Forward along the left side of the pas-
sageway were two doors. Behind the first was
the captain's at-sea cabin. He has a nicely ap-
pointed quarters below, but he takes meals and
sleeps in this cabin during operations that re-
quire him to stay near the bridge. The next door
opened on the charthouse. At the end of the pas-
sageway, about 25 feet away, was a door that
led to the navigation bridge or pilothouse.
 119/1540

The charthouse is headquarters for the Naviga-

tion Department. This small room, crowded
with navigation equipment, two desks, a
 Page 19

safe, and a chart table, enjoys a luxury shared by

only a few spaces on the ship: a single porthole
through which natural light may enter and mix
with light from the fluorescent lamps overhead.
The charthouse is one of several spaces under
the control of the Navigation Department. Nav-
igation personnel not only work in these spaces,
they are also responsible for keeping them
clean. Since the bridge is one of the main work
areas of the ship's captain, it is thought to be es-
pecially important to keep it looking nice. While
in port, Navigation personnel polish the brass on
the bridge. Because the captain's at-sea cabin is
adjacent to the charthouse, members of the Nav-
igation Department tend to work more quietly
there than they might in other parts of the ship.
Since the average age of a sailor is under 20
years, a certain amount of playful horsing
around is expected in many parts of the ship, but
is not tolerated on the 05 level.
 121/1540

The Navigation Department is responsible for

all of the spaces on the 05 level with the excep-
tion of the captain's at-sea cabin. It is also re-
sponsible for the secondary or auxiliary conning
station ("Secondary Conn")a completely redund-
ant navigation bridge located in the bow, just
under the forward edge of the flight deck.
 122/1540

Secondary Conn is manned by the ship's execut-

ive officer and a complete navigation team
whenever the ship is at general quarters (battle
stations). This is done because the primary nav-
igation bridge in the island is very vulnerable if
the ship comes under attack. Modern anti-ship
missiles home in on electromagnetic radiation.
Because the radar antennae on the top of the is-
land are the principal sources of such radiation
on the ship, the island is the most likely part to
be hit by a missile. If the primary navigation
bridge is destroyed, the ship can be controlled
from Secondary Conn under the command of
the executive officer. Secondary Conn is a space
assigned to the Navigation Department and is a
duty station for Navigation personnel, but it will
be of little interest to us with regard to the nor-
mal practice of navigation. The ship's extensive
library of charts and navigation forms is stored
in this space.
 123/1540

The Navigation Department is supervised by the

Navigator. At the time the observations reported
here were made, the Palau's Navigation Depart-
ment consisted of the Navigator and seven enlis-
ted men. The title "Navigator" refers to the posi-
tion as head of the Navigation Department
rather than to the officer's technical speciality.
Though it is expected that an officer who serves
as Navigator aboard any ship will know enough
about navigation to
 Page 20

supervise the working of the Navigation Depart-

ment, Navigators seldom do any navigating
themselves.

The work of the Navigation Department is car-

ried out by enlisted personnel of the quartermas-
ter rating under the direction of the Assistant
Navigator (a quartermaster chief).

Navigating Large Ships

 125/1540

While a naval vessel is underway, a plot of its

past and projected movements is maintained at
all times. Such complete records are not always
kept aboard merchant vessels and are not abso-
lutely essential to the task of navigating a ship in
restricted waters. It is possible for an experi-
enced pilot to "eyeball" the passage and make
judgements concerning control of the ship
without the support of the computations that are
carried out on the chart. Aboard naval vessels,
however, such records are always keptprimarily
for reasons of safety, but also for purposes of
accountability. Should there be a problem, the
crew will be able to show exactly where the ship
was and what it was doing at the time of the
mishap. Day and night, whenever a ship is
neither tied to a pier nor at anchor, navigation
computations are performed as frequently as is
required to ensure safe navigation. During a
long passage, navigation activities may be per-
formed almost continuously for weeks or even
 126/1540

months on end. Most of the time the work of

navigation is conducted by one person working
alone. However, when a ship leaves or enters
port, or operates in any other environment
where maneuverability is restricted, the compu-
tational requirements of the task may exceed the
capabilities of any individual; then the naviga-
tion duties are carried out by a team.
 127/1540

The conning officer is nominally responsible for

the decisions about the motion of the ship, but
for the most part he does not make the actual de-
cisions. Usually, such decisions are made by the
Navigation Department and passed to the con-
ning officer as recommendations, such as "Re-
commend coming right to 0 1 7 at this time."
The conning officer considers the recommenda-
tion in the light of the ship's overall situation. If
the recommendation is appropriate, he will act
upon it by giving orders to the helmsman, who
steers the ship, or tothe leehelmsman, who con-
trols the engines. At all times when the ship may
have need of navigational information, someone
from the Navigation Department is at work and
ready to do whatever is required. The navigation
team per-
 Page 21

forms in a variety of configurations, with as few

as one and as many as six members of the Nav-
igation Department working together. In every
configuration there is one individual, designated
the quartermaster of the watch, who is respons-
ible for the quality of the work performed and
who serves as the department's official interface
with other departments aboard ship.
 129/1540

Navigation is a specialized task which, in its or-

dinary operation, confronts a limited set of prob-
lems, each of which has a wellunderstood struc-
ture. The problem that confronts a navigator is
usually not one of figuring out how to process
the information in order to get an answer; that
has already been worked out. The problem, in
most instances, is simply to use the existing
tools and techniques to process the information
gathered by the system and to produce an appro-
priate evaluation of the ship's situation or an ap-
propriate recommendation about how the ship
should proceed in order to get where it is sup-
posed to go.
 130/1540

The navigation activity is event-driven in the

sense that the navigation team must keep pace
with the movements of the ship. In contrast with
many other decision-making settings, when
something goes wrong aboard a ship, it is not an
option to quit the task, to set it aside momentar-
ily, or to start over from scratch. The work must
go on. In fact, the conditions under which the
task is most difficult are usually the conditions
under which its correct and timely performance
is most important.

The Researcher's Identity

Having said something about how naval person-

nel establish their own identities, I should also
say something about how they and I negotiated
an identity for me.
 131/1540

In the course of this work I made firsthand ob-

servations of navigation practice at sea aboard
two aircraft carriers (the Constellation and the
Ranger) and two ships of the amphibious fleet
(the one known here as the Palau and the Den-
ver). Aboard the aircraft carriers, I worked both
on the navigation bridge and in the combat in-
formation center. I made a passage from San
Diego to Seattle, with several stops, aboard the
Denver. I also interviewed members of the Nav-
igation Departments of five other ships (the En-
terprise, the Beleau Wood, the Carl Vinson, the
Cook, and the Berkeley) and had a number of in-
formal conversations with other navigation
personnel.
Page 22
 133/1540

The events reported here come mainly from op-

erations in the Southern California Operations
(SoCalOps) area aboard the Palau. I also
worked with the crew while the ship was in port.
I logged a total of 11 days at sea over a period
of 4 months. First came a weeklong trip during
which I observed the team, got the members
used to my presence, and got to know them.
During this trip, I only took notes and made a
few still photos and audio tape recordings of
navigation tasks and interviews with crewmen.
On a later trip, I mounted a video camera with a
wide-angle lens in the overhead above the chart
table in the pilothouse. I placed a stereo tape re-
corder on the chart table, with one channel cap-
turing the ambient noise and conversation of the
pilothouse. The other channel I wired into the
sound-powered phone circuit. Because the chief
was both plotting positions and supervising the
work of the navigation team, I wanted to be sure
to capture what he said. I therefore wired him
 134/1540

with a remote transmitter and a lavaliere micro-

phone. I used this signal to feed the audio track
on the video recording. Thus, I had one video
track and three audio tracks to work with.

During my time at sea, I took a normal watch

rotation. I appeared on the bridge on one occa-
sion or another during every watch period, in-
cluding the one from midnight to 4 a.m. I was
accorded privileges appropriate to the military
equivalent of my civilian Government Service
rank: lieutenant commander. I was assigned a
cabin in "officer country," took my meals in the
officer's mess, and spent my waking off-watch
time either in the charthouse with the navigation
crew or in the wardroom with officers.
 135/1540

As to what they thought of me, one must begin

with the understanding that for military folk the
military/civilian distinction stands just below the
friend/foe distinction as an element of the estab-
lishment of identity. A civilian aboard a ship is
an outsider by definition. It was important that
the navigator treated me as a colleague and
friend, and that the captain normally addressed
me as Doctor when we met. Many of the mem-
bers of the navigation team were also aware that
I had lunched at least once in the captain's quar-
ters, an honor reserved for visiting VIPs.

Some evidence of what the crew thought of me

is available in the video record. Early on, a num-
ber of nervous jokes were made on camera
about the dangerous potential of the videotap-
ing. In the first 5 minutes of videotaping with
this crew, the assistant navigator told the navig-
ator "Everything you say around me is getting
recorded for history, for your court-martial."
136/1540
 Page 23

On more than one occasion while he was away

from the chart table, the chief of the navigation
team explained my work to other members. He
apparently forgot that he was being recorded. I
discovered these comments weeks later while
doing transcription. During my second at sea
period, the chief went into the charthouse to
check on the fathometer. The fathometer operat-
or asked who I was. The conversation proceeded
as follows:

Chief: He's studying navigation on big ships.

He's the guy, he makes computer programs for
teaching stuff. Like they got a big computer pro-
gram thing they use in ASW school to teach
maneuvering boards. It's all computerized. He is
the one that makes it. He is the one who makes
things like that. He's a psychologist and anthro-
pologist. Works for the navy. He's a Ph.D.
Makes all kinds of strange things.
 138/1540

Fathometer operator: He makes all kinds of

strange money too.

Chief: Yeah, does he? He knows what he is do-

ing. He's swift. He just sits and watches and re-
cords everything you're doing. Then he puts it
all in data, then he starts putting it in a program.
Figuring out what to do, I don't know.
 139/1540

My most intensive data collection was carried

out on a four-day exercise during which the
Palau left port, steamed around the operations
area for two days, reentered port, and anchored
in the harbor overnight. The next morning the
ship left port again for another day of exercises.
Finally, it entered port again and returned to its
berth at the 32nd Street Naval Station. It was
during the last entry to port that the crisis repor-
ted in the opening pages of this book occurred.
The quality of the recording from the sound-
powered phone circuit was poor until I dis-
covered a better way to capture the signal on the
last entry to port. The two entries to and exits
from port were recorded from the time Sea and
Anchor Detail was set until the navigation team
stood down. This procedure produced video and
audio tape recordings of about 8 hours of team
activity. Additional recordings were made at
various times during Standard Steaming Watch.
In addition to the video and audio records, I
 140/1540

took notes during these events of any aspects of

the situation that I noticed that could not be fully
captured on the tapes. Even with the wide-angle
lens, the video camera captured only the surface
of the chart table. This permitted me to identify
features on the chart and even to know which
buttons of a calculator were pressed, but it
 Page 24

meant that many events of interest were not cap-

tured on tape because they occurred out of cam-
era range.
 142/1540

Transcribing the tape recordings was a very dif-

ficult process. At times there were four or more
conversations happening simultaneously in the
pilothouse. To make matters worse, ships are
noisy places. There are many kinds of equip-
ment on the bridge that create background
noises. The bosun's mate pipes various an-
nouncements from a station just aft and inboard
of the chart table, and his whistle blowing and
his public-address messages sometimes drown
out all other sounds. Helicopters may be operat-
ing on the flight deck or in the air just outside
the pilothouse. It was often necessary to listen to
each of the three audio tracks separately in order
to reconstruct what was being said, and still in
many cases the full content of the tapes cannot
be deciphered. Because of the placement of the
microphones, however, the coverage of the
verbal behavior of the members of the naviga-
tion team was uniformly good. Only rarely was
 143/1540

it impossible to determine what was being said

with respect to the navigation task.
 144/1540

I did much of the transcription myself, for three

reasons. First, this is a technical domain with
many specialized words in it. We know that
hearing is itself a constructive process and that
ambiguous inputs are often unconsciously re-
constructed and cleaned up on the basis of con-
text. Lacking context, other transcribers could
not hear what I could hear in the tapes. For ex-
ample, an untrained transcriber without expecta-
tions about what might be said during an an-
choring detail transcribed "thirty fathoms on
deck" as "thirty phantoms on deck." Naviga-
tionese is a foreign language to most people, and
quality transcription cannot be expected from a
transcriber who is not fluent in it. Second, since
there were many speakers, the fact that I knew
them personally helped me distinguish the iden-
tity of speakers where it was not clearly evident
from the content of a statement who was speak-
ing. Third, and most important, there is no better
way to learn what is actually in a recording than
 145/1540

to listen to it the many times that one must in or-

der to produce a good transcription. (Over a
period of about a year, one transcription assist-
ant did develop enough familiarity with the sub-
ject to provide usable transcriptions.)

The fact that listening is reconstructive intro-

duces the possibility of distortions in the data
driven by my expectations. I will attempt to deal
with that by making the ethnographic grounds
for my interpretations explicit.
 Page 25

In the pilothouse I tried not to participate, but

only observe. On only one occasion did I inter-
vene, and that was a case in which I felt that by
failing to speak I would put a number of people
in serious danger. My intervention was a brief
sotto voce comment to the navigator, who re-
solved the situation without indicating my role
in it.
 147/1540

It was clear that I knew more about the theory of

navigation than the members of the crew I was
studying with the exception of the ship's navig-
ator and the quartermaster chief. Of course,
knowing the theory and knowing the nature of
the practice in a particular setting are two quite
different things. In no case did I know more
about an individual's relation to the practice of
navigation than that individual. Still, this is an
unusual situation for an ethnographer. The web
of constraints provided by cultural practices is
important both to the people doing the task and
to the researcher. For the performers, it means
that the universe of possible activities is closely
bounded by the constraints. For the researcher,
the activities that are observed are interpreted in
terms of their reflection of the constraints. My
many years of studying and practicing naviga-
tion made me a particular sort of instrument, one
in which the constraints of the domain were
present. My interpretations of the actions of the
 148/1540

members of the navigation team were informed

by many of the same constraints that were guid-
ing their behaviors. But there was more. Be-
cause I attempted to continually make these con-
straints explicit, and to conceive of them in a
computational sense as well as in the operational
sense required of the navigation team, my inter-
pretations were not simply those of a native.

A few months of field work is, for an anthropo-

logist, a rather a short visit. Many aspects of the
military culture go unreported here because I am
not confident about their organization and
meaning on the basis of such a short exposure. I
did have 5 years of employment as a civilian
scientist working for the Navy, and that gave me
many opportunities to observe aspects of milit-
ary organization. The coverage of navigation
practice is adequate, I think, because of the op-
portunity on my second at-sea period to video-
tape the navigation operations on the bridge.
 149/1540

How different would the story be if the observa-

tions had been made aboard another ship? I do
not believe that the culture would permit it to be
very different. The information processed by the
navigation team may move more or less effi-
ciently, and the individual quartermasters may
have better or poorer relationships with
 Page 26

one another, but the tasks remain, and the means

of performing the tasks are standardized
throughout the fleet. The crews of different
ships may meet the requirements of navigation
more or less capably, but they must nevertheless
solve these particular tasks in the limited num-
ber of ways possible.

In fact, I made observations aboard several

ships, and my colleague, Colleen Siefert, did so
on yet another ship. The differences we ob-
served across ships were minor. The ship
Colleen observed had more quartermasters
available and was therefore able to organize its
navigation team in a slightly different way; that
however, does not present a challenge to my
framework or to my basic descriptions of the
nature of the cognition at either the individual or
the group level.
 151/1540

On the Bridge: Standard Steaming Watch

At the forward end of the 05 level's passageway

is the door to the navigation bridge or pilot-
house. It is here that the most important part of
the navigation work is done. The pilothouse oc-
cupies the forward 18 feet of the 05 level of the
island (see figure 1.1). Outward-canting win-
dows extend from chest height to the overhead
on both sides and the front of the pilothouse.
The windows on the port side and forward over-
look the flight deck. All work tables are moun-
ted on substantial bases on a light greenish lino-
leum floor. The walls, the cabinets, and the
equipment stands are thickly coated in light gray
paint. The overhead is flat black and tangled
with pipes and cables, their identities stenciled
on them in white. The polished brass of ship's
wheel and the controls for the engineorder tele-
graph stand out in the otherwise drab space.
 152/1540

The activities of the Navigation Department re-

volve around a computational ritual called the
fix cycle. The fix cycle has two major parts: de-
termining the present position of the ship and
projecting its future position. The fix cycle gath-
ers various bits of information about the ship's
location in the world and brings them together
in a representation of the ship's position. The
chart is the positional consciousness of the ship:
the navigation fix is the ship's internal represent-
ation of its own location.

When I first made it known to a ship's navigator

that I wanted to know how navigation work was
performed, he referred me to the Navigation De-
partment Watch Standing Procedures, a docu-
ment that describes the watch configurations.
"It's all in here," he said.
154/1540
 155/1540

Figure 1.1 A plan view of the pilothouse and the charthouse.

the navigation team do most of their work at the chart table,
and in the charthouse. The heavy line represents the exterior
Up in the diagram is forward on the ship.

''You can read this and save yourself the trouble of

watch." Of course it is not all in there, but the norm
scription in the Procedures is not a bad place to star
Navigation Department's "official" version of the o
of its work. This document is one of many symboli
which navigators "represent themselves to themselv
one another" (Geertz 1983).

Because the procedures refer to objects and places t

of shipboard navigation culture, understanding thes
will require us to explore the environment of naviga
conducting this exploration, we should keep in min
 Page 28

descriptions of navigation work that appear in a

ship's documents and in various navigation pub-
lications must be taken as data rather than
analysis.

In this section I will attempt to use the ship's

documents as a guide to the task of navigation.
The specifications presented in the Watch
Standing Procedures describe actions to be
taken and equipment and techniques to be used.
First I will present the normative descriptions
and try to provide the sort of background in-
formation that might be provided by a native of
the navigation culture, in the hope that this will
make these things meaningful to a reader who is
not a practitioner of the art. Later I will present
an analysis of the procedures, tools, and tech-
niques that will be grounded in information-pro-
cessing theory rather than in the world of ship
navigation.
 157/1540

The Palau's normal steaming watch procedures

are introduced as follows:
While in normal steaming condition at sea, the
following watch procedures will be adhered to as
closely as possible, modified as necessary by
82866990447203f57400372e62d765b5.gif 828669
situations beyond the control of the watch
stander.

In normal steaming, a single quartermaster is re-

sponsible for all the navigation duties. The pro-
cedures described in the document are taken ser-
iously, although it is recognized that it may not
be possible to execute them as described in all
circumstances. The normative procedures are an
ideal that is seldom achieved, or seldom
achieved as described.

The Primary Duty of the QMOW

 158/1540

When the Navigation Department is providing

navigation services to the ship, a particular
quartermaster is designated as the quarter master
of the watch (QMOW) at all times. According to
the procedures,

The Primary Duty of the QMOW is the safe

navigation of the ship. To this end he shall:

(a) Fix the position of the ship by any and all

methods available.

(1) All fixes will be plotted.

(2) When information is available, a fix will be

plotted at least every hour, when in open ocean
transit.

(3) When within Visual or Radar sight of land, a

fix will be plotted at least every fifteen minutes.

(i) Visual bearings will take priority.

82866990447203f57400372e62d765b5.gif
 159/1540

(ii) Fill in with Radar as necessary.

82866990447203f57400372e62d765b5.gif

(4) Fixes may be obtained from any combina-

tion of the following sources:

(i) Visual bearings

82866990447203f57400372e62d765b5.gif
 Page 29

(ii) Radar ranges

82866990447203f57400372e62d765b5.gif

(iii) Radar bearings

82866990447203f57400372e62d765b5.gif

(iv) Fathometer (line of soundings, bottom

82866990447203f57400372e62d765b5.gif
contouring, or guyout hopping)

(v) NavSat
82866990447203f57400372e62d765b5.gif

(vi) Omega
82866990447203f57400372e62d765b5.gif

(vii) Celestial observations

82866990447203f57400372e62d765b5.gif

(5) Fixes obtained from visual or radar sources

will consist of at least three LOPs.
 161/1540

(b) Project the ship's track by dead reckoning to

a sufficient length of time that any danger
presented to the ship from land, shoals or other
fixed dangers, or violation of international wa-
ters will be noticed well in advance of the ship
actually standing into danger or departing legal/
assigned waters.
 162/1540

Items a and b in this document describe the two

main parts of the fix cycle: fixing the ship's pos-
ition and projecting its track. The procedures of
dead reckoning will be explained in detail in
chapter 2. The plotted fix is a residue on the
chart of a process that gathers and transforms in-
formation about the ship's position. A succes-
sion of fixes is both a history of the positions of
the ship and a history of the workings of the
process that produced the position information.
The requirement that all fixes be plotted ensures
a complete history of positions and provides cer-
tain opportunities to detect and correct faults in
the process that creates the history. The interval
between fixes is set to 60 minutes in open wa-
ters and no more than 15 minutes when the ship
is in visual or radar contact with land. Near land,
the ship may stand into danger more quickly
than when in the open ocean. Sailors know that
it is not the open ocean that sinks ships, it's all
that hard stuff around the edges. The increased
 163/1540

frequency of fixes near land is intended to en-

sure that dangers are anticipated and avoided.
Visual bearings are given priority because they
are the most accurate means of fixing position.
The potential sources of position information
are listed roughly in order of their accuracy and
reliability.

The procedure states that fixes may be obtained

from any combination of a number of sources.
Let us briefly consider the nature of these
sources and the kinds of information they con-
tribute to fixing the position of the ship.

Sources of Information for Position Fixing

VISUAL BEARINGS

The simplest way of fixing position, and the one

that will concern us most in this book, is by
visual bearings. For this one needs a chart of the
region around the ship and a way to measure the
164/1540
 Page 30

direction (conventionally with respect to north)

of the line of sight connecting the ship and some
landmark on the shore. The direction of a land-
mark from the ship is called the landmark's
bearing. Imagine the line of sight in space
between the ship and a known landmark. Al-
though we know that one end of the line is at the
landmark and we know the direction of the line,
we can't just draw a line on the chart that corres-
ponds to the line of sight between ship and land-
mark, because we don't know where the other
end of the line is. The other end of the line is
where the ship is, and that is what we are trying
to discover.
 166/1540

Suppose we draw a line on the chart starting at

the location of the symbol for the landmark on
the chart and extend it past where we think the
ship isperhaps off the edge of the chart if we are
really unsure. We still don't know just where the
ship is, but we do know it must have been some-
where on that line when the bearing was ob-
served. Such a line is called a line of position
(LOP). If we have another line of position, con-
structed on the basis of the direction of the line
of sight to another known landmark, then we
know that the ship is also on that line. If the ship
was on both of these lines at the same time, the
only place it can have been is where the lines in-
tersect. The intersection of two lines of position
uniquely constrains the location from which the
observations were made. In practice, a third line
of position with respect to another landmark is
constructed. The three lines of position form a
triangle, and the size of that triangle is an indic-
ation of the quality of the position fix. It is
 167/1540

sometimes said that the navigator's level of anxi-

ety is proportional to the size of the fix triangle.

The observations of visual bearings of the land-

marks (direction with respect to north) are made
with a special telescopic sighting device called
an alidade. The true-north directional reference
is provided by a gyrocompass repeater that is
mounted under the alidade. A prism in the alid-
ade permits the image of the gyrocompass's
scale to be superimposed on the view of the
landmark. (The view through such a sight is il-
lustrated in figure 1.2.) The gyrocompass re-
peaters are located on the wings outside the
bridge. Each one is mounted on a solid metal
stand just tall enough to extend above the chest-
high metal railing that bounds the wing.
 168/1540

The most direct access to the port wing from the

chart table is through a door at the back of the
pilothouse just behind the captain's chair. In
cold weather, the captain of the Palau does not
permit traffic through this door. The only other
way to get from the
 Page 31

Figure 1.2 A view through an alidade. A prism inside the

alidade superimposes the images of two compass scales
onto whatever is seen through the telescopic sight. The in-
ner scale is a gyrocompass repeater; the outer scale is
fastened to the ship and indicates bearings relative to the
ship's head.
 170/1540

port wing position to the chart table is to go aft

on the wing to the hatch that leads to the island
stairwell and then come forward through the in-
terior passageway past the captain's at-sea cabin
and the charthouse. This makes it difficult to get
bearings sometimes, because it takes a long time
to go around the entire 05 level.

RADAR
 171/1540

Radar also provides information for position fix-

ing. The radar antenna on the ship's mast trans-
mits pulses of radiomagnetic energy as it ro-
tates. When the pulse strikes a solid object, the
pulse reflects off the object. Some of that reflec-
tion may return to the radar antenna that trans-
mitted it. By measuring the time required for the
pulse to travel to the object and return, the radar
can compute the distance to the object. This dis-
tance is called the range of the object. The dir-
ection in which the antenna is pointing when the
reflected pulse returns gives the bearing of the
object.
 172/1540

Radar ranges are more accurate than radar bear-

ings, so they are given priority in position plot-
ting. In practice, radar ranges plotted as circles
of position are often combined with visual bear-
ings to produce position fixes. The surface
search radar displays are located at the front of
the pilothouse on the starboard side. Each is
equipped with a heavy black rubber glare shield
that improves the visibility of the display in high
ambient light. This glare shield prevents two or
more people from looking at the scope at the
same time. The surface search radar also has
non-navigational uses. The
 Page 32

officer of the deck may use the radar to observe

and track other ship traffic. For this, a short
range is usually desired. The navigation tasks
often require a long range, and there is some-
times conflict between the two users of the
scopes. It is not difficult to change from one
range to another; however, in order to obtain the
required information after changing ranges, the
operator may have to wait for a full rotation of
the radar antenna at the new range setting.

FATHOMETER
 174/1540

The fathometer is a device for measuring the

depth of the water under a ship. It emits a pulse
of sound and measures the time it takes the
sound pulse to bounce off the sea bottom and re-
turn to the ship. The time delay is recorded by
the movement of a pen across a piece of paper.
The sound pulse is emitted when the pen is at
the top of the paper. The pen moves down the
paper at a constant speed and is brought into
contact with the paper when the echo is detec-
ted. The distance the pen travels down the paper
before making its mark is proportional to the
time required for the echo to return, which is in
turn proportional to the depth of the water. If the
water is deep, the sound will take longer to re-
turn, and the pen will have traveled farther down
the paper before coming into contact with it.
The depth of the water can be read from the
scale printed on the paper. Changing the scale of
the fathometer to operate in deeper or shallower
water is accomplished by changing the speed at
 175/1540

which the pen travels. The paper is mounted on

a motor drive that moves the paper to the side a
small amount just before each pulse. This results
in a continuous graphical record of the depth of
the water under the ship.

The Palau's fathometer is located in the chart-

house, so the QMOW must leave the bridge to
use it.

NAVSAT
 176/1540

Satellite navigation systems have now become

commonplace. They are easy to use, and they
provide high-quality position information. Their
major drawback at the time this research was
carried out was that with the number of naviga-
tion satellites then available the mean interval
between fixes was about 90 minutes. After com-
puting the ship's position from the reception of
satellite signals, the satellite navigation system
continuously updates the position of the ship on
the basis of inputs from the gyrocompass (for
direction) and
 Page 33

the ship's log (for speed). The NavSat system

aboard the Palau (located in the charthouse) was
a box, about the size of a small suitcase that
continuously displayed a digital readout of the
latitude and longitude of the ship.

The fact that NavSat systems must update posi-

tion with dead reckoning during the long wait
between fixes puts NavSat near the bottom of
the list of sources of information. With the im-
plementation of the Global Positioning System
(GPS), continuous satellite fixes are now avail-
able; the need for dead-reckoning updates of po-
sition has been eliminated. The military version
of GPS is accurate to within less than a meter in
three dimensions. The civilian versions are in-
tentionally degraded to a considerably lower ac-
curacy. GPS will very likely transform the way
navigation is done, perhaps rendering most of
the procedures described in this book obsolete.
 178/1540

OMEGA

Omega measures the phase difference between

the arrival of signals from multiple stations.
Omega was intended to provide accurate
worldwide position-fixing capability. In practice
it is unreliable. Whatever the source of the prob-
lems, they are perceived to be so serious that the
following warning appears in the Watch Stand-
ing Manual.
CAUTION: Positions obtained from Omega are
highly suspect, unless substantiated by informa-
tion from another source. In recent years, a num-
ber of costly and embarrassing groundings have
82866990447203f57400372e62d765b5.gif
been directly attributable to trusting Omega. No 828669
drastic decisions are ever to be made on unsub-
stantiated Omega fixes without the explicit per-
mission of the navigator.
 179/1540

If this system is considered to be so unreliable

that it merits this strongly worded caution in the
written procedures, what is it doing on the ship?
I believe the answer involves an interaction of
the organization of military research and fund-
ing with the development of technology. Omega
is a system that not only went into service be-
fore all the bugs could be worked out, it has
been overtaken by other superior technologies
before the bugs could be worked out. Still, it
was bought and paid for by the military, and
can, on occasion, provide useful navigation
information.

The Palau's Omega is located in the charthouse.

CELESTIAL OBSERVATIONS

By measuring the angular distance of a star

above the horizon, an observer can determine
his distance from the point on the surface
180/1540
 Page 34

of the earth that the star is directly above. This

point forms the center of a circle of position. In
a celestial sight reduction, each observed celesti-
al body defines a circle of position, and the ves-
sel from which the observations were made
must be located at the intersections of the circles
of position. Celestial observations appear at the
bottom of the list of sources of information.
When properly performed, celestial observations
provide fairly good position information.
 182/1540

There are, however, two major drawbacks to ce-

lestial observations. First, they can be performed
only under certain meteorological circum-
stances. This makes celestial navigation hard to
use and hard to teach. Several senior quarter-
masters have told me that they would like to
teach celestial navigation on training missions in
the Southern California operations area, but the
combination of air pollution and light pollution
(which makes the night sky bright, masking all
but the brightest stars and obscuring the line of
the horizon) produces very few occasions suit-
able for it. Second, the procedures are so com-
putationally complex that, even using a special-
ized calculator, a proficient celestial navigator
needs about half an hour to compute a good ce-
lestial position fix. Together these factors lead
to infrequent practice of this skill. I believe that
in the near future the only navigators who will
know how to fix position by star sights will be
 183/1540

those sailing on cruising yachts who cannot af-

ford a thousand dollars for a SatNav system.

DRAI
 184/1540

The Dead Reckoning Analyzer Instrument

(DRAI) is one of the most interesting naviga-
tional devices. A mechanical analog computer,
it takes input from the ship's speed log and the
gyrocompass and, by way of a system of mo-
tors, gears, belts, and cams, continuously com-
putes changes in latitude and longitude. The out-
put of the DRAI is expressed in the positions of
two dials: one reads latitude and the other lon-
gitude. If these dials are set to the current latit-
ude and longitude, the changes computed by the
motions of the internal parts of the DRAI will
move them so that their readings follow the lat-
itude and longitude of the ship. The crew of the
Palau claimed that when, properly cared for, the
DRAI is quite accurate and reliable. Older ver-
sions of the DRAI, such as the one aboard the
Palau, have been around since the 1940s. Newer
versions that do the same computations elec-
tronically are installed on some of the newer
ships.
185/1540
 Page 35

PIT SWORD AND DUMMY LOG

The pit sword is a device that is extended

through the hull and into the water to measure a
ship's actual speed through the water. The pit
sword extends several feet outside the hull and
measures speed by measuring the water's distor-
tion of a magnetic field. The speed signal gener-
ated by the pit sword is fed to speed indicators
on the bridge and to all the automated instru-
ments that do dead reckoning: the NavSat, the
DRAI, and the inertial navigation systems (if
present).
 187/1540

If the ship is operating in shallow water, the pit

sword cannot be extended from the hull. In this
case, or if for any other reason the pit sword
cannot be used, the dummy log is used. When a
ship is neither accelerating nor decelerating, its
speed can be estimated fairly accurately from
the rate of rotation of the propeller. The dummy
log is a device that senses this rate and provides
a signal that mimics what the pit sword would
produce at the corresponding speed.

Both of these devices are remote from the loca-

tion of the navigation team's normal activities. A
display of speed through the water is available
on the forward port side of the pilothouse, but it
is rarely consulted by the navigation team.

CHRONOMETERS
 188/1540

Three traditional spring-driven clocks are kept

in a special box in the Palau's charthouse. Read-
ings are recorded daily so that trends in the be-
havior of these chronometer's can be noted.
These records are maintained while time signals
are available on radio so that if time signals
should become unavailable the behavior of the
clocks will be known. If, for example, the log
shows that a particular chronometer loses a
second every day, that same rate of change will
be assumed until more reliable time sources are
restored.
 189/1540

The diversity of the many sources of navigation

information and the many methods for generat-
ing constraints on the ship's position produces
an important system property: the fact that posi-
tions are determined by combining information
from multiple, sometimes independent, sources
of information permits the navigation team to
check the consistency of the multiple represent-
ations with each other. The probability that sev-
eral, independently derived, representations are
in agreement with one another and are in error is
much smaller than the probability that any one
representation is in error.
 Page 36

At the Chart Table

The previous section described the sources of

information that the quartermaster of the watch
may use while discharging his primary duty: en-
suring the safe navigation of the ship. The in-
formation provided by these sources converges
on the chart table, where positions are plotted
and tracks are projected.

The Watch Standing Procedures specify addi-

tional constraints on the QMOW that bring us to
other aspects of the navigation team's task
setting:
The chart table and environs will be kept free of
extraneous material at all times. Only the chart(s)
in use, necessary publications, the logs of the
82866990447203f57400372e62d765b5.gif 828669
watch, and necessary writing/plotting
paraphernalia will be on the chart table.
 191/1540

The chart table is mounted against the starboard

wall of the pilothouse, just under the large
outward-canted windows. It is large enough for
full-size navigation charts and toolsabout 4 by 6
feet. Under the chart table are a number of lock-
ing drawers in which charts, publications, and
plotting tools are stored. A locking cabinet for
binoculars is mounted on the aft edge of the
chart table.

Navigation Charts
 192/1540

The most important piece of technology in the

position-fixing task is the navigation chart. A
navigation chart is a specially constructed model
of a real geographical space. The ship is some-
where in space, and to determine or "fix" the po-
sition of the ship is to find the point on the ap-
propriate chart that corresponds to the ship's po-
sition in space. The lines of position derived
from visual observations, radar bearings, radar
ranges, celestial observations, and depth-contour
matches are all graphically constructed on the
chart. Latitude and longitude positions determ-
ined by NavSat, Omega, or Loran are plotted
directly on the chart. A fix may be constructed
from a combination of these types of
information.

Navigation charts are printed on high-quality

paper in color. Natural and "cultural" features
are depicted in a complex symbology (see figure
1.3).
 193/1540

The Palau keeps an inventory of about 5400

charts depicting ports and coastlines around the
world. A complete set of charts for current oper-
ations are kept on the chart table, and a second
complete set in the table's drawers. The rest of
the charts are kept in a chart library in Second-
ary Conn.
 Page 37

Figure 1.3 A navigation chart in use. Such a chart includes

information about features
both above and below the water. This chart shows the en-
trance to San Diego Harbor.
 Page 38

The Secondary Duty of the QMOW

According to the Watch Standing Procedures,

The secondary duty of the QMOW is the keep-

ing of the logs of the watch.
 196/1540

Those who have experience in the merchant

fleet often say that it is not necessary to do all
the work of piloting in order to get a large ship
into port. A good ship driver can, after all, ''eye-
ball" the movement of the ship and get it down
the channel without having positions plotted on
short intervals. To say that it is possible to guide
a ship down a narrow channel without maintain-
ing the piloting record is not to say that it is
easier to do it that way. Even if nothing goes
wrong, the plotted and projected positions of the
ship on the chart are a useful resource to the
conning officer, and while it does require a nav-
igation team to do the work of plotting positions
and computing turn points, the task of the con-
ning officer is greatly simplified by the advice
he receives from the navigation team. If
something does go wrong, the work of the nav-
igation team becomes indispensable in two
ways. First, depending upon what it is that goes
wrong, computing the ship's position and track
 197/1540

may become essential to the process of figuring

out how to keep the ship out of trouble (see
chapter 8 for an example). Second, the records
kept by the navigation team-the chart, the deck
log, and the bearing logare all legal documents.
If the ship is involved in a mishap, as soon as it
is prudent to do so, all these documents are re-
moved from the chart table and locked in the
Executive Officer's safe. This precaution is
taken to ensure that they will not be tampered
with before they are turned over to a board of
inquiry investigating the incident. These records
may be needed to protect the navigation team,
the captain, the ship, and ultimately the Navy
from accusations of negligence or incompet-
ence. The Palau's Assistant Navigator offered
the following justification:
 198/1540

You can go into San Diego by eye. But legally,

you can't. If you haven't matched all the things
and something happens, not necessarily to you,
it don't have to. One of those buoys can float
loose in the goddamn bay and rub up along side
you. Boy, you better have everything covered
here, because they are going to try to hang the
captain. They will try to hang him. Unless he
can prove with data that everything he did was
right. Now ... the merchant ship wouldn't. They
would just say, "We were in the middle of the
channel. The damn thing hit us, and if there is
an expense, fine, charge the company."
 Page 39

Other records are kept as well. There is a separ-

ate log for the gyrocompasses (with entries
made twice daily), and another for the magnetic
compasses. (The DRAI reading is also recorded
in the magnetic compass log at the beginning of
each watch.) There is yet another log for the
ship's chronometers. A fathometer log is kept
with the fathometer during maneuvers in restric-
ted waters. A log of the ship's position is up-
dated daily.

The Tertiary Duty of the QMOW

 200/1540

The tertiary duty of the quartermaster of the

watch is "to give all possible aid to the Officer
of the Deck in the conduct of his watch." The
Officer of the Deck (OOD) is also normally the
conning officer, although he may delegate this
duty to a Junior Officer of the Deck. The im-
portance of the relationship between the
QMOW and the OOD is reflected in the follow-
ing excerpt from the Watch Standing
Procedures:
The QMOW will not leave the Bridge except to
take DRAI and Fathometer readings, and collect
NavSat and Omega fixes as necessary. If he
leaves the bridge, he will inform the OOD, and
82866990447203f57400372e62d765b5.gif
will absent himself for as short a period of time 828669
as possible. (If a Charthouse Quartermaster is as-
signed, there no necessity for the QMOW to
leave the bridge unless properly relieved.)
 201/1540

The control of the ship is a partially closed in-

formation loop. The conning officer senses the
ship's situation in the world by looking out the
window of the bridge. The members of the nav-
igation team also sense the world by looking at
it; in addition, however, they gather information
from other sources, and from that other informa-
tion they synthesize a more comprehensive and
accurate representation of the situation of the
ship. The navigation team uses its representation
to generate advice to the conning officer, who
by acting (or not acting) on that advice affects
the actual situation of the ship in the world
which is sensed and interpreted.
 202/1540

The navigation team relies on the conning of-

ficer to the extent that if the conning officer
turns the ship or changes its speed in other than
the recommended places then the workload of
the navigation team is increased. When the
quartermasters project the position of the ship
into the future, the projections sometimes in-
volve changes in course and or speed. When this
is the case, the projected track is carefully
planned, pre-computed, and plotted. If the ship
remains on the pre-computed track, many parts
of the required computation will have been per-
formed in advance. When the ship deviates from
planned track, new computations may be
 Page 40

required to establish when and where various

maneuvers are appropriate. For example, on one
of the Palau's departures from port an inexperi-
enced conning officer made several turns before
the recommended point. This happened because
the deck of the ship is so big and so high off the
water that from the point of view of the naviga-
tion bridge the surface of the water for several
hundred yards in front of the ship is hidden from
view. When a channel is narrow and some of the
turns are tight, channel buoys disappear beneath
the deck before the turn is commenced. For an
inexperienced conning officer, the temptation to
turn before the buoy disappears under the bow is
great. Once a buoy disappears beneath the deck,
it is difficult to estimate whether or not the ship
will hit it. To keep the ship on track, a conning
officer must be disciplined and must trust the
navigation team.
 204/1540

The conning officer has other obligations and

cannot always do what is easiest for the naviga-
tion team. On one occasion the Palau's engin-
eering department detected a rumbling noise in
the propeller shaft. In order to diagnose the
problem, the engineers requested 5° right> rud-
der, then 5° left rudder, then 10°° right rudder
followed by 10° left rudder. The ship was slalo-
ming along through 80° turns. This happened
while the ship was out of visual and radar range
of land, so its position had to be maintained by
dead reckoning, a very difficult task under these
conditions.

THE COMBAT INFORMATION CENTER

 205/1540

The navigation team also coordinates its activit-

ies with the Combat Information Center (CIC),
which is located below the flight deck. Duplic-
ate position plots are maintained by the Opera-
tions Specialists (OSs) who work in CIC. They
use radar bearings and ranges to fix the position
of the ship. Under conditions of reduced visibil-
ity, CIC is supposed to be the primary source of
navigation advice for the conning officer. The
quartermaster chief in charge of the Navigation
Department on the Palau said the following
about this shift in responsibility:
 206/1540

They've got a whole team down there [in CIC]

and they are pretty good at what they are doing.
They are supposed to be like a backup on what
happens up here. They've got good radars, and
for reduced visibility, they are supposed to be
primary. Now the only way that is going to hap-
pen is if I drop dead. As long as I am on a ship,
and this is the same thing I tell my navigator, as
soon as I walk on
 Page 41

board, "Everything that has to do with naviga-

tion while I am on board, I'm it. I'll hand you
papers to sign, I'll back you up in any way you
need. You will never get in trouble, navigation is
my business. " For OSs, it is a secondary busi-
ness to them. There are people in my business
who will let CIC take it. I won't.

I never saw this claim put to the test.

AIR BOSS

The Navigation Department provides position

information to the Air Boss, who is responsible
for controlling the aircraft that operate from the
flight deck. The most frequent requests for in-
formation from the air boss consist of position
or projected position information to be used by
aircraft coming to the ship, and directions and
distances to land bases for aircraft departing the
ship.
 208/1540

Sea and Anchor Detail

Guiding a large ship into or out of a harbor is a

difficult task. A ship is a massive object; its in-
ertia makes it slow to respond to changes in pro-
peller speed or rudder position. Putting the rud-
der over will have no immediate effect, but once
the ship has started turning it will tend to contin-
ue turning. Similarly, stopping the engines will
not stop the ship. Depending on its speed, a ship
may coast without power for many miles. To
stop in less distance, the propeller must be
turned in the reverse direction, but even this res-
ults in only a gradual slowing. Because of this
response lag, changes in direction or speed must
be anticipated and planned well in advance.
Depending on the characteristics and the velo-
city of the ship, the actions that will bring it to a
stop or turn it around may need to be taken tens
of seconds or many minutes before the ship ar-
rives at the desired turning or stopping point.
 209/1540

In order to satisfy the OOD's need for informa-

tion about the location and movement of the
ship when it is near hazards, the Navigation De-
partments of Navy ships take on a watch config-
uration called Sea and Anchor Piloting Detail.
Piloting waters are defined as follows in the
Watch Standing Procedures:
Piloting waterswithin five miles of land, shoals,
or hazards to navigation, or inside of the fifty
82866990447203f57400372e62d765b5.gif 828669
fathom curve, whichever is further from land.

Restricted watersInside of the outermost aid to

navigation or inside of the ten fathom curve,
82866990447203f57400372e62d765b5.gif 828669
whichever is further from land.
 Page 42

1. When operating within Restricted Waters, the

Sea and Anchor Piloting Detail will be
stationed.

2. The QMOW will ensure that all members of

the Sea and Anchor Piloting Detail are called at
least thirty minutes prior to entering restricted
waters.

3. The Sea and Anchor Piloting Detail will con-

sist of:

a. The Navigator
82866990447203f57400372e62d765b5.gif

b. The Assistant to the Navigator

82866990447203f57400372e62d765b5.gif

c. Navigation Plotter
82866990447203f57400372e62d765b5.gif

d. Navigation Bearing Recorder/Timer

82866990447203f57400372e62d765b5.gif

e. Starboard Pelorus Operator

82866990447203f57400372e62d765b5.gif

f. Port Pelorus Operator

82866990447203f57400372e62d765b5.gif
 211/1540

g. Restricted Maneuvering Helmsman

82866990447203f57400372e62d765b5.gif

h. Quartermaster of the Watch

82866990447203f57400372e62d765b5.gif

i. Restricted Maneuvering Helmsman in

82866990447203f57400372e62d765b5.gif
After Steering

j. Fathometer Operator
82866990447203f57400372e62d765b5.gif
 212/1540

As long as the visibility is adequate for visual

bearings, the primary work of the sea and an-
chor piloting detail is to fix the position of the
ship by visual bearings. The pelorus operators
stationed on the port and starboard wings, just
outside the doors to the pilothouse, measure the
bearings of specified landmarks and report the
bearings to the bearing recorder/timer (hence-
forth referred to as "the recorder"), who records
them in the bearing log. The recorder stands at
the after edge of the chart table in the pilot-
house. The bearing log is kept on the chart table,
adjacent to the chart. The navigation plotter
stands at the chart table and plots the recorded
bearings as lines of position on the chart, thus
fixing the position of the ship. The plotter also
projects the future positions of the ship, and to-
gether with the recorder he chooses landmarks
for the pelorus operators to use on future fixes.
The restricted-maneuvering helmsman stands at
the helm station in the center of the pilothouse
 213/1540

and steers the ship in accordance with com-

mands from the conning officer. In sea and an-
chor detail, the quartermaster of the watch is
mainly responsible for maintaining the ship's
log, in which all engine and helm commands
and other events of consequence to the naviga-
tion of the ship are recorded. The quartermaster
of the watch stands at the forward edge of the
chart table and keeps the ships log on the chart
table. The restricted-maneuvering helmsman is
stationed in the after steering compartment, at
the head of the rudder post in the stern of the
ship. In case of a problem with the ship's wheel,
the steering function can be taken over more dir-
ectly by the helmsman in aftersteering. The
fathometer operator is stationed in the chart-
house, which is separated from the pilothouse
by a bulkhead. The fathometer operator reports
the depth of the water under the ship for each
position fix. The navigator is responsible for the
 Page 43

work of the navigation team but does not nor-

mally participate directly in that work. Aboard
the Palau, even the supervision of the naviga-
tion team was done by the quartermaster chief,
who acted as Assistant to the Navigator. If the
crew had been more experienced, the Assistant
to the Navigator would not have taken up a
functional role in the performance of the task.
Because the Palau was understaffed and the
available personnel were inexperienced,
however, the assistant to the navigator also
served as navigation plotter.

Narrative: Sighting
 215/1540

In the late afternoon of a clear spring day the

U.S.S. Palau completed several hours of engin-
eering drills that left it alternately steaming in
tight circles and lying dead in the water. The
Palau had been at sea for a few days on local
maneuvers and was now just south of the en-
trance to San Diego Harbor. The crew was
anxious to go ashore, and going in circles and
lying dead in the water when home was in plain
sight was very frustrating. It was therefore
something of a cause for celebration in the pilot-
house when the engineering officer of the watch
called the bridge on the intercom and said "Main
engine warmed, ready to answer all bells." The
officer of the deck acknowledged the ready state
of the propulsion plant and advised the engin-
eering officer to "stand by for 15," meaning that
they should be prepared to respond to an order
for 15 knots of speed. Shortly thereafter, the
conning officer ordered the engine ahead stand-
ard speed. Pilothouse morale rose swiftly.
 216/1540

Quartermaster Second Class (QM2) John Silver

stood at the chart table in the pilothouse. He was
wearing a sound-powered telephone set (head-
phones and a collar-mounted microphone) that
connected him to other members of the naviga-
tion team who were not in the pilothouse. When
he learned that the ship would be getting under-
way again soon, he pressed the transmit button
on his microphone and said "We're baggin' ass!"

On a platform on the starboard side of the ship,

just outside the door to the pilothouse and about
50 feet above the surface of the water, Seaman
Steve Wheeler had been leaning on the rail,
studying the patches of foam that lay motionless
next to the hull, and wondering when the engin-
eering drills would end and the ship would
move again. When he heard Silver's exclamation
in his headphones, he looked up and began to
scan the city skyline for major landmarks.
Wheeler was the starboard pelorus operator, and
217/1540
 Page 44

it was his job to sight landmarks and measure

their direction from the ship. A novice, he had
done this job only once before, and was not sure
how to identify all the landmarks, nor was he
entirely clear on the procedure he was to
perform.

Inside the pilothouse, Quartermaster Chief Rick

Richards moved to the forward edge of the chart
table and looked over the shoulder of QM2
James Smith as Smith recorded the conning
officer's orders in the deck log. "Ahead stand-
ard, left 10 degrees rudder, come to course 3 0
5."
 219/1540

Chief Richards turned and leaned over the chart

table with QM2 Silver. As happy as they were
to be heading for their pier at last, they also
knew it was time to begin the high-workload job
of bringing the Palau into port. They examined
the chart of the approaches to San Diego Har-
bor. Silver found the symbolic depictions of
several important landmarks on the chart and
used his fingers to draw imaginary lines from
them to the last charted position of the ship.
These imaginary lines represented the lines of
sight from the ship to the landmarks. He
checked the angles at which the lines intersec-
ted. Pointing to the chart, he said to Richards
"How about these?" "Yeah, those are fine," the
chief replied.
 220/1540

Silver was the navigation team's bearing record-

er. It was his job to control the pelorus operators
on the wings of the ship and record the measure-
ments they made. Once Silver had chosen his
landmarks, he assigned them to the pelorus op-
erators: "Hey Steve, you'll be keeping Hotel del
and Dive Tower as we go in, and John, you got
Point Loma." Steve Wheeler answered "OK"
and heard his opposite number on the port wing,
Seaman John Painter, say "Aye." Wheeler
looked out across the water, found the conical
red roofs of the Hotel del Coronado on the
beach, and searched to the south along the
strand for the building called the Dive Tower.
There it was. Wheeler's hands were resting on
the alidade that was mounted on a shoulder-high
pedestal at his station. He quickly pointed the al-
idade in the rough direction of the Dive Tower
and leaned down, pressing his right eye against
the rubber eyepiece to look through the sight.
He saw the beach and some low buildings back
 221/1540

from the water's edge. He swung the sight left

and then right until the Dive Tower came into
view, then carefully rotated the sight on its ped-
estal until the vertical hairline in the sight fell
right down the middle of the tower. Near the
bottom of his field of view
 Page 45

through the alidade, he could see a portion of

the scale of a gyrocompass card. The hairline
crossed the scale three small tick marks to the
right of a large mark labeled 030. Another large
tick mark, labeled 040, was still further to the
right. Wheeler counted the little tick marks and
noted that the the Dive Tower bore 033°.

Once Silver had assigned the landmarks to the

pelorus operators, he wrote the name of each of
the chosen landmarks at the head of a column in
the bearing record log, which was lying on the
chart table between him and the chart.
 223/1540

Silver kept an eye on his wristwatch. It was a di-

gital model, and when he had come to his duty
station several hours ago he had synchronized it
with the ship's clock on the wall at the back of
the pilothouse. Now he had taken the watch off
his wrist and placed it on the chart table in front
of him, just above the pages of his bearing re-
cord log. As the ship began to move and turn to
its course for home, the plotter, Chief Richards,
told Silver to take a round of bearings. It was 13
minutes and 40 seconds after 4 pm. Silver de-
cided to make the official time of the next set of
bearing observations 16:14, using the 24-hour
notation standard in the military. He wrote
"1614" in the time column of the bearing record
log, and at 16:13:50 he said into his phone set:
"Stand by to mark. Time 14."
 224/1540

Seaman Ron White sat on a high stool at the

chart table, looking at the display of the fatho-
meter. On the chart table in front of him was a
depth sounder logbook. When he heard the
"Stand by to mark" signal in his headset, he read
the depth of the water under the ship from the
display and reported on the phone circuit: "Fif-
teen fathoms." He then logged the time and the
depth in his book. Silver recorded the depth in
the bearing record log.
 225/1540

Out on the starboard wing, Wheeler heard the

recorder say "Stand by to mark, time 14." As he
made a small adjustment to bring the hairline to
the center of the Dive Tower, he heard the
fathometer operator report the depth of the water
under the ship as 15 fathoms. The hairline now
crossed the scale at 034°. Wheeler pressed the
button on the microphone of his phone set and
reported "Dive Tower, 0 3 4." That was a mis-
take. The bearing was correct; however, in his
excitement Wheeler blurted out his bearing im-
mediately after the fathometer operator's report.
He was supposed to track the landmark and re-
port its bearing only after the recorder gave a
"mark" signal. The port pelorus operator noticed
the mistake and barked, "He didn't say 'Mark'."
 Page 46

But by then it was time to mark the bearings.

Wheeler's mistake was not a serious timing er-
ror; he was only a few seconds early. The im-
portant thing was to make the observations as
close the "mark" time as was possible. Stopping
to discuss the mistake would have been more
disruptive than continuing on. There was no
time for lessons or corrections now. The bearing
recorder quickly restarted the procedure from its
current state by giving the "mark" signal, ac-
knowledging the premature bearing, and urging
the pelorus operators to get on with their re-
ports: ''Mark it. I got Dive Tower, Steve. Go
ahead." Silver then wrote "0 3 4" in the column
labeled Dive Tower in the bearing record log.
 227/1540

The plotter, Chief Richards, was standing next

to Silver, waiting for the bearings. He leaned
across the chart table and read the bearing of
Dive Tower even as Silver was writing it in the
log. Silver noticed that Richards was craning his
neck to read the bearing from the book. Softly
he said "0 3 4" to Richards, whose face was
close to his. As Richards moved away from the
bearing log, he looked to the plotting tool in his
hands and acknowledged: "Uh huh."
 228/1540

Chief Richards held in his hands a one-armed

protractor called a hoey. The hoey has a circular
scale of 180 degrees on it, and a straight-edged
arm about 18 inches long that pivots in the cen-
ter of the scale. It is used to construct lines on
the chart that correspond to the lines of sight
between the ship and the landmarks. Richards
aligned the straightedge with the fourth tick
mark to the right of the large mark labeled 030
on the scale of the hoey and turned a knob at the
pivot point of the arm to lock its position with
respect to the scale. He then laid the hoey on the
chart and found the symbol on the chart that rep-
resented the Dive Tower. He put the point of his
pencil on the symbol on the chart. Holding it
there, he brought the straightedge up against the
pencil point. Keeping the straightedge against
the tip of the pencil and keeping the protractor
scale further away from the charted location of
the landmark than the anticipated location of the
fix, Richards slid the hoey itself around on the
 229/1540

chart until the directional frame of the protractor

scale was aligned with the directional frame of
the chart. The edge of the arm now lay on the
chart along a line representing the line of sight
from the ship to the landmark. Richards held the
hoey firmly in place while he removed his pen-
cil from the symbol for the landmark and drew a
line segment along the protractor arm in the vi-
cinity of the expected location of the ship on the
chart. By drawing only the sec-
 Page 47

tions of the lines of position that were in the vi-

cinity of the expected location of the ship,
Richards kept the chart neat and avoided the
creation of spurious triangles formed by the in-
tersection of lines of position from different
fixes.
 231/1540

While Chief Richards was plotting the line of

position for the Dive Tower, the port wing
pelorus operator reported the bearing of Point
Loma. By the time Silver had acknowledged the
port pelorus operator's report ("Three three nine,
Point Loma"), Richards was ready for the next
bearing. Because he was standing right next to
Silver, he could hear everything that Silver said
into his phonecircuit microphone. He could not
hear what the pelorus operators or others on the
circuit were saying to Silver or to one another;
however, he could hear what Silver said, and he
got the bearing to Point Loma by hearing
Silver's acknowledgement.
 232/1540

While the port pelorus operator was making his

report, "Point Loma, 3 3 9" and while Chief
Richards was plotting Dive Tower, Seaman
Wheeler swung his sight to the tallest spire of
the Hotel del Coronado, aligned the hairline, and
read the bearing from the scale. In his headset
he heard the recorder acknowledge "Three 3 9,
Point Loma." But he was trying not to listen, be-
cause he had his own numbers to report as soon
as the phone circuit became quiet: "Hotel del, 0
2 4." Then he listened as the recorder acknow-
ledged his report: "0 2 4, Hotel del.'' The report
was heard and echoed without error, so Wheeler
said no more.
 233/1540

About 30 seconds passed between the "Stand by

to mark" signal and the acknowledgement of the
third bearing. The pelorus operators relaxed at
their stations for a minute or so while the bear-
ings they had reported were processed by other
members of the navigation team to determine
the position of the ship at the time of the obser-
vations. The pelorus operators themselves did
not know exactly what had been done with the
bearings after they had reported them.

Less than 10 seconds after the acknowledge-

ment of the last bearing, Chief Richards had his
fix triangle constructed and was ready to label it
with the time of the observations. He asked Sil-
ver "OK, what time was that?" Silver looked in
the 'time' column of the bearing record log and
replied "One 4," meaning 14 minutes after the
hour.
 234/1540

With the fix plotted and labeled, Richards and

Silver turned to the tasks of predicting the posi-
tion of the ship at the time of the next fix (3
minutes hence) and deciding which course to
take for
 Page 48

the best approach to the harbor. Speaking slowly

while plotting, Chief Richards said: "He's still
turning. That's gonna put us about right here."
He made a mark on the chart at the end of an arc
he had drawn to represent the track of the ship
through the turn. Silver looked at the projected
position and determined that the same three
landmarks used for the previous fix would be
appropriate for the next fix.

At 16:16:50 Silver pressed the transmit button

on his mike and said: "Stand by to mark. Time 1
7."

"Fifteen fathoms," said the fathometer operator.

Silver said "Mark it." The pelorus operators re-

ported their bearings, and Silver read each one
back.

"Point Loma, 3 3 8."

 236/1540

"3 3 8, Point Loma."

"Dive Tower, 0 3 5."

"0 3 5, Dive Tower."

"Hotel del, 0 2 4."

"0 2 4, Hotel del."

 237/1540

Chief Richards plotted the ship's position, but it

was not as far along the track as he had projec-
ted it would be. Silver commented on the radius
of the projected turn: "That was a big ellipse."
Richards looked at the plot. "Oh, yeah," he said.
"It's just that propulsion is coming up really
slooooow. I only figured it to come up to 9, but
it didn't even come up to 4." Both men laughed,
and Silver said "Recompute." For half a minute
they worked together silently, jointly redoing
the computations of the speed calculation. They
checked the lines of position for the new fix and
measured the distance between the previous po-
sition and the latest one: 400 yards. Chief
Richards shook his head and said "Four knots."
Silver nodded and said "Right." Richards poin-
ted to the projected track of the ship up the
channel. "Four knots for the first 3 minutes," he
said. "At this rate we better change the timing a
little."
238/1540
 Page 49

2 Navigation as Computation
 240/1540

Navigation is the process of directing the move-

ments of a craft from one point to another.
There are many kinds of navigation. This
chapter lays the foundation for the construction
of an analysis of the information processing car-
ried out by those who practice a form of naviga-
tion referred to in The Western technological
culture as surface ship piloting. Piloting (or pi-
lotage) is navigation involving determination of
position relative to known geographic locations.
Rather than present what passes in our cultural
tradition as a description of how pilotage is
done, this chapter attempts to develop a compu-
tational account of pilotage. This account of pi-
lotage overlaps portions of the computational
bases of many other forms of navigation, includ-
ing celestial, air, and radio navigation. Aspects
of these forms of navigation will be mentioned
in passing, but the focus will be on the pilotage
of surface vessels in the vicinity of land. Unless
 241/1540

otherwise indicated, the term 'navigation' will

henceforth refer to pilotage.
 242/1540

Having taken ship navigation as it is performed

by a team on the bridge of a ship as the unit of
cognitive analysis, I will attempt to apply the
principal metaphor of cognitive sciencecogni-
tion as computationto the operation of this sys-
tem. In so doing I do not make any special com-
mitment to the nature of the computations that
are going on inside individuals except to say
that whatever happens there is part of a larger
computational system. But I do believe that the
computation observed in the activity of the lar-
ger system can be described in the way cogni-
tion has been traditionally describedthat is, as
computation realized through the creation, trans-
formation, and propagation of representational
states. In order to understand navigation practice
as a computational or information-processing
activity, we need to consider what might consti-
tute an understanding of an information-pro-
cessing system. Working on vision but thinking
of a much wider class of information-processing
 243/1540

systems, David Marr developed a view of what

it takes to understand an information-processing
system. The discussion here is based on Marr's
(1982) distinctions between several levels of de-
scription of cognitive systems.
 Page 50

Marr's Levels of Description

 245/1540

In his work on vision, Marr suggests that there

are several levels of description at which any
information-processing system must be under-
stood. According to Marr, the most important
three levels are as follows: The first level is the
computational theory of the task that the system
performs. This level of description should spe-
cify what the system does, and why it does it. It
should say what constraints are satisfied by the
operation of the system. Here, "the performance
of the [system] is characterized as a mapping
from one kind of information to another, the ab-
stract properties of this mapping are defined pre-
cisely, and its appropriateness and adequacy for
the task at hand are demonstrated" (Marr 1982).
Such a description is defined by the constraints
the system has to satisfy in order to do what it
does. The second level of description concerns
the "choice of representation for the input and
output and the algorithm to be used to transform
one into the other." This level specifies the
 246/1540

logical organization of the structures that encode

the information and the transformations by
which the information is propagated through the
system from input to output. The third level con-
cerns "the details of how the algorithm and rep-
resentation are realized physically." Marr points
out that there are many choices available at each
level for any computational system, and that the
choices made at one level may constrain what
will work at other levels.

Marr intended his framework to be applied to

cognitive processes that take place inside an in-
dividual, but there is no reason, in principle, to
confine it to such a narrow conception of cogni-
tion. In this chapter I will attempt to apply
Marr's prescription to the task of navigation.
 247/1540

Navigation is an activity that is recognizable

across cultures, yet in each cultural tradition it is
accomplished within a conceptual system that
makes certain representational assumptions. In
the next section, I give a computational account
of navigation that is independent of the repres-
entational assumptions of any established tradi-
tion of navigation practice. It is an account that
specifies the nature of the navigation problem
and the sorts of information that are transformed
in the doing of the task, yet spans the differ-
ences between even radically different traditions
of navigation.

Unfortunately, the computational account by it-

self is quite abstract and difficult to convey in
the absence of examples that embody the satis-
faction of the constraints that are described. I
will
Page 51
 249/1540

therefore illustrate aspects of the computational

account with a few examples taken from the
Western tradition of piloting. This should help
the reader to understand the nature of the con-
straints discussed. However, these examples are
inevitably grounded in the representational as-
sumptions of the Western cultural tradition, and
frequently have implications for algorithms, and
will probably suggest particular implementa-
tions. The inclusion of this sort of material
seems unavoidable. I will try to keep the ex-
amples as sparse as is possible and to make clear
distinctions between those aspects that properly
belong to the computational account and those
that belong to other levels of description. The
importance of keeping the computational de-
scription free of representational assumptions
will become apparent in the two subsequent sec-
tions, which briefly contrast the culturally spe-
cific representations and algorithms used by our
technological Western culture with those used
 250/1540

by a nonliterate Micronesian culture to solve the

navigation problem. Much of the remainder of
the book can be seen as a further elaboration of
the representational/algorithmic level of de-
scription and a thorough exploration of the im-
plementation of navigation computations by
navigation teams on large ships.
 251/1540

The implementational details have been largely

ignored in the past. This may be due in part to
the notion that in informationprocessing systems
what is important is the structure of the compu-
tation, not the means of implementation. One of
the most important insights of computer science
is that the same program can run on many dif-
ferent machinesthat is, the same computation
can be performed many different ways. When
we consider a system like ship navigation,
however, the situation is complicated by a nest-
ing of computational systems. What is the im-
plementational level for the navigation system
as a whole is the computational level for the
people who operate the tools of the system. The
material means in which the computation is ac-
tually performed are implementational details
for the system, but they set the task constraints
on the performance of the navigation staff. The
distinction between what is computed by the
system as a whole and what is computed by the
 252/1540

individual navigation practitioners in the system

will be developed in later chapters. For the mo-
ment, let us take it simply as a justification for
attending to a level of detail that is often missing
from accounts of organizations as computational
systems.
A Computational Account of Navigation

In a computational sense, all systems of navigation

question "Where am I?" in fundamentally the same
representational assumptions of the navigation syst
this question is answered are enormously variable a
their ingenuity, all of them answer the question by c
dimensional constraints on position.

The surface of the sea is, of course, actually a three

surface on a nearly spherical body, the earth. As lon
concerned only with positions on this surface, we n
mensions to uniquely specify a position. Thus, a mi
one-dimensional constraints are required to specify
ship navigation. Navigating in three dimensionsa ra
activityrequires at least three one-dimensional cons
cify position.

Lines of Position
 254/1540

Figure 2.1 depicts the one-dimensional constraint th

by a known position and a given direction. Such a c
produces a line of position. Thus, if we know that p
particular direction from known position A, we kno
lie on a line extended from A in the specified direct
constraint alone, however, we still don't know wher
ally is; we know only that it must lie on the line of
by point A and the specified direction. If, for examp
that a treasure is buried due east of a certain split ro
are considerably narrowed but we still don't know w

Figure 2.1 Graphical and conceptual depiction of the line-of-

255/1540
Circles of Position

Figure 2.2 shows another type of one-dimensional c

one consists of a known position and a specified dis
defines a circle of position. If we know that point B
cified distance from point A, then we know that B m
circle of position centered on A with a radius of the
tance. Given this constraint alone, we cannot yet lo
only that it is somewhere on the circle of position s
known point and the distance from the point. In pra
position is often plotted as an arc in the vicinity of t
tion of point B rather than as a complete circle.

Combining Positional Constraints: Position Fixing

 257/1540

One-dimensional constraints can be combined in m

produce two-dimensional constraints on position. F
some of the possibilities. In the Western tradition, t
constraint is the computational basis of position fix
bearings and by radio direction finding (figure 2.3a
cedures, position is determined by finding the inters
more lines of position. A radar fix is constructed fro
a range (figure 2.3b). The circle-of-position constra
celestial navigation, although the circles of position

Figure 2.2 Graphical and conceptual depiction of the arc-of

 258/1540

Figure 2.3 A conceptual depiction of the combinations of

constraints.
 Page 54

are so large that they are treated as lines of posi-

tion in the vicinity of the fix (figure 2.3c). In a
celestial sight reduction, each observed celestial
body defines a circle of position, and the vessel
from which the observations were made must be
located at the intersections of the circles of posi-
tion established with respect to celestial bodies.
 260/1540

Systems such as Loran, Decca, and Omega

measure time or phase differences between the
arrival of signals from multiple stations. Con-
sider position fixing by Loran. If stations A and
B emit signals at precisely the same time, where
must I be if I receive the signal from station A 3
microseconds before I receive the signal from
station B? The answer is that I must be some-
where on a hyperbolic line of position that is
defined by all the intersections of circles of pos-
ition around A and B for which the circle of po-
sition around station A is 3 microseconds closer
to station A (at the speed of light) than the circle
of position around station B is to station B. Each
pair of stations received provides a time differ-
ence that defines a hyperbolic line of position.
The vessel's position is fixed by finding the in-
tersection of two or more such one-dimensional
lines of position. Radar combines a circle of po-
sition, expressed as a range (distance), with a
line of position, expressed as a bearing
 261/1540

(direction), to provide a two-dimensional con-

straint on the relative position of the object
detected.

The Position-Displacement Constraint

 262/1540

Two other important questions in navigation are

"Given that we are where we are, how shall we
proceed in order to arrive at a particular some-
where else?" and "Given that we are where we
are, where shall we be if we proceed in a partic-
ular way for a particular period of time?" Both
of these questions concern relationships among
positions. To answer the first is to use the spe-
cification of two positions to determine the rela-
tionship between them. To answer the second is
to use the specification of a position and a posi-
tional relationship to determine the specification
of another position. Both of these constraints are
captured by a single constraining relationship
that holds among positions and the spatial dis-
placements that lie between them. Figure 2.4 de-
scribes this constraint. It simply says that the
specification of any two of the items in the rela-
tionship fully constrains the specification of the
third item. There is no commitment to represent-
ation or algorithm in this. Positions
263/1540
 Page 55

Figure 2.4 A conceptual depiction of the position and dis-

placement constraint.
 265/1540

and displacements may be represented in a wide

variety of ways; however, if they are to be part
of a system that does navigation, they will have
to be represented in a way that satisfies this gen-
eric constraint. Things work out especially
nicely if the displacement is given in the form of
a direction and a distance. Then the determina-
tion of a new position from a given position and
a displacement is simply the familiar case of
combining the one-dimensional constraint
defined by the starting position and a direction
and a second one-dimensional constraint defined
by the same starting position and a distance. Let
me illustrate the satisfaction of this constraint
with two procedures from the Western tradition,
course planning and dead reckoning.

COURSE PLANNING
 266/1540

The fact that the specification of any two posi-

tions uniquely constrains the displacement that
lies between them is the basis of course plan-
ning. If I know where I am and where I want to
be, how can I determine a plan that will get me
where I want to go? In some representational
systems it is possible to compute a description
of how to move from one position to the other
from the description of the displacement
between two positions. For example, on some
types of nautical charts it is easy to measure the
direction (course) and the distance between any
two locations represented on the chart. Starting
at one point and sailing the specified course for
the specified distance will deliver the traveler to
the other point. In this case, the representational
medium, the chart, has been carefully designed
so that an easily obtained description of the dis-
placement between positions is also a descrip-
tion of a plan for getting from one point to the
other. We tend to take this property for granted,
 267/1540

but it is itself an impressive technical

accomplishment.
 Page 56

To appreciate what a nice property it is that dis-

placement on a chart is a plan for travel, one
need only consider how often such is not the
case. If the positions are represented as street
addresses to be looked up in a phone book, for
example, it may not be easy to get any descrip-
tion of the displacement between them at all.
And if one can construct a description of the dis-
placement from the addresses, unless the places
are on the same street it is unlikely that the de-
scription by itself will be a useful plan of travel.

DEAD RECKONING
 269/1540

The fact that the specification of a position and a

displacement uniquely constrains another posi-
tion is the basis of dead reckoning. In dead reck-
oning the navigator monitors the motion of the
vessel to determine its displacement from a pre-
vious position. If the distance and the direction
of the vessel's travel can be determined, the
measured displacement can be added to the pre-
vious position to determine the current position.
Or a planned future displacement can be added
to the current position to determine a future pos-
ition. Thus, if I know where I started and in
which direction and how far I have traveled, I
can compute my position.
 270/1540

According to Bowditch (1977), the term "dead

reckoning" is derived from deduced or ded reck-
oning, a procedure (predating modern charts) in
which a ship's position was computed, or de-
duced, mathematically from a displacement and
a known starting position. Even though modern
charts permit simple graphical solutions to this
problem, the term "dead reckoning" remains.
And even though the representation of informa-
tion and the procedure used in the computation
changed with the advent of modern charts, both
the old and the new version of dead reckoning
are based on the satisfaction of the position-dis-
placement constraint.

Depth-Contour Matching
 271/1540

There is one additional one-dimensional con-

straint to consider. Nautical charts sometimes
have depth contours indicating lines of equal
depth of water. If the depth of the water under a
ship can be measured, the position of the ship
can be constrained to be over a contour of that
depth. This is a one-dimensional constraint al-
though the line that defines it is usually not a
straight line. The utility of this method depends
on the shape of the bottom of the sea in the area.
If it is a featureless plain, the constraints im-
posed by
 Page 57

measured depth are weak. Almost any location

on the chart will satisfy the constraint imposed
by the measured depth, because the measured
depth is almost everywhere the same. If there
are many hills and valleys of nearly equal size,
then again there may be many contours in many
locations that satisfy the constraint. An inclined
plane with a moderate slope is a useful bottom
shape for simple contour navigation. A measure-
ment of depth in such an area yields a one-di-
mensional constraint that is typically combined
with other one-dimensional constraints, such as
circles or lines of position, to generate an estim-
ated position. Another useful bottom shape is
encountered in the Central Pacific, where a uni-
form abyssal plain is dotted with small raised
plateaus called guyots. There, one can hop from
guyot to guyot, identifying them by their depths.
 273/1540

If the depth-measuring apparatus is more soph-

isticated and can match changing depth contours
against patterns of changing depths rather than
simply matching single depth measurements
against single contours, then additional features
on the bottom may provide additional con-
straintsenough, in fact, to permit a twodimen-
sional position determination from depth data
alone. A similar sort of positional constraint can
be achieved on land through the use of an alti-
meter and a topographic map of terrain that in-
cludes altitude contours.

The Distance-Rate-Time Constraint

 274/1540

Just one more constraint is required to complete

the description of the computational core of nav-
igation. This constraint relates distance, rate,
and time. Figure 2.5 shows the form of this con-
straint. As with the constraint that holds among
positions and displacements, the specification of
any two values uniquely constrains the value of
the third. The constraint on distance, rate, and
time is

Figure 2.5 A conceptual depiction of the distance-rate-

time constraint.
 Page 58

often used to determine the distance portion of a

planned displacement in dead reckoning. This is
a commonly used constraint in the Western cul-
tural tradition outside the realm of navigation as
well. It is an important part of logistical plan-
ning, for organizations and individuals alike. If I
walk 4 miles per hour, how far can I get during
a 50-minute lunch break? How long will it take
me to drive the 118 miles to Los Angeles if I
can average 50 miles per hour? If the circumfer-
ence of the earth's orbit is 584 million miles,
how fast is the earth moving along its orbital
track?

Summary of Constraints
 276/1540

The computational account of navigation con-

sists of four principal constraints. Two of them
provide one-dimensional constraints on position
from a given position and a component of a spa-
tial displacement. The third relates positions and
the spatial displacements that lie between them,
where a displacement is composed of a pair of
one-dimensional constraints (one a distance and
the other a direction). The fourth constrains the
relations among distance, rate, and time as de-
scriptions of the motion of an object. Part of the
art of the actual practice of navigation lies in in-
tegrating information from many kinds of simul-
taneous constraints to produce a single solution
that satisfies them all.

Representational Assumptions of Western

Navigation
 277/1540

This section and the next describe sets of struc-

tures that have arisen in the Western cultural tra-
dition in terms of which the computational con-
straints outlined above are represented. The ac-
tual mechanics of the techniques for propagating
the constraints across representational structures
will be discussed in detail in a later chapter.

Units and Frames of Reference

In Western navigation, the units of direction are

based on a system of angular measurement. This
abstract system consists of a circle composed of
360 equal angular units called degrees. By con-
vention, north is 0 degrees, east is 90 degrees,
south is 180, and west is 270 degrees. Tradition-
al magnetic compasses had 32 named com-
Page 59
 279/1540

pass points. If the compass rose is oriented to

true north and south (as defined by the geo-
graphic poles), the directions are called true dir-
ections. If the compass rose is oriented to mag-
netic north or south (as defined by the magnetic
poles), the directions are called magnetic direc-
tions. The magnetic north pole is currently west
of Greenland, about 15° from the north geo-
graphic pole; the south magnetic pole is off the
coast of Antarctica, toward Australia, about 22°°
from the south geographic pole. These differ-
ences between the locations of the geographic
and magnetic poles cause magnetic instruments
to show considerable but largely predictable and
compensable errors in some locations. For finer
resolution, each degree is subdivided into 60
equal minutes of arc, and each minute is further
subdivided into 60 equal seconds. Thus, a
second of arc is 1/1,296,000 of a full circle. It is
not widely realized that the coordinates of geo-
graphic position (latitude and longitude) and the
 280/1540

basic unit of distance in modern navigation (the

nautical mile) are based on this same system of
angular measurement.

GEOGRAPHIC POSITION

The coordinate system in which locations on the

face of the earth are specified is based on a map-
ping of this circle onto the earth itself. Every
location has a latitude and a longitude. The latit-
ude of a place is its angular distance from the
equator. Points on the equator itself have latit-
ude 0°°. The north and south poles, which are
defined by the axis of rotation of the planet, are
each a quarter of a circle away from the equator
and are therefore at latitude 90°°. Locations in
the northern hemisphere are said to have north
latitude; those in the southern hemisphere have
south latitude.
 281/1540

A geometric plane passed through the earth such

that it contains the Planet's axis of rotation will
define two meridians where it intersects the
earth's surface. Longitude is the angular distance
of the meridian of a place from an arbitrarily se-
lected meridian that passes through Greenwich,
England.

The Greenwich or Prime meridian defines lon-

gitude 0°; its partner, stretching down the Pa-
cific ocean on the other side of the globe,
defines longitude 18°. Locations that lie in the
180°to the west of Greenwich are given west
longitudes; those in the 180°to the east are given
east longitude. Positions are given in terms of
these two one-dimensional constraints. Global
positions are specified in terms of this general
system. Specific positions are fixed, as
 Page 60

described in the examples given in the previous

section, by their relation to actual locations in
the immediately surrounding local space. The
nautical chart is a medium in which the specific-
ation of positions can be transformed from the
local to the global and vice versa.

THE NAUTICAL MILE

 283/1540

The nautical mile, the primary unit of distance

in maritime navigation, is based on this system
of angular measurement. A nautical mile is one
minute of arc on the surface of the earth. Thus,
there are 360 x 60 = 21,600 nautical miles
around the circumference of the earth. The size
of this unit has varied historically with vari-
ations in the estimation of the size of the earth.
Columbus and Magellan assumed a smaller
earth that had 45.3 modern nautical miles per
degree of latitude. The earth turned out to be
about 32 percent larger than they thought. The
statute mile (now established as 5,280 feet in the
United States) is a descendant of an earlier Ro-
man mile that was also intended to be 1/21,600
of the circumference of the earth. As measure-
ment of the earth improved and previous estim-
ates were found to be in error, there were pro-
posals to change the length of the mile itself and
proposals to change the number of miles in a de-
gree. For navigation at sea the easy mapping
 284/1540

between position descriptions by angular dis-

placement and the size of the major unit of dis-
tance is extremely useful. Having one minute of
arc equal one nautical mile simplifies many
computations at sea. Since this relationship
would have been destroyed by changing the
number of miles in a degree, the length of the
nautical mile was changed. The modern nautical
mile6,076.11549 feetis an attempt to preserve
that relationship. However, the modern nautical
mile is still an approximation. Because the earth
is not a sphere, the length of a minute of latitude
varies from about 6080.2 feet at the equator to
6108 feet at the poles. One minute of longitude
at the equator is about 6087 feet. The current
nautical mile is 1852 meters exactly. (Bowditch
1977)
 285/1540

The knot, or nautical mile per hour, is the stand-

ard unit of velocity in navigation. This knot ties
the circumference of the earth to the angular ve-
locity of the earth. Because an hour is 1/24 of a
day (a complete revolution of the earth), a point
on the surface of the earth at the equator moves
to the east at a rate of 1/24 of the circumference
of the earth in nautical miles per hour. That is,
900 knots.
 Page 61

Charts

In the Western tradition of pilotage, virtually all

computations involving position are carried out
on nautical charts. While there are many other
ways to represent the data and carry out the
computations of navigation, the chart is the key
representational artifact. The most obvious
property of maps and charts is that they are spa-
tial analogies. Positions on a map or a chart
have correspondence with positions in a depic-
ted large scale space. That is always true. But
charts designed for navigation are something
more than this. A navigation chart is a carefully
crafted computational device.
 287/1540

In algebra and analytic geometry, many compu-

tations can be performed on graphs; in fact,
graphs are essential in motivating the symbolic
manipulations that form the real heart of compu-
tation in algebra and analytic geometry. One can
compute all the points that lie between any two
points by drawing a line between them. Or one
can identify all the points that lie at a specified
distance d from a given point by drawing a
circle with radius d around the given point.
Using graphs for computation, however, intro-
duces errors, because plotted lines are less pre-
cise than the abstractions they depict (infinitely
small points and truly one-dimensional lines).
The infinite set of points lying between any two
given points is accurately and economically rep-
resented by the equation of the line that contains
the two points (and a range on x or y to constrain
points to be between the reference points), and
the set of points lying a specified distance from
a given point is accurately and economically
 288/1540

represented by the quadratic equation for the

circle. Of course, the utility of these representa-
tions depends on the subsequent computations
that they are supposed to support and the sort of
computational systems that are available to
carry out the computation.

It is essential to realize that a nautical chart is

more akin to a coordinate space in analytic geo-
metry than to the sort of simple map I may pro-
duce to guide a new acquaintance to my office.
All maps are spatial analogies in the sense that
they preserve some of the spatial relationships
of the world they depict, but navigation charts
depict spatial relationships in special ways that
support certain specialized computations. A
navigation chart is an analog computer. Clearly,
all the problems that are solved on charts could
be represented as equations and solved by
symbol-processing techniques. Plotting a posi-
tion or a course on a nautical chart is just as
much a
289/1540
 Page 62

computation as solving the set of equations that

represent the same constructs as the plotted
points and lines. A chart contains an enormous
amount of informationevery location on it has a
specifiable address, and the relationships of all
the locations to all the others are implicitly
represented.
 291/1540

Finally, charts introduce a perspective on the

local space and on the position and motion of
the vessel that is almost never achieved directly
by any person. Standing over a chart, one has a
''bird'seye" view that, depending on the scale of
the chart, could be duplicated with respect to the
real space only from an aircraft or a satellite.
Furthermore, the perspective is that of a spectat-
or rather than that of a participant. This is one
reason why establishing the correspondences
between the features on a chart and the features
in the local space is so difficult. In order to re-
concile the chart to the territory, one must ima-
gine how the world that is seen from a location
on the surface would appear from a point of
view from which it is never seen. The chart de-
piction assumes a very different perspective
than that of the observer on the vessel. The ex-
perience of motion for the observer on the ves-
sel is of moving through a surrounding space,
while the depiction of motion on a chart is that
 292/1540

of an object moving across a space. This other

perspective created by the chart is so compelling
that a navigator may have difficulty imagining
his movements, especially over large spaces,
from the traveler's perspective. Conversely,
people who have had no experience with maps
and charts may find them completely baffling.

THE COMPUTATIONAL PROPERTIES OF

CHART PROJECTIONS

Not all charts are equally useful for all sorts of

computations. For example, compare rhumb-
line sailing with radio-beacon navigation.

Rhumb-line sailing
 293/1540

A rhumb line is a line on the surface of the earth

that represents a constant direction from some
location. Rhumb-line sailing refers to a form of
navigation in which one sets a course to a des-
tination and then maintains a constant heading
until the destination is reached. When one is
steering a ship by any sort of compass, the
simplest route is a constant heading. For this
task it is very useful to have a chart on which
rhumb lines are straight lines. However, if one
were to plot the course that would result from
steering a constant heading on a globe rather
than on a chart, one would produce a line that
 Page 63

wraps around the globe and spirals up to the

pole. This line is called a loxodrome.
 295/1540

The Mercator projection overcomes this prob-

lem and transforms the spiral into a straight line.
Imagine the transformation in two steps. First,
the meridians of longitude that actually con-
verge with one another at the poles of the globe
are made parallel to one another, so that they are
just as far apart at high latitudes as they are at
the equator. This introduces a systematic distor-
tion. At the equator there is no distortion, but
with increasing latitude the east-west distance
shown between the meridians on the chart ex-
ceeds by an increasing margin the distance
between them on the globe. At the poles, of
course, the distortion is infinitewhat was zero
distance between meridians where they con-
verged at the pole of the globe would appear as
a finite distance on the chart. To compensate for
the effects of this distortion on direction, the
parallels of latitude are expanded by the same
ratio as the meridians of longitude. At the poles
this would require infinite expansion, which is
 296/1540

why the poles never appear on Mercator projec-

tions. This expansion also results in a distortion
of the relative areas depicted on the chart. This
distortion is more pronounced at higher latit-
udes. Thus, while Greenland actually has only
1/9 the area of South America, they appear to
have roughly the same area on a Mercator chart.

Radio-beacon navigation
 297/1540

Radio beacon navigation uses radio antennae

that are sensitive to direction. Such an antenna
can determine the direction from which it re-
ceives a radio signal. By tuning the antenna to a
station whose location is known and identifying
the direction from which the signal comes, one
can establish a one-dimensional position con-
straint. However, a radio signal does not follow
a rhumb line; it takes the shortest route. These
shortest routes are called great-circle routes. A
great-circle route is defined by the intersection
with the surface of the earth of a plane that con-
tains the center of the earth and the two points
on the surface between which the route is to be
constructed. Great-circle routes can be approx-
imated by stretching a piece of yarn over the
surface of a globe. The meridians of longitude
define great circles, and so does the equator. All
the circles of latitude other than the equator
define rhumb-line courses that are not great
circles. While the rhumb-line course from Los
 298/1540

Angeles to Tokyo is almost exactly due west

(and the heading is constant for the entire trip),
the great-circle route leaves Los Angeles head-
ing to
 Page 64

the northwest and arrives in Tokyo heading to

the southwest. To plot a position from radio-
beacon bearings, one would like a chart on
which great circles are straight lines. Over short
distances, great circles approximate straight
lines on all projections; however, over long dis-
tances (and radio signals travel long distances)
great circles are significantly different from
rhumb lines. There is no chart projection on
which both rhumb lines and great circles appear
as straight lines.

In addition to the properties of having rhumb

lines and great circles represented as straight
lines, it is easy to imagine navigation tasks in
which the following would be desirable chart
properties:

true shapes of physical features

correct angular relationships among positions

 300/1540

equal area, or the representation of areas in their

correct relative proportions

constant scale values for measuring distances

Whenever the three-dimensional surface of the

earth is rendered in two dimensions, some of
these properties are sacrificed. For example, the
Mercator projection sacrifices true shape of
physical features, equal area, and constant scale
values for measuring distances in the interest of
providing correct angular relationship and
rhumb lines as straight lines. These features are
most apparent on charts of large areas. As the
area of the earth's surface represented by the
chart decreases, the differences between projec-
tions becomes less noticeable.
 301/1540

Chart projections make it clear that different

representational systems have different compu-
tational properties and permit differing imple-
mentations of the computations. For example, it
is possible to draw a great circle on a Mercator
projection; it is just very difficult to compute
where the points should go. On a Lambert con-
formal chart it is quite easy to draw a great
circle, because on this projection a straight line
so nearly approximates a great circle that it is
more than adequate for navigational purposes.
One can see the work that went into construct-
ing a chart as part of every one of the computa-
tions that is performed on the chart in its life-
time. This computation is distributed in space
and time. Those who make the chart and those
who use it are not known to one another (per-
haps they are not even contemporaries), yet they
are joint participants in a computational event
every time the chart is used.
302/1540
 Page 65

Summary
 304/1540

Large-scale space is represented as small-scale

space on a chart. The primary frame of reference
is the system of earth coordinates. Objects that
are unmoving with respect to earth coordinates
are given fixed locations on the chart. Every loc-
ation can be assigned an absolute address in a
global coordinate system. Direction, position,
and distance are all defined in terms of a single
universal framework, established by applying a
scheme of angular measurement to the earth it-
self. A universal time standard in combination
with the measurement of distance yields a uni-
versal unit of rate of movement. These units are
universal in the sense that their interpretations
do not change with changing location or circum-
stances of their use. Directions, positions, dis-
tances, and rates can all be represented as num-
bers, and any of the first three can also be
modeled in the small-scale space of a chart.
Line-of-position constraints are represented as
lines on a chart; circles of position are
 305/1540

represented as circles on a chart; position-dis-

placement constraints are represented as posi-
tions and displacements on a chart. Distance,
rate, and time are represented as numbers, and
computations of the constraints among them are
accomplished by digital arithmetic algorithms.
All the major computations in this system are
based on procedures that involve measurement
(which is analog-todigital conversion), followed
by digital manipulation, followed by digital-to-
analog conversion in the plotting of results on a
chart.

Representational Assumptions of Micronesian

Navigation
 306/1540

The computational account presented above also

describes the computations carried out by Mi-
cronesian navigators (Hutchins 1983). Microne-
sian navigators establish their position in terms
of the intersections of one-dimensional con-
straints. Substantial differences between
Western and Micronesian navigation become
apparent as soon as we consider the representa-
tions and the algorithms that the two cultural
traditions have developed to satisfy the con-
straints of the task. A major problem with earlier
Western studies of Micronesian navigation was
that the representations used in the performance
of Western navigation were assumed to be the
most general description. Because they failed to
see the computational level at all Gladwin
(1970), Lewis (1972), Sarfert (1911), and
Schück (1882) attempted to interpret the repres-
entations used in
 Page 66

Micronesian navigation in terms of the repres-

entations used in Western navigation, rather
than interpreting both sets of representations in
terms of a single, more general, computational
account.
 308/1540

This brief discussion of Micronesian navigation

is inserted here in the hope that it will make the
importance of the distinction between the com-
putational and representational level of descrip-
tion clearer. I also hope to show that even the
most commonplace aspects of thinking in
Western culture, as natural as they may seem,
are historically contingent. In this light, the or-
ganization of systems of cultural representations
may become visible and, once noticed, may
come to seem much less obvious than before.
Furthermore, because the representational and
implementational levels constrain each other
more closely than do the computational and rep-
resentational, it is useful to see the relationship
between the representational level and its imple-
mentation in cultures that are technologically
quite different from each other.
 309/1540

For more than a thousand years, long-distance

non-instrumental navigation has been practiced
over large areas of Polynesia and Micronesia,
and perhaps in parts of Melanesia. In Polynesia,
the traditional techniques atrophied and were ul-
timately lost in the wake of contact with coloni-
al powers. Only the Micronesians have main-
tained their traditional skills, and in the past two
decades they have been the wellspring of navig-
ation knowledge for a renaissance of traditional
voyaging throughout the Pacific Basin (Finney
1979, 1991; Kyselka 1987; Lewis 1976, 1978).
 310/1540

Without recourse to mechanical, electrical, or

even magnetic devices, the navigators of the
Central Caroline Islands of Micronesia routinely
embark on ocean voyages that take them several
days out of the sight of land. Their technique
seems at first glance to be inadequate for the job
demanded of it, yet it consistently passes what
Lewis (1972) has called "the stern test of land-
fall." Of the thousands of voyages made in the
memory of living navigators, only a few have
ended with the loss of a canoe. Western re-
searchers traveling with these people have found
that at any time during the voyage the navigat-
ors can accurately indicate the bearings of the
port of departure, the destination, and other is-
lands off to the side of the course being steered,
even though all of these may be over the horizon
and out of sight. These navigators are also able
to tack upwind to an unseen island while keep-
ing mental track of its
311/1540
 Page 67

changing bearing-a feat that is simply im-

possible for a Western navigator without
instruments.

In the neighborhood of the Caroline Islands, less

than 0.2 percent of the surface is land. The sur-
face is a vast expanse of water dotted with about
two dozen atolls and low islands. Experienced
navigators in these waters routinely sail their
outrigger canoes up to 150 miles between is-
lands. The knowledge required to make these
voyages is not held by all, but is the domain of a
small number of experts.
 313/1540

The world of the navigator, however, contains

more than a set of tiny islands on an undifferen-
tiated expanse of ocean. Deep below, the pres-
ence of submerged reefs changes the apparent
color of the water. The surface of the sea undu-
lates with swells born in distant weather sys-
tems, and the interaction of the swells with is-
lands produces distinctive swell patterns in the
vicinity of land. Above the sea surface are the
winds and weather patterns which govern the
fate of sailors. Seabirds abound, especially in
the vicinity of land. Finally, at night, there are
the stars. Here in the Central Pacific, away from
pollution and artificial light, the stars shine
brightly and in incredible numbers. All these
elements in the navigator's world are sources of
information. The whole system of knowledge
used by a Micronesian master navigator is well
beyond the scope of this book. Here I will treat
only a portion of the navigators' use of celestial
cues.
 314/1540

The most complete description of this system

comes from the work of Thomas Gladwin, who
worked with the navigators of Puluwat Atoll in
what is now the Republic of Micronesia. Glad-
win (1970) divides the pragmatics of Puluwat
navigation into three parts. First one must set
out in a direction such that, knowing the condi-
tions to be expected en route, one will arrive in
the vicinity of the island of destination. Second,
one must hold the canoe steady on its course and
maintain a running estimate of its position. Fin-
ally, when nearing the destination one must be
able to locate it and head toward it.
 315/1540

One of the most widespread notions employed

in Pacific non-instrumental navigation is the
concept of the star path. From the point of view
of the earth, the positions of the stars relative to
one another are fixed. As the earth rotates about
its axis, the stars appear to move across the sky
from east to west. As the earth moves through
its orbit around the sun, the stars that can be
seen at night
Page 68
 317/1540

(that is, from the side of the earth away from the
sun) change. But from any fixed location on the
earth, any given star always rises from the same
point on the eastern horizon and always sets into
the same point in the western horizon, regard-
less of season. Movement to the north or south
does change the azimuth of the rising and set-
ting of any star. Within the range of the Caroline
Islands navigator, however, the effects of such
movements are small (on the order of 3° or less).
A star path, also known as a linear constellation
(Aveni 1981), is a set of stars all of which "fol-
low the same path" (Gladwin 1970). That is,
they all rise in succession from the same point
on the eastern horizon, describe the same arc
across the sky, and set into the same point on the
western horizon. Star paths are typically com-
posed of from six to ten stars fairly evenly
spaced across the heavens (Lewis 1972). Thus,
when one star in the linear constellation has ris-
en too far above the horizon to serve as an
 318/1540

indication of direction, another will soon take its

place. In this way, each star path describes two
directions on the horizon, one in the east and
one in the west, which are visible regardless of
season or time of night as long as the skies are
clear. A "connect the dots" drawing of such a
linear constellation is simply an arc across the
sky, anchored at fixed azimuths in the east and
in the west. While the stars themselves make
their nightly journeys across the sky, the arcs of
the linear constellations remain stationary.

Seeing the night sky in terms of linear constella-

tions is a simple representational artifice that
converts the moving field of stars into a fixed
frame of reference.
 319/1540

This seeing is not a passive perceptual process.

Rather, it is the projection of external structure
(the arrangement of stars in the heavens) and in-
ternal structure (the ability to identify the linear
constellations) onto a single spatial image. In
this superimposition of internal and external,
elements of the external structure are given cul-
turally meaningful relationships to one another.
The process is actively constructive. The posi-
tions of a few stars may suggest a relationship
which, when applied, establishes the identity of
yet other stars. Anyone who can identify the tra-
ditional Western constellations knows that, in
the subjective experience of this seeing, not just
the stars but the constellations themselves seem
to be "out there." The little lines holding the
stars together seem nearly visible in the sky.
These relations are expressed in verbal formu-
las. For example, the formula "Follow the arc
(of the handle of the Big
320/1540
Page 6
 322/1540

Figure 2.6 A Caroline Island sidereal compass.

Dipper) to Arcturus, and drive a spike into Spica"

guides the observer's eye across the sky, construct-
ing part of a constellation. In sky charts for amateur
star watchers, the lines are drawn in on the charts-
like mental training wheelsto make the constella-
tions easier to imagine when looking at the sky.
 323/1540

It is known that star paths have long been used to

define the courses between islands in many parts of
Oceania (Lewis 1972). The navigators of the
Caroline Islands have combined fourteen named
star paths with the position of Polaris (the North
Star) to form a sidereal compass that defines 32 dir
ections around the circle of the horizon. Figure 2.6
shows a schematic representation of the Caroline
Island sidereal compass. As can be seen, most of
the recognized star bearings are named for major
stars whose paths intersect the horizon at those
points. Those which are not so named are the true-
north bearing, named for Polaris which from the
Caroline Islands is always about 8° above the north
ern horizon, and three bearings in the south which
are defined by orientations of the Southern Cross
above the horizon. Of course, the names given to
these stars are not the same as the names given to
them in the
 Page 70

Western tradition, nor are all the constellations

grouped in the same way. The cardinal direction
in the Micronesian system is east, at the rising
point of the star Altair. It is interesting that
Altair is part of a Micronesian constellation
called the "Big Bird" (hence Gladwin's title East
is a Big Bird). The Western tradition has inher-
ited many of its star names from Arabic roots,
and Altair is the brightest star in the constella-
tion Aquila, the eagle. East was the cardinal dir-
ection in the Western tradition (consider the two
meanings of the word 'orient') before the advent
of the magnetic compass.
 325/1540

The inclusion of other stars which travel the

same path guarantees that as long as the weather
is clear the complete compass is available to the
navigator no matter what time of year he is sail-
ing. In fact, a practiced navigator can construct
the whole compass mentally from a glimpse of
only one or two stars near the horizon. This abil-
ity is crucial to the navigator's performance, be-
cause the star bearings that concern him during
a voyage may not be those he can readily see.
The star compass is an abstraction which can be
oriented as a whole by determining the orienta-
tion of any part. During the day, the orientation
of the star compass can be maintained by ob-
serving the star bearings from which the major
ocean swells come and the star bearings at
which the sun and the moon rise and set.
 326/1540

Courses between islands are defined in terms of

this abstract sidereal compass. For every island
in a navigator's sailing range, he knows the star
point under which he must sail to reach any oth-
er island in the vicinity. Thus, the sidereal com-
pass provides the directional reference in terms
of which displacements can be specified.
 327/1540

The sidereal compass has a second function in

navigation: the expression of distance traveled
on a voyage. For every course from one island
to another, a third island (over the horizon and
out of sight of the first two) is taken as a refer-
ence for the expression of the distance traveled.
In the language of Puluwat Atoll, this system of
expressing distance traveled in terms of the
changing bearing of a reference island is called
etak (Gladwin 1970). Since he knows the star
bearings for all the inter-island courses in his
area, the navigator knows the star bearing of the
reference island from his point of origin and the
bearing of the reference island from his destina-
tion. In the navigator's conception, this reference
island starts out under a particular star (at a par-
ticular star bearing) and moves back
Figure 2.7 An etak diagram. This diagram, based on the work
nographer E.
Sarfert, reflects the conventional method of drawing the rel
between islands and star points for a typical voyage
 329/1540

abeam of the canoe during the voyage through a su

of star bearings until the canoe reaches its destinatio
which time the reference island is under the point th
defines the course from the destination island to the
ence island. The changing star bearing of the refere
land during the voyage is illustrated in figure 2.7.

The movement of the reference island under the suc

of star bearings divides the voyage conceptually int
segments called the etaks of the voyage. Each voya
known number of etak segments defined by the pas
the reference island under the star bearings.

A fundamental conception in Caroline Island navig

that a canoe on course between islands is stationary
islands move by the canoe. This is, of course, unlik
tion of the vessel moving between stationary island
sage from Gladwin (1970: 182) amplifies this:
 330/1540

Picture yourself on a Puluwat canoe at night. The w

clear, the stars are out, but no land is in sight. The
a familiar little world. Men sit about, talk, perhaps
around a little within their microcosm. On either si
canoe, water streams past, a line of turbulence and
merging into a wake and disappearing into the dark
Overhead there are stars, immovable, immutable. T
swing in their paths across and out of the sky but in
come up again in the same places. You may
 Page 72

travel for days on the canoe, but the stars will

not go away or change their positions aside
from their nightly trajectories from horizon to
horizon. Hours go by, miles of water have
flowed past. Yet the canoe is still underneath
and the stars are still above. Back along the
wake however, the island you left falls farther
and farther behind, while the one toward which
you are heading is hopefully drawing closer.
You can see neither of them, but you know this is
happening. You know too that there are islands
on either side of you, some near, some far, some
ahead, some behind. The ones that are ahead
will, in due course, fall behind. Everything
passes by the little canoe-everything except the
stars by night and the sun in the day.
 332/1540

Here we have a conceptualization in which the

known geography is moving past the navigator,
his canoe, and the stars in the sky. Off to the
side of the course being steered is the reference
island. It cannot be seen because of its distance
over the horizon, yet the navigator imagines it to
be moving back slowly under a sequence of star
points on the horizon. Observations of navigat-
ors during voyages have shown that the navigat-
ors can accurately judge the relative bearing of
the reference island at any time during the voy-
age (Lewis 1972). Since the navigator has not
actually seen the reference island at any point
during the voyage, his ability to indicate where
it lies represents an inference that could not be
made in the Western system without recourse to
tools.

Gladwin (1970: 184) describes the Micronesian

navigator's use of this judgement as follows:
 333/1540

When the navigator envisions in his mind's eye

that the reference island is passing under a par-
ticular star he notes that a certain number of
segments have been completed and a certain
proportion of the voyage has therefore been
accomplished.

The navigator uses this information to estimate

when he will be in the vicinity of his destina-
tion, and therefore when he should start looking
for signs of land. Since land-based birds venture
as far as 20 miles to sea, seeing them arrive at a
fishing ground from land or seeing them depart
a fishing ground for land can give information at
a distance about the direction in which land lies.
This information is available only in the early
morning and at dusk, when the birds are moving
from or to their island. A navigator who arrives
at what he believes to be the vicinity of his des-
tination at midday is
 Page 73

therefore well advised to drop sail and wait for

dusk. The danger of failing to make an accurate
judgement of when land is near is that one might
be close to land when no indications are avail-
able and then sail past and be far away from the
destination when homing signs are available.

Because traditional Micronesian culture is non-

literate, navigators are required to commit a
large body of information to memory. Riesen-
berg (1972) has documented some of the elabor-
ate mnemonic devices used by navigators to or-
ganize their knowledge of geography, star
courses, and etak segments. An interesting find-
ing of Riesenberg's work is that the memorized
systems of knowledge make frequent reference
to islands that do not exist. Riesenberg (1972:
20) explains:
 335/1540

In a few instances, when unknown geographical

features were mentioned and when enough
courses from identifiable islands to them have
been given, an attempt has been made to locate
them by projecting the courses on a chart. The
intersections of the projected courses generally
coincide poorly with known bathymetric
features.

The role of these phantom islands will be taken

up in a later section of this chapter.

Some Anomalous Interpretations

The history of attempts to understand how the

Micronesian navigators accomplished their way-
finding feats reads like a detective story in
which we know who did it but not how it was
done. Each of several researchers has provided
us with both useful clues and a few red herrings.
 336/1540

There is little dispute about the nature of course-

keeping with the sidereal compass. Western ac-
counts of the star compass go back at least to
1722 (Schück 1882), and its use seems relat-
ively easy to observe and document. The most
detailed description of the star compass of the
Caroline Islands was provided by Goodenough
(1953). Although his diagram reproduced above
as figure 2.6 is, as far as we know, a completely
accurate depiction of the stars used by the
Caroline Island navigators, and although it gives
the first complete tabulation of the azimuths
(true bearings on the horizon) and names of the
star points, it contains a potentially misleading
distortion that was probably incorporated to
make the compass concept more accessible to
Western readers. Goodenough drew the
 Page 74

compass as a circular compass rose, the way

compasses are traditionally represented in our
culture. The original records of native depictions
of the star compass, however, are all box-
shaped.

To date there have been two attempts to explain

just how the Caroline Island navigators use the
concept of etak to keep track of their progress
on a voyage: Sarfert's (1911) and Gladwin's
(1970). Sarfert's (1911: 134) description is rich
and compact and bears careful consideration:
 338/1540

In an arbitrary voyage between two determined

islands, the native captains have still a third is-
land in mind, besides the starting point and goal
of the trip. For the voyage between every pair of
islands, this is a specific island. Henceforth, I
will refer to this island simply as ''emergency is-
land" [Notinsel] corresponding to the purpose
that it serves as a last place to flee to in case of
extenuating circumstances that make it im-
possible to reach either the starting point or
goal of the trip. This island is placed off to the
side of the course. In rare situations the natives
established two islands as emergency islands,
specifically in such a way that one lies to the left
and the other to the right of the direction of
travel.
 339/1540

Riesenberg's (1972) discovery that the reference

islands for some voyages are phantoms,
however, makes the "emergency island" inter-
pretation unlikely. No navigator would attempt
to take refuge in a location known to be devoid
of land. Another possibility is that knowing the
location of the reference island as well as the
origin and destination of the voyage allows the
navigator to estimate accurately where many
other islands in the area are, so that, should he
need to take refuge, a choice based on the exist-
ing conditions of the wind and the sea might be
made among several possible islands. The spe-
cification of the placement of the islands is no
doubt important; but if they were places in
which to take refuge, why would it not be just as
well to have two "emergency islands" on the
same side of the course?

Sarfert continues:
 340/1540

In figure [2.7 of this chapter-E.H.], the island

Biseras, a small island of the Onona atoll,
serves as emergency island in the already given
voyage from Polowat to Ruk [Truk]. If the emer-
gency island is to fulfill its purpose, the captain
must be capable of determining at any moment
the direction in which the island lies, and there-
fore the course to it, from an arbitrary point of
the voyage. As far as I
 Page 75

have experience about it, he ... does this by

rather simple means:

1) The direction of the island Biseras from Po-

lowat as well as from Ruk is known.

2) The native captain may undertake a bearing

of the area during the trip by means of calculat-
ing the already-traveled distance. This is done
with the aid of experience, knowledge of the
normal duration of the voyage and with the help
of an estimate of the speed that the canoe travels
through the water. This last means, the so-called
dead reckoning, was also in general used by us
for the same purpose before the introduction of
the log at the end of the sixteenth century.
 342/1540

3) To determine the bearing of the emergency

island from the vantage point of the canoe, the
observation must necessarily be done such that,
as [figure 2.7] clearly demonstrates, it describes
the emergency island Biseras, from the canoe as
a visible movement on the horizon in the oppos-
ite direction of the voyage. This visible move-
ment of the emergency island appears, with the
interpretation of the horizon as a straight line,
in direct relationship to the already-traversed
distance. If the captain estimates, for example,
the covered path as being a quarter of the total
voyage length, then the emergency island must
have completed likewise a quarter of its visible
path along the horizon. If the total length of the
visible path totals eight (etak) lines, then after
one quarter of the trip they would have reached,
accordingly, the third line. By means of this
simple calculation, the course to the emergency
island is confirmed and the captain is capable of
seeking it out. (135)
 343/1540

The major issue raised by Sarfert's proposed cal-

culation technique involves the method used to
express the proportion of the total voyage that
has been completed. It is easy enough to ima-
gine how the navigator might represent the fact
that "the emergency island must have completed
a quarter of its visible path along the horizon,"
although it is doubtful that proportions like "a
quarter" are involved. But how does the captain
compute that he has covered some proportion of
the total length of the voyage? Further, the ex-
pression of the movement of the emergency is-
land in terms of the proportion of the number of
etak segments will work only if the etak seg-
ments themselves are all nearly the same size.

Gladwin's descriptive model, like Sarfert's,

relates the bearing of the etak reference island to
the distance traveled. However, Sarfert believed
that the navigator computed the apparent bear-
ing of the
344/1540
 Page 76

etak island so that he could take refuge there,

whereas Gladwin asserted that the navigator
used that apparent position as an expression of
the proportion of the voyage completed. Glad-
win states:

When the navigator envisions in his mind's eye

that the reference island is passing under a par-
ticular star he notes that a certain number of
segments have been completed and a certain
proportion of the voyage has therefore been ac-
complished. (184)
 346/1540

This is similar to Sarfert's proportional-deriva-

tion model, but the subtle difference raises an
interesting question. What is the nature of the
computation? Is it, as Sarfert maintains, that the
navigator uses his estimate of the proportion of
the voyage completed to establish the bearing of
the reference island, or, as Gladwin maintains,
that the navigator uses his estimate of the bear-
ing of the reference island to establish the pro-
portion of the voyage that has been accom-
plished? Clearly, these concepts are closely re-
lated for the navigator.

In practice, not every inter-island course is situ-

ated such that there is an island to the side of the
course with the desired properties of an etak is-
land. Gladwin notes:
 347/1540

If the reference island is too close, it passes un-

der many stars, dividing the journey into a lot of
segments. Worse, the segments are of very un-
equal length. They start out rather long (slow)
and then as the canoe passes close by, they be-
come shorter (fast) as the reference island
swings under one star after another, and then at
the end they are long again, a confusing effect.
A distant reference island has the opposite effect
making the segments approximately equal, but
so few in number that they do not divide the
journey into components of a useful size. (187)
 348/1540

The effect of having a close reference island is

confusing because when a voyage is divided in-
to segments of very different lengths the estima-
tion of the number of segments remaining is a
poor measure of the distance remaining in the
voyage. Gladwin describes another situation,
also noted by Sarfert, in which this same sort of
confusion was bound to arise. In a discussion
with the master navigator Ikuliman, Gladwin
discovered that for the voyage between Puluwat
and Pulusuk atolls, a distance of about 30 miles,
the navigator used two etak islands-one to the
west of the course and nearby, the other to the
east and quite distant:
 Page 77

This case well illustrates one of the difficulties

with the practice: when two reference islands
are used in this way, the segments are almost
certain to be markedly different in length. Ikuli-
man was not able to offer a good explanation
for using two islands, insisting only that this is
the way it is taught. When I pressed him further,
he observed dryly that Puluwat and Pulusuk are
so close together that a navigator does not
really need to use ETAK at all in order to estab-
lish his position on this seaway, so in this case
my question was irrelevant. (188)
 350/1540

Another feature of the system in use that seems

to give rise to the same sort of conceptual diffi-
culty is that the first two and last two segments
of the voyages are all about the same length, re-
gardless of the positioning of the reference is-
land relative to the courses and regardless of the
density of star points in the portion of the hori-
zon through which the reference island is ima-
gined to be moving. Gladwin states:
 351/1540

Upon leaving an island, one enters upon the

"ETAK of sighting," a segment which lasts as
long as the island remains in view, usually
about 10 miles. When the island has at last dis-
appeared, one enters the "ETAK of birds" which
extends out as far as the flights of birds which
sleep ashore each night. This is about twenty
miles from land, making the first two and there-
fore also the last two, segments each about ten
miles long. Having four segments of the voyage
absolute in length is logically incongruous (by
our criteria) with the proportional derivation of
the remainder of the ETAK divisions. (188)
 352/1540

Again, the problem with this conception is that

it interferes with the computation of the distance
remaining in the voyage because it destroys the
consistency of the etak segments as units of dis-
tance. Gladwin explored this inconsistency with
his main informant, the navigator Hipourwho
later sailed with Lewis to Saipan and back using
the system described here (Lewis 1972, 1976,
1978). Gladwin continues:

When I tried to explore with Hipour how he re-

solved the discrepancy he simply replied that
beyond the ETAK of birds he uses the reference
island to establish distance. When I asked how
he handled the problem of segments ending in
different places, under the two methods, he said
he did not see this as a problem. As with
 Page 78

Ikuliman's answer to my "problem" over the

dual reference islands, this ended the discus-
sion. (189)

The major difficulty with Sarfert's model, and

all the "problems" that Gladwin raised with his
navigator informants, spring from the observa-
tion that etak segments are unsuitable units for
the measurement of distance covered on a voy-
age. One interpretation of this state of affairs is
that what appeared to be a logical organizing
principle in navigation may be a useful descrip-
tion in the abstract, but that in the exigencies of
use it is not strictly adhered to. Gladwin
concludes:
 354/1540

Although ETAK has for us much the quality of a

systematic organizing principle or even logical
construct, the Puluwat navigator does not let lo-
gical consistency or inconsistency, insofar as he
is aware of them, interfere with practical utility.
(189)
 355/1540

There is, of course, another possible interpreta-

tion: that the apparent anomalies result from the
unwarranted assumption that the etak segments
are units of measurement. The notion that con-
sistent units of measurement are necessary for
accurate navigation is one of the fundamental
representational assumptions of our system of
navigationso much so, in fact, that it is hard for
us to conceive of a system of navigation that
does not rely on such units and a set of opera-
tions for manipulating them. Yet there is no
evidence in the record that the etak segments
perform that function, nor is there any evidence
of any set of mental arithmetic operations that
would permit a navigator to manipulate etak
segments as though they were units of distance.

A Conceptual Blind Spot

 356/1540

The following revealing incident occurred while

Lewis was working with the master navigators
Hipour of Puluwat and Beiong of Pulusuk. Ac-
cording to Lewis:

On one occasion I was trying to determine the

identity of an island called Ngatikthere were no
charts to be consulted of coursethat lay
somewhere south-west of Ponape. It has not
been visited by Central Carolinian canoes for
several generations but was an ETAK reference
island for the OrolukPonape voyage and as
such, its star bearings from both these islands
were known to Hipour. On his telling me what
they were, I drew a diagram to illustrate that
Ngatik
 Page 7

Figure 2.8 Lewis's method of determining the position of the

island Ngatik.

must necessarily lie where these ETAK bearings in-

tersected. [See figure 2.8.] Hipour could not grasp
this idea at all. His concept is the wholly dynamic
one of moving islands. (1972: 142)
 358/1540

This passage raises several important questions:

Why did Lewis use the technique of drawing the in
tersecting bearings in order to determine the loca-
tion of the island called Ngatik? Why did Lewis as-
sume that posing the question the way he did would
make sense to Hipour? Why did Hipour not grasp
the idea of the intersecting bearings?
 359/1540

Let us consider the questions about Lewis first. The

technique Lewis used is clearly an effective one for
the solution of this particular problem. It establishe
a two-dimensional constraint on the location of Ng-
atik by combining two one-dimensional constraints
It also contains some very powerful assumptions
about the relation of the problem solver to the space
in which the problem is being solved. First, it re-
quires a global representation of the locations of the
various pieces of land relative to each other. In ad-
dition, it requires a point of view relative to that
space which we might call the "bird's-eye" view.
The problem solver does not (and cannot without an
aircraft) actually assume this relation to the world i
which the problem is posed. We can guess that
Lewis did this because it is for him a natural frame-
work in which to pose questions and solve prob-
lems having to do with the relative locations of ob-
jects in a two-dimensional space. Western navigat-
ors make incessant use of this point of view. When
a Western navigator takes a bearing of landmark, h
 360/1540

has a real point of view on a real space. However,

as soon as he leans over his chart, he is no longer
conceptually on the boat; he is over the sea surface,
looking down on the position of his craft in a rep-
resentation of the real local space. Novice
navigators
 Page 80

sometimes find this change of viewpoint disori-

enting, especially if the orientation of the chart
does not happen to correspond to the orientation
of objects in the world.

Beiong was also puzzled by Lewis's assertion,

and in reaching an understanding of it he
provides us with an important insight into the
operation of the Micronesian conceptual system:

He eventually succeeded in achieving the mental

tour de force of visualizing himself sailing sim-
ultaneously from Oroluk to Ponape and from
Ponape to Oroluk and picturing the ETAK bear-
ings to Ngatik at the start of both voyages. In
this way he managed to comprehend the dia-
gram and confirmed that it showed the island's
position correctly. (143)
 362/1540

The nature of Beiong's understanding indicates

that for the Caroline Island navigator the star
bearing of an island is not simply the orientation
of a line in space but the direction of a star point
from the position of the navigator. In order to
see that the star bearings would indeed intersect
each other at the island, he had to imagine him-
self to be at both ends of the voyage at once.
This allowed him to visualize the star bearing
from Oroluk to Ngatik radiating from a navigat-
or at Oroluk and the star bearing from Ponape to
Ngatik radiating from a navigator at Ponape.
What Hipour probably imagined when Lewis
asserted that the island lies where the bearings
cross must have been something like the situ-
ation depicted in figure 2.9. Contrast this with
what Lewis imagined he was asserting (figure
2.8). Hipour's consternation is now perhaps
more understandable. The star bearings of the
etak island radiate out from the navigator him-
self. From this perspective they meet only at
 363/1540

him. In his conception of this voyage, the etak

island begins under one of these bearings and
ends under the other. That two relative bearings
might meet anywhere other than at the navigator
is inconceivable.

Because the Caroline Island navigator takes a

real point of view on the real local space to de-
termine the star bearings, it does not seem likely
that the mapping of etak segments onto an ab-
stract representation of the expanse of water
between the islands is faithful to his conception.
Gladwin's (1970) statement about the
navigator's noting that "a certain number of seg-
ments have been completed" and the diagrams
that Lewis, Gladwin, and Sarfert use to repres-
ent the changing relative bearing of the etak ref-
erence island all contain two implicit assump-
tions: that the navigator uses
Figure 2.9 Hipour's way of thinking about star bearing
all bearings originate at himself and radiate outward. T
conception in the Western bird's-ey
 365/1540

some sort of bird's-eye view of the space he is in, an

in terms of changes in the position of his canoe in a
changing point of view. These assumptions are true
ception of a voyage, but they appear not to be true o
conception. These assumptions are at odds with the
the islands moving relative to the navigator) and wi
sternation in the face of what ought to be a trivial in

It is tempting to criticize the Caroline Island naviga

centric perspective on the voyage when the global p
much more powerful. Before concluding that the W
the following thought experiment: Go at dawn to a
the center of the rising sun. That defines a line in sp
place at noon and point again to the center of the su
space. I assert that the sun is located in space where
seem wrong? Do you feel that the two lines meet w
else? In spite of the fact that the lines seem to be or
cross at the sun. This is not intuitively obvious to u
ceiving of the
Figure 2.10 A heliocentric depiction (not to scale) of pointin
then again at noon.
The sun is indeed located where the lines

sun's location is not to conceive of its location at all

of its orientation relative to a frame defined by the h
zenith on earth. The rotation of the earth is not expe
movement of the surface of the earth around its cen
ment of the celestial bodies around the earth. From
side the solar system, however, the intersection of t
and it is immediately apparent that the sun is in fact
lines cross (figure 2.10).
 367/1540

Our everyday models of the sun's movement are ex

the Caroline Island navigator's conception of the lo
ence island. The choice of representations limits the
that make sense. Because we Westerners have all b
ideas of Copernicus, we can sit down and convince
we experience is an artifact of our being on the face
et. That is, after all, the "correct" way to think of it,
sarily the most useful way. Modern celestial naviga
pre-Copernican precisely because a geocentric conc
ent movements of bodies on a rigid celestial sphere
inferences about the positions of celestial bodies m
pute than they would be in a heliocentric representa
spective outside the galaxy, of course, the heliocent
is seen to be a fiction which gives an improved acco
movements of bodies within the solar system but w
accounting for the motion of the solar system relati
in the universe. Such a "veridical cosmology" is irre
present-day navigator's concerns.
 368/1540

These observations place strong constraints on cand

how the Caroline Island navigators use the etak sys
must not rely on arbitrary units of distance, nor sho
 Page 83

involve a bird's-eye view of the navigator and

his craft situated in some represented space.

An Alternative Model
 370/1540

What does the Caroline Island navigator gain by

using the conception of the moving reference is-
land? Western navigators find the use of a chart
or some other model indispensable for express-
ing and keeping track of how much of the jour-
ney has been completed and how much remains.
While the Caroline Island navigators are fully
capable of imagining and even drawing charts of
their island group, these conceptions are not
compatible with the movingisland and star-bear-
ing conceptions they use while navigating.
Lewis's diagram was nonsense to Hipour be-
cause Hipour never takes a bird's-eye point of
view when he is thinking about star bearings. In
addition, even though the necessary technology
is available to them, we know that the navigat-
ors carry nothing like a chart with them on their
voyages.
 371/1540

Consider the Caroline navigator's conception in

its context of use. At the outset of any voyage,
the navigator imagines that the reference island
is over the horizon ahead of him and to one side.
It is, for him, under the point on the horizon
marked by the rising or setting of a particular
line of stars. During the course of the voyage,
the reference island will move back along its
track, remaining out of sight of the navigator.
As it does so, it will assume positions under a
succession of star bearings until it lies under the
star bearing that marks the course from the des-
tination to the reference island. If the helmsman
has kept a straight course, then the canoe will be
at the destination when this happens. An import-
ant aspect of this imagined sweep of the refer-
ence island back along its track, out of sight of
the navigator, has been ignored by recent writers
on Caroline navigation but was noticed by Sar-
fert in 1911. Sarfert was struck by the fact that
the navigators conceive of the horizon as a
 372/1540

straight line lying parallel to the course of the

canoe. For a Western navigator, who normally
conceives of the horizon as a circle around him,
this is a puzzling observation. Why should these
navigators make such a counterfactual
assumption?

Sarfert realized the importance of the fact that

the Caroline navigator conceives of the horizon
as a straight line and imagines the apparent
movement of the reference island beyond it.
With this
374/1540
375/1540
 376/1540

Figure 2.11 (a) The standard Western representation of the m

bearings to the etak isla
(b) The Micronesian representation of the same phenomenon
the star bearings.
(c) Illustration that the imagined movement of the etak island
along the course
 377/1540

image the horizon itself becomes a line, parallel to

progress of the reference island from initial bearing
bearings to the final bearing is exactly proportional
the island of departure across the sea to the goal isla
navigator does not think of it from the bird's-eye pe
Rather, the imagined movement of the etak referenc
a complete model of the voyage which is visualizab
ural point of view of the navigator in the canoe (fig
 Figure 2.12 The horizon with star points as see
When the navigator looks at the horizon, he imagines the l
In this diagram, the constellation Orion is shown rising. This
struction
of the entire star compass, including points defined by stars th
shaded
region below the horizon represents the water between
 379/1540

sentation of the spatial extent of the voyage, and of

that does not require either the construction of a ma
point. The straight-line-horizon conception is essen
of angular displacement into linear displacement.
 380/1540

The image of the etak reference island moving alon

can be quite naturally tied to the passage of time. P
navigator has about every voyage is the amount of
to take under various conditions. Suppose that the n
ticular voyage that, under favorable conditions, he w
one day of sailing. If he leaves his island of departu
parture time), he can estimate that he will arrive at h
noon on the following day. In terms of the moveme
this means that the island will move from a position
to a position under the final bearing in one day (figu
normal rate of travel, he can associate other times d
er bearings of the reference island (figure 2.14). In
visual image that represents the extent of the voyag
that represents the voyage and its subparts in time.
are as expected, the task of determining where the r
tioned over the horizon at any point in time is trivia
do is determine
Figure 2.13 The superimposition of starting and ending bearin
the etak
island at the start of the voyage is under the star point define
the star
bearing of the etak island is under the Pleiades. The etak is im
zon
from the star point defined by Antares to the star p
 382/1540

Figure 2.14 Temporal landmarks superimposed on star po

The expected duration of the voyage is mapped uniformly on
ending star bearings of the eta
 Figure 2.15 Just before midnight the navigator
All he needs to do is point to the location of the curr
is superimposed on the spatial landmarks pro
 384/1540

the time of day and refer to the image of the referen

the horizon. By pointing to the position on the horiz
time of day, the navigator has pointed directly at th

The assumption that etak segments are units of dist

apparent inconsistencies: the supposedly confusing
be of different lengths, the conflicting boundaries o
more than one etak island at once, and the conflictin
at the beginning and end of a voyage (caused by us
of sighting in addition to the star-bearingdefined eta
these conceptions ''completely inconsistent with the
(189).

In my model, there is no need to assume that the eta

We dispense with the notion that the numbers of eta
ical computation of the proportion of the voyage co
equality of their lengths is not an awkward concept
that on a typical voyage the navigator will have mo
by star bearings in the middle of the voyage than at
the navigators, we
 Page 88

find that they are not talking about the spatial

duration (length) of the etak segments, but of
their temporal duration. As Gladwin (1970: 187)
notes, "They start out being rather long ('slow')
and then as the canoe passes close by, they be-
come shorter ('fast') as the reference island
swings under one star after another, and then at
the end they are long again, a confusing effect."
The concern of the navigator is not how far he
travels in a particular etak segment, but how
long he will travel before asserting that the ref-
erence island has moved back under the next
star bearing.
 386/1540

When the concept of the etak segment is freed

from the notion of a unit of distance, the appar-
ent problem of using more than one etak island
at once, and the apparent problem of overlap-
ping the star-bearing-determined etak segments
with those determined by the range of birds and
the range of sighting disappear. Using one etak
island to each side of a voyage gives the navig-
ator more conceptual landmarks on his voyage.
There is no reason for it to be a problem to the
navigator. If two reference islands were on the
same side of the voyage, however, the navigator
would have two complete but non-coextensive
sets of time-bearing correspondences superim-
posed on a single horizon, and that probably
would be a source of confusion. But Sarfert
(1911: 134) was quite clear on this issue; he said
that when two etak islands are used, they are
chosen "specifically in such a way that one lies
to the left and the other to the right of the direc-
tion of travel." The confusion that Gladwin
 387/1540

imagined with one reference island to each side

does not arise, since the etak segments are
mapped not onto the course line but onto the im-
agery on the horizon in front of the reference is-
lands (figure 2.16).
 388/1540

The strategy of including the etak of sighting

and the etak of birds is entirely consistent with
the notion of the star-bearingdefined etak divi-
sion as a conceptual landmark. The star-bearing-
defined etak segments are conceptual landmarks
derived in a particular way, and the etak of
sighting is a conceptual landmark determined in
another way. Once established, they function for
the navigator in the same way. They do not
enter into a numerical computation; rather, they
give the navigator a more direct representation
of where he is (or, actually, where land is). In
addition, since the star-bearing etak segments
are slow in passing near the beginning and near
the end of the voyage, it may be helpful to the
navigator to have the other conceptual land-
marks at those points.
 Figure 2.16 The effect of using two etak islands, one on
The star bearings of the two etak islands do not interfere wit
are mapped
onto separate images constructed on the horizons on oppo
 390/1540

What of the phantom etak islands that correspond to

ric features? If the conception of etak presented her
need for there to ever be an island present at the eta
decide, for any particular voyage, that one is going
of the voyage as the movement of an unseen point t
star bearing ahead of and to the side of the course a
bearing behind and to the same side of the course. S
struct does all the conceptual work required of the e
marked that the error of assuming that etak islands
to which one sails in case of danger is "an overly co
the navigators abstract idea" (Neisser 1976, cited in

This conception and this technique make computing

trivial when conditions are favorable. Suppose, how
must be made under conditions which differ from th
outset of the voyage. How could the navigator upda
movement of the reference island to reflect
Page 90
 392/1540

what is happening to his rate of travel? The key

to this problem lies in the judgement of speed
and in the way that this judgement is expressed.
Any experienced Western yachtsman can make
fairly accurate judgements of his boat's speed
through the water without the aid of instru-
ments. By attending to the feel of the boat as it
moves through the water, the accelerations de-
veloped as it moves over waves, the feel of the
apparent wind, the appearance and sound of the
wake (it sizzles at speeds in excess of about 5
knots), the response of the helm, and many other
sensations, the smallboat sailor can make judge-
ments that he normally expresses as a number of
unitsusually knots. The knot is a good choice for
the yachtsman; as one nautical mile per hour, it
is convenient for subsequent numerical calcula-
tions. One might have expressed the speed as
furlongs per fortnight, or on a scale of how
thrilling it is, but neither of these fits especially
well with useful subsequent calculations. The
 393/1540

same must be true for the Caroline Island navig-

ators. There is no doubt that they can make ac-
curate judgements of speed; however, express-
ing those judgements in terms of knots would
not be advantageous at all for them, because that
unit is not compatible with any interesting com-
putations on a visual image of the moving refer-
ence island.
 394/1540

Clearly what is wanted is an expression of speed

that bears a compatible relationship to the im-
agery. Consider the following hypothetical
scheme. At some point in the voyage (and it
could be any point, including the very begin-
ning) the speed of the canoe changes. The nav-
igator reconstructs his image of the movement
of the reference island with the time landmarks
placed in accordance with the previous speed. If
the change occurs at the very beginning of the
voyage, the usual or default speed will be taken
as the previous speed. Let the segment of the
horizon from the present position of the refer-
ence island to any convenient future time land-
mark represent the previous speed (see the seg-
ment labeled "old rate" in figure 2.17). This rep-
resents the expected movement of the reference
island at the previous speed during the period
between the present time and the temporal land-
mark chosen. The problem is to determine the
movement of the reference island during the
 395/1540

same time period at the new speed. If the new

speed is greater than the old speed, then the ref-
erence island will move further along the hori-
zon in the same period; if the speed is less, the
movement will be less. Using the old rate as a
scale, imagine another segment
Figure 2.17 Reconstructing the etak imagery to reflec
 397/1540

("new rate" in figure 2.17), starting at the present po

island and extending in the direction of the apparen
erence island. This segment represents a judgement
the new speed relative to the old speed. Now simply
mark from the end of the old-rate segment to the en
ment. The new-rate segment now defines the new t
speed. The other time landmarks for subsequent po
can be moved accordingly, as in the figure, and a co
pectations for the times at which the etak reference
ture positions is achieved. This procedure can, of co
time there is a noticeable change in the rate of trave
the water. Thus the navigator can always keep an u
bearing correspondences for the etak reference islan
gauge how much of his voyage has been completed
remains.

The notion of the changing bearing of the reference

modated by our usual way of thinking, in which the
while the islands remain fixed. Why, then, would
 Page 92

Micronesian navigators insist on what they

know to be a fiction and imagine that the canoe
is stationary, with the islands in motion about it?
 399/1540

All navigation computations make use of frames

of reference. The most prominent aspect of the
Micronesian conception is the apparent motion
of the etak island against the fixed backdrop of
the star points defined by the sidereal compass.
Here there are three elements to be related to
one another: the vessel, the islands, and the dir-
ectional frame. In order to preserve the observed
relationships of motion parallax, one can have
the vessel and the direction frame move while
the islands stay stationary (the Western solution)
or one can have the vessel and the directional
frame stationary while the islands move (the Mi-
cronesian solution). In the Western case, the dir-
ectional frame is a compass, or a gyrocompass,
and it is carried with the ship. In the Microne-
sian case, the directional frame is defined by the
star points of the sidereal compass, and the star
points are fixed. Each of these schemes makes
some things easy to compute and others
difficult.
 400/1540

The islands move for the Micronesian navigator,

because it is computationally less expensive to
update their positions with respect to the frame
defined by the navigator and the star points than
it is to update the positions of both the navigator
and the star points with respect to the positions
of the islands (Hutchins and Hinton 1984).

Summary
 401/1540

The position-displacement constraint is repres-

ented locally in the Micronesian system in every
inter-island course. Sailing a constant heading
from a known location implicitly represents a
line of position. A second line of position is es-
tablished by the imagined bearing to the etak is-
land. The position of the canoe is established as
simultaneously satisfying these two one-dimen-
sional constraints, although the two representa-
tions are not superimposed directly on each oth-
er, as they are on the Western navigator's chart.
The line of position representing the track of the
canoe is implicit in the steered course of the ca-
noe. The concepts of the etak of birds and the
etak of sighting provide a circle of position con-
straint. Depth contours are also used, and the
Micronesian navigators practice a form of guyot
hopping on some voyages by sailing from
 Page 93

seamount to seamount. Even though they do not

encounter land, they are able to determine their
position by the discoloration in the water caused
by the presence of the submerged seamount.
The distance-rate-time constraint is explicitly
represented in the superimposition of temporal
landmarks on the spatial landmarks defined by
the star bearings of the etak island. In this sys-
tem there are no universal units of direction, po-
sition, distance, or rate, no analogto-digital con-
versions, and no digital computations. Instead,
there are many special-purpose units and an el-
egant way of "seeing" the world in which intern-
al structure is superimposed on external struc-
ture to compose a computational image device.
By constructing this image, the Micronesian
navigator performs navigation computations in
his "mind's eye."

Pre-Modern Western Navigation

 403/1540

The practice of modern navigation is of more re-

cent origin than many of us probably imagine.
Before the introduction of the magnetic compass
(around 1100 A.D.), navigation in European wa-
ters looked a good deal like a rather unsophistic-
ated version of Micronesian navigation. We do
not know the extent to which the similarities
between the two systems are due to independent
invention or how much they share from a com-
mon origin. Some scholars have attempted to
find a common Arab origin for some of the fea-
tures (Lewis 1976), but the evidence of such a
connection is scanty at best. Whatever the reas-
ons for their existence, consider the following
parallels.
 404/1540

Before the discovery of the magnetic compass

needle, the sun and the stars were the guides for
Western navigation. In the Odyssey, Homer has
Odysseus come home from the west by keeping
the bear (the Big Dipper) on his left and sailing
toward the rising of the Pleiades and Arcturus.
The Pleiades and Arcturus have similar declen-
sions (they rise out of the same point in the east-
ern horizon) and are 11 hours different in right
ascension (they are on opposite sides of the
night sky), so one or the other would be in the
sky on any night regardless of season (Taylor
1971). This is clearly a linear constellation con-
struct, although having only two stars in the
constellation is of limited utility (since the nav-
igator will not always have one of the stars near
enough the horizon to be useful for course
setting).
 Page 94

In ancient Greece, very short distances were

given in stadia (a stade is about a tenth of a
mile), but longer distances in early voyages
were given in terms of a day's sail. This was the
distance a "normal ship would accomplish dur-
ing a twenty-four-hour run with a fresh follow-
ing wind" (Taylor 1971: 51). The units in which
the distances between islands are given in the
Micronesian system are based on exactly the
same concept, the only difference being that Mi-
cronesians are interested in a day's sail of a ca-
noe (Riesenberg 1972). This still requires the
navigator to recognize the conditions under
which a "day's sail" will be accomplished in a
day. Making this judgement is probably the sort
of skill that no practitioner can describe in
detail''But ever since sailing began, masters and
pilots have always prided themselves on know-
ing the 'feel' of their ship and how much way
she was making" (Taylor 1971: 52).
 406/1540

The kenning, "a unit of distance used by early

mariners, equivalent to the distance at which the
shore could first be seen from the offing when
making landfall" (Cotter 1983b: 260), appears to
be a European version of the etak of sightingal-
though, since the decks of European ships are
generally higher than the decks of Micronesian
canoes, it is a greater distance. This is a salient
concept for mariners of all kinds. In the Western
system it became the basis of a unit of distance.
Once determined, it was used as a unit of dis-
tance in sailing directions that give "the ken-
nings between headlands and ports" (Cotter
1983b: 255).

The sighting of birds has been important in the

Western tradition since biblical days. Fuson
(1987) reports the following entries in the log of
Christopher Columbus on his first voyage to the
New World:
 407/1540

Later in the day I saw another tern that came

from the WNW and flew to the SE. This is a sure
sign that land lies to the WNW because these
birds sleep ashore and go to sea in the morning
in search of food, and they do not fly sixty miles.
(65)

I know that most of the islands discovered by the

Portuguese have been found because of birds.
(71)

The first quote shows that Columbus was not

only using the behavior of birds to find land, he
was also making the same sort of inferences as
are made by the Micronesian navigators. The
second quotation gives an indication of
Columbus's estimation of the importance of this
technique. Since in the century before his
voyage
 Page 95

no European nation had discovered more islands

than Portugal, this is a strong endorsement of
the technique.
 409/1540

When Europeans first ventured into the open

ocean, they could roughly determine latitude by
measuring the altitude of the North Star, or of
the sun as it passed the local meridian. Yet they
had no way to determine their longitude with
any accuracy. To find an island known to be at a
particular latitude and longitude, a European
navigator would attempt to arrive at the target
latitude well upwind of the target longitude; he
could then simply sail downwind, maintaining
the specified latitude until the island was
sighted. This technique of "latitude sailing" was
probably practiced by traditional Pacific navig-
ators too, although because of the nature of the
traditional practices the evidence is simply lack-
ing. It is interesting to note, however, that a
young Hawaiian navigator, Nainoa Thompson,
who apprenticed himself to an experienced
Caroline Island navigator, has invented or dis-
covered a technique for determining the latit-
udes of specific islands at sea, and has used this
 410/1540

technique to support the latitude-sailing strategy

in longdistance voyages between Hawaii and
Tahiti without the aid of instruments. The tech-
nique relies on the observation of pairs of stars
rising out of or setting into the horizon. At a par-
ticular latitude, if one can find two stars that rise
out of the eastern horizon at the same instant,
then the more northerly of the two will rise be-
fore the other when the observer is north of that
latitude, and the more southerly of the two will
rise before the other when the observer is south
of that latitude. By identifying a few pairs of
stars for each target island, it is possible to use
the latitude-sailing strategy with great accuracy.

The Divergence of Traditions

 411/1540

The similarities between early European naviga-

tion and Micronesian navigation are based on
regularities in the world that are just too useful
to miss. The differences between the two tradi-
tions are many and appear to have increased in
number over time. The divergence of the tradi-
tions can be traced through three closely related
trends in the development of Western naviga-
tion: the increasing crystallization of knowledge
and practice in the physical structure of artifacts,
in addition to in mental structure; the develop-
ment of measurement as analog-to-digital con-
version, and the concomitant reliance on techno-
logies of arithmetic computation;
 Page 96

and the emergence of the chart as the funda-

mental model of the world and the plotted
course as the principal computational metaphor
for the voyage.

The Crystallization of Knowledge and Practice

in the Physical Structure of Artifacts
 413/1540

The Micronesian navigator holds all the know-

ledge required for the voyage in his head. Dia-
grams are sometimes constructed in the sand for
pedagogical purposes, but these (of course) are
only temporary and are not taken on voyages. In
the Western tradition, physical artifacts became
repositories of knowledge, and they were con-
structed in durable media so that a single artifact
might come to represent more than any indi-
vidual could know. Furthermore, through the
combination and superimposition of task-relev-
ant structure, artifacts came to embody kinds of
knowledge that would be exceedingly difficult
to represent mentally (Latour 1986). Many of
the instruments of Western navigation are based
on the principle of building computational con-
straints of the task into the physical structure of
the artifact. I will illustrate this pervasive
strategy with just a few examples.

THE ASTROLABE
 414/1540

The astrolabe (figure 2.18), a portable mechan-

ical model of the movements of the heavens,
was invented in Greece around 200 B.C. Pre-
served during the Dark Ages by the Byzantines,
it was not much modified by the Arabs, via
whom it returned to the West around 1000 A.D.
 415/1540

An astrolabe is a memory for the structure of the

heavens. As we saw in the discussion of Mi-
cronesian navigation, it is possible for an indi-
vidual navigator to learn an internal image of
the heavens so rich that he can recognize ar-
rangements of stars, and even imagine the loca-
tions of stars that are obscured by cloud or the
horizon. However, it is not possible with such
mental representations to control all those spa-
tial relationships with the sort of precision that
is possible in a durable external representation.
In an external representation, structure can be
built up graduallya distribution of cognitive ef-
fort over timeso that the final product may be
something that no individual could represent all
at once internally. Furthermore, the astrolabe
encodes a kind of knowledge that cannot be rep-
resented internally. In this respect, it is a
physical
 Page 97

Figure 2.18 An astrolabe, which superimposes several

kinds of structure
to create a celestial computer.
 417/1540

residuum of generations of astronomical prac-

tice. It is a sedimentation of representations of
cosmic regularities.

The astrolabe also enables its user to predict the

positions and movements of the sun and the
stars:

Because the astrolabe can be set to show the po-

sitions of these heavenly bodies at different
times of day or night, on different dates and or
different latitudes, the instrument is also a com-
puter, serving to solve problems concerning the
positions of the Sun and stars at any given time.
(National Maritime Museum 1976)

Any map of the heavens can capture the rela-

tionships among the stars. The astrolabe goes
further. The physical structure of the
 Page 98

moving parts of the instrument captures regular-

ities in the movements of the heavens and the
effects of latitude and time on the observations
of the heavens. Thus, the astrolabe is not just a
memory for the structure of the sky; it is also an
analog computer.

The major components of an astrolabe are the

mater, the limb, the plate, and the rete. The ma-
ter is the framework that holds the other pieces
together. The limb is a circular scale around the
perimeter of the mater. The limb is inscribed
with a 360° scale and/or a 24-hour scale. In
either case, the limb is a representation of the
structure of sidereal time. Each astrolabe is
really a kit that can be assembled differently ac-
cording to the circumstances of its use:
 419/1540

As the configuration of the celestial coordinates

changes according to the latitude of the observ-
er, a set of removable platessometimes as many
as six, engraved on both sidesis usually sup-
plied, fitting into the hollow of the mater, so that
the user can select the plate most appropriate to
his own latitude. (National Maritime Museum
1976: 14)

The interchangeable plates capture regularities

in the effects of observer latitude on the rela-
tions of the celestial coordinates to the local ho-
rizon. Of course, it is not possible to provide a
plate for every observer latitude, since latitudes
are infinite in number. The plates provide a
coarse discrete representation of the effects of
latitude. Even with a large number of plates, the
representation of observer latitude will be ap-
proximate most of the time. The rete captures
the locational relationships of the stars to one
another and that of the sun to the stars.
 420/1540

The assembled astrolabe brings these three

kinds of structure (and much more) into co-
ordination just the right way so that the interac-
tions of these variables can be controlled in the
manipulation of the physical parts of the instru-
ment. An astrolabe can be made of durable ma-
terials because the regularities it captures
change only very slowly. The variables that do
change, observer latitude and time of observa-
tion, are represented in the physical structure of
the astrolabe either by changeable parts (plates
for each latitude) or by changeable relations
among parts (the rotation of the rete about the
axis with respect to the plate and the limb). The
constraints of the represented world are thus
built into the physical structure of the device.
The astrolabe is a manipulable model of the
heavensa simulator of the effects of time and
latitude on the relationships of the heavens to
the horizon. The astrolabe is an early
421/1540
 Page 99

example of a general trend toward the represent-

ation and solution of computational problems
via physical manipulations of carefully con-
structed artifacts.

THE COMPASS ROSE AND RECKONING

THE TIDES

Frake (1985) provides an especially interesting

example of the ways in which a variety of kinds
of structure are combined in a single artifact to
create a computational system. Frake is inter-
ested in what Northern European sailors knew
about the tides and how they went about know-
ing what they know. Although he is interested in
the tides, his account begins with the so-called
wind rose:
 423/1540

The schema of directions ... resulted from a suc-

cessive division of the quadrants of a horizon
circle formed by north-south and eastwest lines
into 8, 16, and finally 32 named (not numbered)
points.... Similar schemata for segmenting the
circle of the horizon with invariant directional
axes characterize all known early seafaring tra-
ditions: those of the Pacific, the China Sea, the
Indian Ocean and Europe. In the various tradi-
tions, compass directions could be thought of as,
and named for, star paths (as in the Pacific and
the Indian Ocean) or wind directions (as in is-
land southeast Asia and Europe). In all cases,
the compass rose provided an invariant repres-
entation of directions which were, in fact, de-
termined at sea by a variety of means: the sun,
stars, winds, swells, landmarks, seamarks, sea
life and, in later times, the magnetic needle.
(Frake 1985: 262)
 424/1540

The wind rose is an ancient schema that, for

most of its history and in most places, had noth-
ing in particular to do with representing know-
ledge of the tides. In the Mediterranean, for ex-
ample, the tides vary so little that mariners can
safely disregard them. In Northern Europe, by
contrast, tidal variations are large, and the abil-
ity to predict the tides is of great value to mar-
iners. The use of the medieval compass rose in
the prediction of tides is a fine example of the
empirical construction of an artifact in the ab-
sence of a theory of the phenomenon it permits
navigators to predict.

The compass rose as a schema for the expres-

sion of directions was appropriated as a schema
for the representation of time as well (see figure
2.19):
 425/1540

In whatever manner time was determined at

some moment, it was thought of and expressed
as a compass bearing. The sun bears
Figure 2.19 Computing the tide from the superimposition of
the compass rose.
 427/1540

south at noon. It was therefore thought of as bearin

east at 6 a.m. and west at 6 p.m. Only the first of th
practical daily use in northern Europe for determin
bearings were ways of expressing time. (Frake 198

Here we have the superposition of two kinds of stru

structure of the 24-hour solar day on the 32-point c
yields a set of correspondences between direction a

If the bearing of the sun is an expression of solar tim

the moon can likewise be seen as an expression of l
result from the gravitational pulls of the moon and
of the moon predominate. Although the tide does n
moon in any obvious manner, the phase of the tide
place is always the same when the moon
 Page 101

is at any given bearing. That is, for any particu-

lar location, the high tide always comes at a par-
ticular lunar time. Medieval mariners noticed
this fact:

Medieval sailing directions, and presumably the

memories of sailors before written directions,
specify the tidal regime of a given place by stat-
ing the lunar time, named as a compass bearing,
of a given state of the tide, usually high. (Frake
1985: 265)
 429/1540

With both solar and lunar time superposed on

the compass rose, the relationships between sol-
ar time and lunar time can be expressed as direc-
tional relationships. A sailor who knows the lun-
ar time of high tide for a given location can use
the superposed lunar and solar time representa-
tions to compute the solar time of high tide. For
example, if it is known that at a given location
the high tide will occur when the moon bears
WSW,

the sailor has to determine the solar time cor-

responding to "WSW moon" on a given date
and also calculate the state of the tide at any
other solar time. It is in the solution of this
problem that the compass rose as cognitive
schema shows its merits. (Frake 1985: 265)
 430/1540

The simplest case occurs when the phase of the

moon is new. In that case, solar time and lunar
time are the same, and the time of high tide will
be when the moon (and therefore the sun as
well) bears WSW. That is, high tide will come
at 4:30 p.m. If the phase of the moon is other
than full or new, the sailor will have to first de-
termine the relation of solar time to lunar time in
order to compute the time of the high tide. It just
so happens that dividing the 24hour day into 32
equal intervals yields intervals of 45 minutes
each:

Each day, lunar time, and the tide following it,

lags behind the sun by about 48 minutes. Our
compass points divide time into 45-minute inter-
vals, close enough to 48 for tidal calculation.
(Frake 1985: 265)
 431/1540

Suppose a sailor finds himself approaching this

harbor five days past the full moon. Since the
moon and the tide lag 48 minutes behind the sun
each day, "we can count five points of the com-
pass past WSW to NW by W, a point which
marks the solar time of 8:15" (Frake 1985: 265).
In this way, the sailor can compute the solar
time of the high tide (and therefore the other
tides as well) by knowing
 Page 102

the phase of the moon and the establishment of

the port (figure 2.19).

The tidiness of the compass rose as a representa-

tion of these relationships is an entirely fortuit-
ous property of the mapping of the 24-hour day
onto the 32 points of the compass rose. The seg-
mentation of the compass rose into 32 points
and the segmentation of the day into 24 hours
arose independently. Their relationship just hap-
pens to map approximately onto the 48-minute
daily lag of the moon behind the sun that results
from the relation of the 29.5-day lunar cycle to
the 24-hour day.
 433/1540

The superposition of the scheme for the 24-hour

day on the scheme for the 32-point wind rose
yields a system of temporal and spatial land-
marks on which the correspondences of the
states of the tide and time can be imagined and
represented. This is reminiscent of the superpos-
ition of temporal and directional landmarks that
the Micronesian navigators use to compute their
progress on a voyage. Frake noted the abstract
similarities of the two systems:

It is the relationship between determining direc-

tion and determining time that makes the use of
a single schema, the compass rose, appropriate
for representing both direction and time. But the
compass rose is not a time-finding instrument. It
is a very abstract model, a cognitive schema, of
the relations of direction to time, of solar time to
lunar time, and of time to tide. It is an etak of
medieval navigation. (Frake 1985: 266)
 434/1540

Frake's comparison of the compass rose used to

compute tides to the Micronesian concept of
etak is based in the abstract properties of both as
organizing schemata. I believe that the links are
even stronger in that both systems achieve their
computational power by superimposing several
kinds of representational structure on a single
framework.

Both of these devicesthe astrolabe and the com-

pass rose as tide computerinvolve the creation of
physical artifacts whose structures capture regu-
larities in the world of phenomena in such a way
that computations can be performed by manipu-
lating the physical devices. It should be noted,
however, that the use of the compass rose as a
tide computer is a bit more like the Micronesian
navigation case, in that an important part of the
structure is not explicitly represented in the arti-
fact itself but is instead supplied by the situated
looking of the navigator.
435/1540
 Page 103

Measurement and Technologies of Digital

Computation

A second clear difference between the Microne-

sian and Western navigation traditions is the reli-
ance of the latter on measurement and digital
computation. This difference is apparent in the
history of the chip log.

THE CHIP LOG

 437/1540

The spread of the use of the chip log in about

1600 marks an important turning point in the his-
tory of Western navigation. Before this, European
navigation was based primarily on analog compu-
tations. The log gave rise to a computational pro-
cess that begins with analog-to-digital conversion,
which is followed by digital computation, then
either digital-to-analog conversion for interpreta-
tion or digital-to-analog conversion followed by
analog computation. Western navigators have
been practicing this style of navigation for less
than 400 years.

The chip log is a simple analog-to-digital convert-

er that converts the rate of travel of a ship through
water into a number by making a direct measure-
ment of the distance the ship moves in a given
unit of time. A panel of wood, called the chip, is
tied to a line and thrown over the side of the ship
(figure 2.20). It remains stationary
 438/1540

Figure 2.20 A chip log. (From Maloney 1985.)

 Page 104

in the water while the ship sails away from it.

The line attached to the chip is allowed to pay
out, and the amount of line that pays out in a
given period of time is the distance the ship has
traveled in that same period. Since speed is dis-
tance per unit time, this distance is, by defini-
tion, directly proportional to the speed of the
ship. In the early days of the use of the chip log,
the interval of time was measured by the dura-
tion of a spoken prayer. Later, to increase accur-
acy, a sand glass was used instead. To measure
the distance covered, one could reel the line
back in and then measure the amount that had
paid out during the given interval.
 440/1540

It would be exceedingly difficult for a single

person to perform this procedure with accuracy.
I have witnessed the use of a traditional chip log
aboard the restored late-nineteenth-century
Swedish cargo schooner Westkust. The proced-
ure requires three people working in close co-
ordination. One manages the chip and the line,
throwing the chip overboard, letting the line run
through his fingers, and calling out when the
end of the ''stray line" has passed his hand. A
second person inverts a sand glass when the first
indicates that the measured portion of the line is
now streaming out, and calls out when the sand
has run out. When the time is up, the first person
grips the line and stops paying out the line. This
stopping is assisted by a third person, who has
been, up to this point, holding the spool on
which the line is wound so that it can flow out
smoothly. The line is dressed with tassels
hanging from the knots so that the number of
each knot can be discerned at a glance. The
 441/1540

number of the knot nearest to hand is noted, and

the line is pulled in and wound onto the spool.

Columbus did not mention the use of a chip log,

although his logbooks do contain entries record-
ing speeds. It is assumed that he either estimated
his speed by eye or used a precursor of the chip
log that involved "dropping a piece of wood into
the water and timing the passage from bow to
stern" (Fuson 1987: 44). As with the early chip
logs, the interval of time was measured by the
recitation of a chant or a prayer. The first certain
use of the chip log was on Magellan's voyage in
1521.

The use of the log or any other technique based

on the distance covered by the ship requires
both a consistent unit of distance and a means of
reliably measuring distance in that unit. This
was accomplished by preparing the log line in a
special way:
Page 105
 443/1540

By [1633] it had become the general practice to

mark the log line so as to facilitate the calcula-
tion of speed. This was done in the following
way. If a half minute glass was used then the
length of line necessary to indicate a speed of
one mile (of 5000 feet) per hour was (30 x 5000
ft)/(60 x 60) or 41 2/3> feet. In other words, at
one mile per hour the ship would advance, and
the line would run out 41> 2/3 feet in 30
seconds. The line was then divided as follows:
From 10 to 20 fathoms, depending on the size of
the ship, were allowed as "stray line" next to the
log chip, to ensure it being clear of the effect of
the wake. The end of this stray line was marked
either by a knot or a piece of red or white rag,
and then from there the line was divided into
sections of 41> 2/3 feet or 42 feet, each section
being marked by a knot in the line. Thus came
into being the term known as the measure of a
ship's speed in nautical miles per hour. (Hewson
1983: 160)
 444/1540

Even with these refinements, the chip log was

not a very accurate instrument. Many things
could induce errors in the readings. The friction
of the spool, shrinkage of the rope, the surge of
the ship working in steep seas, the effects of
currents, and the yawing of the ship with a swell
on the quarter were among the many things that
could cause significant errors. For a navigator
relying on a log, there is no choice but to expect
error and attempt to allow for it. Just as a car-
penter would rather err on the long side in cut-
ting a piece of wood (so that any error can be
corrected with minimal waste of material) a nav-
igator prefers to overestimate the distance sailed
in order to avoid an unexpected landfall. If an
error is made, it is better to have overestimated
the distance sailed, so that the problem can be
corrected without losing the ship.
 445/1540

Log lines can shrink with use, so it is important

to check the length of the segments between the
knots. This "was facilitated in most ships by
having permanent marks of nails driven into the
deck" (Hewson 1983: 166). Decks don't stretch
and shrink as ropes do. Putting the calibrating
nails into the deck is a way of creating a
memory for the lengths between knots in the log
line in a medium that has physical properties
that match the computational needs of the task.
In this case, the marks on the deck are a memory
for distance.

In the late eighteenth century many attempts

were made to develop more accurate ways to
measure speed or distance run
 Page 106

through the water. These included the taff-rail

and paddle-wheel logs (Hewson 1983). Al-
though the details of their implementation var-
ied, these were all simple analog-to-digital con-
verters that stood in the same relation computa-
tionally to other navigation tools that the chip
log had.

The importance of the chip log is that it changed

the way navigation was done. Rather than
knowing a journey should take some number of
days and counting days until the required num-
ber had elapsed, a navigator using a chip log
used the concept of distance between points and
the integration of speed over time to determine
the distance covered by the ship. Having created
a digital representation of speed, the chip log
created a need for a method of calculation that
could operate on that representation to tell the
navigator what he needed to know.
 447/1540

The chip log and its descendants are among the

many measuring instruments that entered the
navigation tool kit during the European expan-
sion. Others include a succession of instruments
for measuring the altitudes of stars (astrolabe,
quadrant, cross-staff, sextant), range-measuring
instruments, instruments to measure bearings,
azimuths, and courses, and instruments to meas-
ure depths. All of these are analog-to-digital
converters. All of them create representations
that are subsequently processed using a special
arithmetic technology in order to produce in-
formation that is of use to the navigator.
 448/1540

Consider the enormous importance of common

logarithms. With a table of logarithms, one can
transform multiplication and division into addi-
tion and subtraction. That is, when numerical
values are expressed as logarithms, the complex
typographic operations required for multiplica-
tion and division (the algorithms of placevalue
arithmetic) can be replaced by a simpler set of
typographic operations that implement addition
and subtraction. Speaking of Edmund Gunter
(1581-1626), Cotter (1983a: 242) says:
 449/1540

He introduced the first tables of logarithmic tri-

gonometrical functions, without which a seaman
would find almost insurmountable difficulty in
solving astronomical problems. It was Gunter's
Tables, published in 1620, that paved the way to
the new phase of "arithmetic navigation."
Armed with the new logarithmic canon, a navig-
ator who memorized the necessary rules could
solve nautical astronomical problems with relat-
ive ease.
 Page 107

But the seamen of the time found even the sim-

plified calculations daunting, so Gunter de-
signed a ruler with a number of scales:

Among these are a logarithmic scale of natural

numbers, logarithmic scales of sines, tangents
and versines; ... and a "meridian line" to facilit-
ate the construction of sea charts on Wright's
projection.... With the advent of "arithmetical
navigation," in which Gunter played the domin-
ant role, the common log for measuring a ship's
speed became commonplace. To the careful sea-
man using a Gunter scale the proportional
problem of finding speed was mechanical and,
therefore, trivial. (ibid.)
 451/1540

The predecessor of the slide rule is apparent

here. In fact, it appears that two of Gunter's
scales were sometimes "laid down on rulers to
slide by each other" (Oxford English Dictionary,
1971). Again we have an artifact on which com-
putations are performed by physical manipula-
tion. However, there is an important difference
between the astrolabe and Gunter's scale in this
regard. In both cases the constraints of a repres-
ented world are built into the physical structure
of the device, but in the case of Gunter's scale
the represented world is not literally the world
of experience. Instead it is a symbolic world: the
world of logarithmic representations of num-
bers. The regularities of relations among entities
in this world are built into the structure of the
artifact, but this time the regularities are the syn-
tax of the symbolic world of numbers rather
than the physics of a literal world of earth and
stars. The representation of symbolic worlds in
physical artifacts, and especially the
 452/1540

representation of the syntax of such a world in

the physical constraints of the artifact itself, is
an enormously powerful principle. The chip log
and Gunter's scale are representative elements of
a cognitive ecology based on measurement and
digital computation.

The Chart as a Model of the World

The navigation chartperhaps the best available

example of the crystallization of practice in a
physical artifactis intimately involved in the
prototypical cycle of measurement, computa-
tion, and interpretation that characterizes so
much of Western navigation. These characterist-
ics of the chart will be developed in much more
detail in the coming chapters. At this point it is
useful to examine another contribution of the
chart that marks one of the most im-
 Page 108

portant elements of the Western conception of

navigation. The chart, by virtue of its interpreta-
tion as a model of an expanse of actual space,
encourages a conception of a voyage as se-
quence of locations on the chart.
 454/1540

Descriptive sailing directions were the principal

navigation aids up until the end of the eight-
eenth century. These documents describe to the
sailor how to proceed with the voyage and what
he can expect to see. Then, with the continued
improvement of survey techniques and the in-
creasing range of areas accurately surveyed,
sailing directions were supplanted by the
pictorial chart. This marks an important change
in perspective. Where the sailing directions
presented the world from the perspective of the
deck of the ship, the coastal chart presented the
world from above-from a virtual perspective (a
"bird's-eye view") that navigators would never
actually experience. Modern navigators may
take to the air and adopt something very like a
bird's-eye view, but this is not in fact the per-
spective presented by the navigation chart. The
navigation chart presents the world in a per-
spective that can never be achieved from any ac-
tual viewing point.
 455/1540

A chart must be more than an accumulation of

observations. The structure of the chart is cru-
cial (Cotter 1983b). The importance of the com-
pass in the actual practice of navigation was par-
alleled by its contribution to the quality of chart
production. The compass made it practical to
make accurate charts. It was possible before (by
means of the stars) to get directions for bearings
and courses, but not nearly so conveniently.
Even when a compass was used, serious prob-
lems in chart construction remained. For ex-
ample, early charts of the Mediterranean showed
a pronounced upward tilt in the eastern end.
This tilt was produced by the difference in mag-
netic variation between the western and eastern
reaches of the Mediterranean Sea. If the carto-
grapher uses a magnetic compass to make the
chart, and the navigator uses a magnetic com-
pass to determine courses, and if both com-
passes show the same errors in the same places,
why would anyone care and how could anyone
 456/1540

ever notice that the charts put the land in the

wrong places?

TAKING THE MEASURE OF THE EARTH

The distortions in charts produced by changes in

magnetic variation became an issue when the
chart became a point of articulation between the
measure of the earth and celestial observations.
In or-
Page 109
 458/1540

der for the effects of distance covered on the

face of the globe to be reconciled with the at-
tendant, and also measurable, changes in latit-
ude (a relation to the celestial sphere), the unit
of measure for distance had to be grounded in
the measure of the earth itself. That is, a degree
of arc on the earth's surface is a particular dis-
tance, and navigators wanted to be able to com-
bine and interrelate measurements made in
terms of distance traveled with measurements of
latitude. For example, if I am currently 2° south
of my home port, sailing north, how far, in units
of distance on the surface of the ocean, must I
sail to arrive at the latitude of my home? The
question is: How much linear distance on the
surface of the earth corresponds to a degree of
arc on the same surface? As we saw in the dis-
cussion of the historical changes in the length of
the nautical mile, establishing a standard that
permitted the chart to be a point of articulation
 459/1540

between the measure of the earth and the meas-

ure of the heavens was no simple task:

The north-up convention is clearly related to the

concept of defining position in terms of latitude
and longitude. For coastal navigation this
concept is of no consequence: a coastal navigat-
or is interested in defining his ship's position not
in terms of these spherical coordinates, but in
terms of bearings and distances from prominent
landmarks of hazards such as rocks and shoals.
Early coastal charts, therefore, (and with good
reason) were orientated relative to the run of
the coast rather than to the compass. (Cotter
1983b: 256)
 460/1540

The modern chart incorporates the global con-

vention of north-up depiction of a plane surface
having a discrete address in terms of latitude
and longitude for every location. This global
framework permits the combination of any num-
ber of observations from any number of loca-
tions. With this scheme it is possible to compute
the relationship between any two locations on
earth even though that relationship has never
been measured.

The virtual perspective created in the chart does

not privilege any actual perspective. A naviga-
tion chart is a representation that is equally use-
ful (or not useful) from any actual perspective. It
attempts neutrality with respect to the perspect-
ives from which the world will be seen by nav-
igators. Since in the Western tradition nearly all
navigable space is represented from this virtual
perspective, it is from this virtual perspective
that voyages come to
461/1540
 Page 110

be conceived. We imagine the voyage as the

movement of our ship over a stretch of water.
There is the ship, and there we are, like tiny
imagined specks on the tiny imagined ship that
is moving in our mind's eye across the expanse
of paper that represents the water between origin
and destination.
 463/1540

Yet there are moments in which this perspective

does not serve the needs of the navigator, as
when one attempts to determine what the land
depicted should look like from the perspectives
that are actually achieved in ships. Here coastal
profiles may be included. There is a problem at
the moment in which one moves conceptually
from being "on the chart" to being "in the
world." The coastal profile is a concession to
this problem. The first coastal profiles appeared
in 1541, in Pierre Garcie's book Le grand routi-
er. Coastal profiles are representations that priv-
ilege particular perspectives that the chart
makers anticipate will be encountered often by
users of the chart.
 464/1540

The common framework of locations also per-

mits the superposition of a wide variety of struc-
tures. In addition to the obvious boundaries of
bodies of water and land, the locations of cultur-
al features and of geographical features (both
above and below water) are depicted. This su-
perposition of these structures, which underlies
much of the computational power of the chart, is
so obvious as to go unnoticed by virtually all the
users of the chart. Soundings were first shown
reduced to a standard half-tide datum in 1584, in
Janzsoon Waghenaer's Speighel der Zeevaert.

SOCIAL PROBLEMS OF CHART

CONSTRUCTION
 465/1540

The birth of astronomical navigation was much

less a scientific problem than a question of or-
ganization. Jean II of Portugal had the great
merit to have knownbefore any other head of
stateto organize the technical exploitation of the
theoretical knowledge of his epoch.Beaujouan,
Science Livresque et Art Nautique au XVe
Siecle; cited in Waters 1976: 28 (translation by
E.H.)

There is a great deal of knowledge embodied in

any navigation chart. To add a new feature to a
chart, one must determine its relationship to at
least two other features. Since a chart implicitly
represents a spatial relationship between the
members of every pair of features depicted, any
the new feature acquires relationships to all the
other features on the chartnot just the ones that
were
 Page 111

used to establish its location. If the number of

relationships depicted is a measure of the know-
ledge in a chart, there may be more knowledge
in a chart than was put into the chart. In fact,
most of the relationships depicted on any chart
have never actually been measured. Even so, a
great many observations are required in order to
construct a useable navigation chart. A naviga-
tion chart represents the accumulation of more
observations than any one person could make in
a lifetime. It is an artifact that embodies genera-
tions of experience and measurement. No navig-
ator has ever had, nor will one ever have, all the
knowledge that is in the chart. The really diffi-
cult technical problem in the production of
charts is the collection of reliable information.
(See Latour 1987 on centers of calculation.)
 467/1540

Compare the problem faced by the Portuguese

during their expansion with that faced by the
Micronesians. Every Micronesian navigator
knows the courses and distances between all the
islands in his sailing rangeincluding, as we have
seen, courses between islands that have not been
visited for many generations. How could a Mi-
cronesian navigator come to have this know-
ledge? Clearly, it is acquired over generations,
and what any navigator knows is much more
than could be learned by direct observation. The
knowledge is a compilation of the experiences
of many navigatorssome of whom, one must as-
sume, set out on voyages of discovery, knowing
which way they were sailing, and how to get
home, but not what they would find. Over the
years the knowledge accumulated, expressed in
the framework of star courses and etak images.
Today the knowledge of a Micronesian navigat-
or exceeds what could be acquired by direct
 468/1540

observation, but it does not exceed what could

be remembered by one individual.

The world of the Portuguese fleet in the early

fifteenth century was much larger than one
group of islands. The total knowledge of the
world not only exceeded what could be ob-
served by any individual, it exceeded what
could be known by any individual. Like the Mi-
cronesians, the Portuguese needed a consistent
set of techniques for making observations and a
representational framework in which all obser-
vations could be expressed. They also needed to
train a large number of observers in these tech-
niques so that the experiences of all of them
could accumulate in a common store. This was
the creation of an enormous system for gather-
ing and processing informationa cognitive sys-
tem of many
 Page 112

parts that operated over many years to create a

collection of representations of the spatial or-
ganization of the surface of our planet (Law
1987).

The Computational Ecology of Navigation

Tools

The mutual dependencies among the various in-

struments and techniques is clearly visible in the
history of navigation. Even though the chip log
was available for use in the sixteenth century,
for example, it was not generally adopted until
the middle of the seventeenth. Why weren't sail-
ors using the log more widely? Because they
had no convenient way to carry out the compu-
tations required to turn the readings gained from
the log into useful information about the ship's
position.
 470/1540

Why was there, before 1767, no nautical alman-

ac giving the positions of the stellar sphere, the
sun, the moon and the planets? Astronomy was
certainly advanced enough to provide these data.
The answer is that these data are useless for
marine navigation in the absence of an accurate
way to determine time at sea. The need was well
known, and in 1714 the English Parliament
passed an act "providing a Publick Reward for
such person or persons as shall discover the
Longitude at Sea." The reward went unclaimed
until 1762, when Harrison constructed a chrono-
meter that would work reliably at sea (Taylor
1971: 261). The nautical almanac soon
followed.
 471/1540

Before seagoing chronometers were perfected,

there was little incentive to develop better sex-
tants. At the equator, an error of one minute in
time produces an error of 15 miles in east-west
position. Since the earth turns on its axis one de-
gree of arc in 4 minutes of time, there is no util-
ity in having an instrument that can measure ce-
lestial angles even to the nearest degree unless it
is coupled with a chronometer that is accurate to
within 2 minutes. Thus, both the development of
the sextant and the development of accurate
navigation tables were arrested by the lack of
the chronometer. Both were technological pos-
sibilities before the development of the chrono-
meter, but there was no use for them until time
could be reckoned accurately.
 472/1540

Similar dependencies can also be seen in the

history of the chart and the plotting tools. Charts
were in wide use by the thirteenth century, but
the most basic of plotting toolsthe parallel ruleas
not invented until the late sixteenth century
(Waters 1976).
 Page 113

Why? Because a straight line has no special

meaning on an early chart. Not until the Mercat-
or projection did a straight line have a computa-
tionally useful meaning. But the earliest Mercat-
or chart came with no explanation. It is unlikely
to have been used at sea (Waters 1976). Navig-
ators needed instruction in the use of exotic
technologies. (Note: For some time, the Mercat-
or projection was known to the English-speak-
ing world by the name of the man who pub-
lished an English version of it in 1599: Edward
Wright.)

The early astrolabes and quadrants were uni-

versity equipment. Ordinary seamen couldn't
use them. The tools had to be simplified, and
there had to be instructions in their use:
 474/1540

By themselves these instruments (quadrant and

astrolabe) were, of course, powerless. The mere
fact of sighting a heavenly body through pin-
holes of an alidade had nothing per se to do
with navigation. That sighting, or the reading
that corresponded to it, had to undergo a num-
ber of complex transformations before it could
be converted into a latitude. The construction of
a network of artifacts and skills for converting
the stars from irrelevant points of light in the
night sky into formidable allies in the struggle to
master the Atlantic is a good example of hetero-
geneous engineering. (Law 1987: 124)
 475/1540

Sometimes, as the nature of the practice has

changed, the role of particular instruments has
changed. For example, the astrolabe was origin-
ally used both to measure the altitudes of celes-
tial bodies and to predict the altitude and azi-
muth of a star. The observationmaking duties
were subsequently taken over by the quadrant,
then by the cross staff, and finally by the sex-
tant. The function of computing the expected
altitudes and azimuths of stars was taken over
by a complex set of tables. Even though the
quadrant and the cross staff were eventually re-
placed by the sextant (which is much easier to
use), their ancestor, the astrolabe, survives as
the modern star finder. It is now usually made
of plastic instead of brass, but it is easily recog-
nizable. A star finder is not considered accurate
enough for the purposes of computing expected
altitudes, but it is used to set the sextant before
making the observation. It is used to get the set-
ting of the precision instrument into the right
 476/1540

neighborhood. It has been moved to a new job

in the navigation process.

In attempting to understand the history of navig-

ation from a cognitive perspective, it is import-
ant to consider the whole suite of instruments
that are used together in doing the task. The
tools of
 Page 114

navigation share with one another a rich net-

work of mutual computational and representa-
tional dependencies. Each plays a role in the
computational environments of the others,
providing the raw materials of computation or
consuming the products of it. In the ecology of
tools, based on the flow of computational
products, each tool creates the environment for
others. This is easy to see in the history of the
physical tools, but the same is certainly true of
the mental tools that navigators bring to their
tasks. Frake's compass rose is there for all to
see, but it becomes a tide computer only in in-
teraction with the establishment of the port and
with a particular way of seeing the circle of dir-
ections as a representation of the temporal rela-
tionships of the periodic cycles of the sun and
the moon.
 478/1540

Every argument showing why a particular tool is

easy to use is also an argument showing why
both internal and external tools are part of the
very same cognitive ecology. It is a truism that
we cannot know what the task is until we know
what the tools are. Not only is this true of both
internal and external tools, it is also true of the
relationships among them.

The Transparency of Cultural Representations

How We Fail to See Culture (Ours and Theirs)

I have presented this comparative and historical

treatment to remind us all that the ways we have
of doing things, the ways that seem to us to be
natural and inevitable or simply the con-
sequences of the interaction of human nature
with the demands of a given task, are in fact his-
torically contingent. As Benedict (1946: 14)
notes,
 479/1540

The lenses through which any nation looks at

life are not the ones another nation uses. It is
hard to be conscious of the eyes through which
one looks. Any country takes them for granted,
and the tricks of focussing and of perspective
which give to any people its national view of life
seem to that people the god-given arrangement
of the landscape. In any matter of spectacles, we
do not expect the man who wears them to know
the formula for the lenses, and neither can be
expect nations to analyze their own outlook
upon the world.
Page 115
 481/1540

Of all the many possible ways of representing

position and implementing navigation computa-
tions in the Western tradition, the chart is the
one in which the meaning of the expression of
position and the meaning of the operations that
produce that expression are most easily under-
stood. As was noted above, lines of position
could be represented as linear equations, and the
algorithm applied to find their intersection could
be that of simultaneous linear equations. As a
physical analog of space, the chart provides an
interface to a computational system in which the
user's understanding of the form of the symbolic
expressions (lines of position) is structurally
similar to the user's understanding of the mean-
ings of the expressions (relations among loca-
tions in the world) (Hutchins, Hollan, and Nor-
man 1986). In fact, the similarity is so close that
many users find the form and the meaning indis-
tinguishable. Navigators not only think they are
doing the computations, they also invest the
 482/1540

interpretations of events in the domain of the

representations with a reality that sometimes
seems to eclipse the reality outside the skin of
the ship. One navigator jokingly described his
faith in the charted position by creating the fol-
lowing mock conversation over the chart: ''This
little dot right here where these lines cross is
where we are! I don't care if the bosun says we
just went aground, we are here and there is
plenty of water under the ship here." For the
navigator, the ship is where the lines of position
intersect.
 483/1540

It is really astonishing how much is taken for

granted in our current practice. The difficulties
that were overcome in the creation of all these
techniques, and the power they provide relative
to their predecessors, are not at all apparent to
the modern practitioner. Only when we look at
the history can we see just how many problems
had to be solved and how many could have been
solved differently in the course of the develop-
ment of the modern practices. A way of thinking
comes with these techniques and tools. The ad-
vances that were made in navigation were al-
ways parts of a surrounding culture. They ap-
peared in other fields as well, so they came to
permeate our culture. This is what makes it so
difficult to see the nature of our way of doing
things and to see how it is that others do what
they do. We see in the divergence of these tradi-
tions not just the development of the tools of
measurement, but a passion for measuring and a
 484/1540

penchant for taking the representation more seri-

ously than the thing represented.
 Page 116

While all navigation computations seem to be

describable by a small number of abstract prin-
ciples, there is great variation in the representa-
tional systems and concomitant algorithmic pro-
cedures that may be employed to organize the
computations. The actual devices and processes
in which these representations and algorithms
are implemented have a complex evolutionary
history. In the next chapter we will consider in
much greater detail the implementation of the
computations of Western navigation.
 Page 117

3 The Implementaion of Contemporary

Pilotage

A Broader Sense of Computation

Having considered the computational nature of

the problems of navigation and the representa-
tional assumptions on which Western navigation
is based, let us now take up the question of how
the basic computations of navigation are actu-
ally implemented.
 487/1540

In his seminal book The Sciences of the Artifi-

cial Herbert Simon (1981: 153) said that "solv-
ing a problem simply means representing it so
as to make the solution transparent." Of course,
the meaning of 'transparent' depends on the
properties of the processor that must interpret
the representation. Simon had theorem proving
in mind when he made this point, but it applies
very nicely to navigation. The basic procedures
of navigation are accomplished by a cycle of
activity, called the fix cycle, in which represent-
ations of the spatial relationship of the ship to
known landmarks are created, transformed, and
combined in such a way that the solution to the
problem of position fixing is transparent.
 488/1540

The fix cycle implements a computation. Since

some of the structure involved in this computa-
tion is internal to the individuals and some ex-
ternal, it is useful to adopt a concept of compu-
tation that does not require a change of theory to
cross the skin. The fix cycle is accomplished by
the propagation of representational state across
a series of representational media. The repres-
entations of the position of the ship take differ-
ent forms in the different media as they make
their way from the sighting telescopes of the al-
idades to the chart. I will refer to a configuration
of the elements of a medium that can be inter-
preted as a representation of something as a rep-
resentational state of the medium. Representa-
tional states are propagated from one medium to
another by bringing the states of the media into
coordination with one another.
 489/1540

Simon's prototypical case of problem solving by

re-representation was theorem proving in which
the computational system is a set of axiomatic
propositions and a set of rules for operating on
the propositions. The rules describe operations
that preserve the truth of the axioms. The system
contains many potential conclusions. A
 Page 118

"problem" in this world is defined by a proposi-

tion about a relationship between terms. The
solution to the problem consists of a sequence of
rule applications that demonstrate that the target
relationship was true in the axioms. The most
straightforward way to prove a theorem is to
make a sequence of rule applications that derive
the target proposition itself from the axiomsthat
is, to rerepresent what the axioms say in such a
way that they become the target proposition. In
this way, the problem is represented in a way
that makes the solution transparent. David Kirsh
(1990) speaks of this process as one of making
explicit that which is only implicit in the starting
state.
 491/1540

Sequential symbol processing of the sort exem-

plified by theorem proving can certainly be de-
scribed in terms of coordination of structure.
Rule application is the means of coordination
between the rule and the state to which it is ap-
plied. The consequence of rule application is a
transformed system state. Still, there are many
interesting types of coordination of information-
bearing structure that are poorly described as ex-
plicit symbol processing. This is not to say that
they could not be implemented as symbol-pro-
cessing routines; it is simply to say that, when
they are implemented that way, their essential
character is lost in the details of the implementa-
tion. I propose a broader notion of cognition be-
cause I want to preserve a concept of cognition
as computation, and I want the sort of computa-
tion that cognition is to be as applicable to
events that involve the interaction of humans
with artifacts and with other humans as it is to
events that are entirely internal to individual
 492/1540

persons. As we shall soon see, the actual imple-

mentation of many interesting computations is
achieved by other than symbolic means. For our
purposes, 'computation' will be taken, in a broad
sense, to refer to the propagation of representa-
tional state across representational media. This
definition encompasses what we think of as pro-
totypical computations (such as arithmetic oper-
ations), as well as a range of other phenomena
which I contend are fundamentally computation-
al but which are not covered by a narrow view
of computation.

The Fix Cycle as an Implementation of the

Computation

The navigation system captures several one-di-

mensional constraints in the world, then repres-
ents them and re-represents them
 Page 119

until they arrive at the chart. The chart is a me-

dium in which the multiple simultaneous one-di-
mensional constraints can be combined to form
the solution. At the computational level, we say
that the inputs to the system are two or more
lines of position and the output is a position fix.
The representation utilized is a twodimensional
model of space, and the algorithm defined in
that representation for combining lines of posi-
tion is to find the intersection of the graphical
depictions of the lines in two-dimensional space.
As we have seen, there are many ways to imple-
ment this algorithm on this representation. The
algorithm is implemented by manipulating the
devices described in chapter 1 in such a way that
particular physical states that can be taken to
represent the spatial relationship of the ship to
its surroundings are propagated from one device
or medium to another until they arrive at the
chart.
 494/1540

Mappings across Representations

What lies between the problem and its solution?

Between the relation of the ship to its environ-
ment and the position plotted on the chart lie a
number of representational media across which
the representations of the spatial relationship of
the ship to the world are propagated. Some of
the representations through which the informa-
tion about the ship's relation to the world passes
are easily observable. We begin our discussion
of the nature of the navigation computation with
a consideration of the propagation of representa-
tional state across these easily observable media.

THE WORLD

Imagine a ship in a harbor. The ship has a spa-

tial relationship to every object in the surround-
ing world. Each of these relationships is spe-
cifiable as a direction and a distance.
 495/1540

THE ALIDADE

The navigation process begins when a spatial re-

lationship between the ship and a fixed land-
mark is transformed into a state of the alidade-
gyrocompass system. (See figure 3.1 for this and
the points to follow.) This is accomplished by
aiming the alidade at the landmark so that the
hairline in the alidade sight is superimposed on
the target. The alidade now has a particular rota-
tional orientation on its base. To be useful, this
rotational orientation must be
Page
497/1540
498/1540
 499/1540

Figure 3.1 The cascade of representations in piloting. A repre

ation of the ship's relationship to the
landmark is propagated from the alidade (a) to the bearing log
then to the hoey (c), and finally from the hoey to the chart (
 500/1540

expressed as an angle with respect to something. Th

hairline in the alidade sight also falls over the scale
the gyrocompass card (figure 3.1a). If the gyrocom
repeater is working, the superimposition of the hair
on the gyrocompass card is an explicit representatio
of the angle of the line of sight between the ship an
the landmark with respect to true north. One end of
hairline provides the coordination of an object in th
world with the sighting device; the other end provid
the coordination of the sighting device with the true
north reference. The whole system hangs by the thr
of simultaneous coordination provided by this hairl
Page 1
502/1540
 503/1540

There are three spaces to consider here (see figure

3.2). First, the space in which the ship and the land-
mark lie is a macrospace. We would like to measur
the directional relationship of the ship to the land-
mark in this space. To do that, we must reproduce
that directional relationship in a second space: the
microspace of the alidade. When the alidade is
aimed and the hairline falls on the image of the land
mark in the sight, the directional relationship of the
ship to the landmark is reproduced in the directiona
relationship of the alidade eyepiece to the hairline.
The physical structure of the alidade guarantees tha
the directional relationship of the eyepiece to the
hairline will, in turn, be reproduced in the direction
relationship of the center of the gyrocompass card t
the point on the edge of the gyrocompass card over
which the hairline falls. Thus, the directional rela-
tionship of the ship to the landmark is reproduced i
a third space: the microspace of the gyrocompass
card.
504/1540
506/1540
507/1540
 508/1540

Figure 3.2 The three spaces of the bearing sighting. The spati
landmark
(a) is reproduced in the physical orientation of the alidade (b
pass card (c).
 Page 123

This last space is a thoroughly domesticated

space (Goody 1977). It is culturally constructed,
measured, and labeled. The locations on the
perimeter of the compass card bear labels that
are names of directions. When the directional
relationship of the ship to the landmark is repro-
duced in this space, the relationship can be giv-
en the name of the point on the perimeter of the
compass card that bears the same relationship to
the center of the card as the landmark bears to
the ship in the world.
 510/1540

Taking the navigation system as our cognitive

unit of analysis, we can see the operation of the
alidade as an instance of situated seeing imple-
mented in hardware. It is a part of the cognitive
system that projects internal structure (the com-
pass rose) and external structure (the landmark)
onto a common image space and, in so doing,
gives meaning to the thing seen that goes bey-
ond the features of the thing itself. The prism in
the alidade that superimposes the image of the
gyrocompass card on the image of the landmark
is a simple technological device that produces
the superposition of internal and external
structure.
 511/1540

The printed scale on the gyrocompass card per-

mits the analog angular state of the alidade to be
converted to a digital representation. This digital
representation may have intermediate external
representations, depending on how the task is
being done. During Sea and Anchor Detail, for
example, the digital representation is spoken
over the phone circuit. During Standard Steam-
ing Watch, the single watchstander on duty may
mentally rehearse the bearing, or may jot it
down on a sheet of paper or even on the palm of
his hand while taking other bearings. In any
case, this digital representation of the state of
the alidade subsequently appears without fail as
a written number in the column labeled with the
name of the landmark in the bearing record log
(figure 3.1b).
 512/1540

The hairline and the telescopic sight add accur-

acy to the alidade, but they are not essential.
Many hand-bearing compasses use a prism or a
half-silvered mirror to produce a similar super-
position of internal and external structure. The
mirror or prism is a simple way to implement
the superposition of images that brings the vari-
ous structural elements into coordination with
one another.

THE PHONE CIRCUIT AND BEARING

RECORD LOG

The azimuthal orientation of the alidade is an

analog representation of the directional relation-
ship between the ship and the
Page 124
 514/1540

landmark. This analog representation is conver-

ted to a digital representation in the act of read-
ing the bearing from the scale of the gyrocom-
pass card. In the previous chapter, it was argued
that the analog-to-digital conversions served the
application of arithmetic operations. The digital
representations produced by reading the gyro-
compass scale are not subjected to arithmetic
operations before being transformed back into
analog form for interpretation. Instead, the
analog-to-digital conversion serves another pur-
pose. It provides a representation that is portable
and transmissible over a restrictedbandwidth
channel. It is easy to imagine a functionally
equivalent system in which the result of the
sighting with the alidade would be an analog
signal. For example, if a two-arm protractor was
placed in the alidade and recorded the angle of
the observation, this could be used directly to
plot the line of position. In that case, though, the
protractor would have to be physically
 515/1540

transported from the alidade to the chart. This

scheme would also require some other technique
to support the recording function now served by
writing the bearings in the bearing record log.
One can imagine the difficulty of storing the
history of observations if each one was ex-
pressed in an angle on a physical device such as
a two-arm protractor. The analog-to-digital con-
version employed here creates a representation
of the analog angle that is transportable. It can
be spoken, and thus the information can move
without moving the physical medium in which it
is represented. It is also easily recorded, the
storage requirements of the digital representa-
tion being much less than those required by the
analog representation.
 516/1540

The bearing record log is a representational

format that is at least 4500 years old in the
Western cultural tradition. Sumerian account-
ants developed similar layouts for recording ag-
ricultural transactions as early as 2650 B.C.
(Ifrah 1987). The column-androw format is one
of the earliest known devices for superposing
representations. Representational structure em-
bodied in the organization of the rows is super-
posed on representational structure embodied in
the organization of the columns to produce a
system of coordination between the two
structures.

THE HOEY

The digitally represented bearing is sub-

sequently represented as an angle on a one-arm
protractor called the hoey (figure 3.1c). Here the
digital representation of the angular relationship
of the ship to the
517/1540
 Page 125

landmark is converted back into an analog form.

The angular relationship of the ship to the land-
mark with respect to true north is now repro-
duced (with some error) in the angular relation-
ship between the center of the hoey and the
point on the hoey scale that bears the name of
the bearing. The hoey is another culturally con-
structed, domesticated space. Again we find la-
bels for directions, the very same labels that are
on the gyrocompass card. The direction on the
hoey is with respect to some directional referent,
which is at this point still unspecified. Ul-
timately, it must be the same directional referent
with respect to which the original measurements
were made.
 519/1540

The angle represented in the space of the hoey

scale is now reproduced in the space of the hoey
arm. The arm is rotated so that its edge (or the
hairline index aligned with the edge) is over the
point on the hoey scale that represents the bear-
ing of the landmark. This establishes the direc-
tion as a physical state of the hoey. This state is
protected from inadvertent upset by tightening a
friction lock at the center of the hoey.

THE CHART
 520/1540

The hoey, thus configured, is then brought into

coordination with the chart (figure 3.1d). In this
step, the angle that was measured between the
ship and the landmark in the world will at last
be reproduced in the space of the chart. There
are two essential aspects to this coordination.
First, the edge of the arm of the hoey must pass
through the symbol on the chart that represents
the observed landmark. Whenever this con-
straint is satisfied, the edge of the hoey arm de-
scribes a line of position with respect to the
landmark. Second, the base of the hoey must be
aligned with the directional frame of the chart.
Whenever this constraint is satisfied, the hoey
scale represents directions with respect to true
north in the space of the chart. When these two
constraints are simultaneously satisfied, the
edge of the hoey arm describes a line on the
chart that is a line of position relative to the
landmark with the same angular relationship to
 521/1540

true north as the line of sight between the ship

and the real landmark in the world.

The latitude-and-longitude grid on the chart

plays an essential role in the computation of the
fix. The Mercator projection is a computational
artifice in which straight lines have special
meaning: they are lines of constant direction.
The role of the grid in the
 Page 126

mechanics of the construction of a line of posi-

tion (LOP) highlights another important func-
tion. It provides a frame of reference that serves
as a common anchor for both the locations of
the features that are depicted on the chart and
the relationship of the ship's position to those
depicted features. When the chart was construc-
ted, the locations of the symbols were fixed with
respect to the grid. In the course of plotting the
LOP, one aligns the base of the plotting device
with a line of latitude on the chart. This ties the
observation to the reference grid of the chart.
Once established by the edge of the arm of the
hoey, the line of position can be ''saved" by
drawing a line on the chart.
 523/1540

Thus one creates the line of position by

propagating representational state across a set of
structured representational media until it arrives
at the chart. The directional relationships of the
ship to landmarks in the world are reproduced in
a set of spaces: the alidade, the gyrocompass
scale, the hoey scale, the hoey arm, and finally
the space of the chart. Between the gyrocompass
scale and the hoey scale, the direction is repres-
ented as a string of digits. The chart is a special
medium in which the constraints of several lines
of position can be simultaneously represented.
The fix is literally a superposition of lines of po-
sition. This neatly fits Simon's characterization
of problem solving. The ship's situation is rep-
resented and re-represented until the answer to
the navigator's question is transparent. The ship
is where the lines of position intersect. Notice
that, although all the information required to fix
the position of the ship was present in the
 524/1540

bearing log, the ship's position was not apparent

in that representation.

The plotted fix position is compared against the

position that was projected after the previous
fix. Since both of these positions were graphic-
ally constructed in the space of the chart, the
comparison operation is implemented as a per-
ceptual judgement. This is not to say that it is a
simple process. What constitutes a significant
difference between the positions? When should
a navigator be worried about the process that
produces the position fixes? The "seeing" in-
volved in seeing the quality of the fix and its re-
lation to the projected fix is quite complex. If
the discrepancy is thought to be significant, it
may be used as input to a process that revises
the representation of the ship's speed and course.

The ship's future positions are projected from

the current fix position. These projected posi-
tions have several uses. They are
525/1540
 Page 127

used to ensure that this ship is not standing into

danger. They also are used to choose landmarks
for the next round of observations. And of
course, after the next fix is plotted, it will be
compared with the projected position to detect
the effects of current or changes in speed.

THE FATHOMETER
 527/1540

Meanwhile, the observed depth of the water un-

der the ship is represented as a mark on the
graduated paper of the fathometer. The paper of
the fathometer is another domesticated space,
arranged with culturally meaningful names for
depths. Reading the depth indicated on the paper
transforms the position of the pen mark in this
domesticated space into a portable digital rep-
resentationa numberwhich is also recorded in
the bearing record log. As was explained in
chapter 1, the lateral movement of the recording
paper under the depth-recording pen converts
elapsed time into distance traveled along the
ship's track. The movement of the pen from top
to bottom of the chart paper turns elapsed time
of the signal and echo return into a representa-
tion of the depth of the water under the ship in
the space of the strip chart. Running the two
motions simultaneously and superimposing
them on each other creates a representation of
the relationship of the ship's track to the depth of
 528/1540

the water under the ship. Here is another physic-

al device that superimposes elements of a relat-
ively direct representation of the external world
(the distance the pen moves before making con-
tact with the paper) onto elements of a culturally
constructed space (the marked paper used to re-
cord depth) in order to give meaning to the
world. It is another hardware implementation of
situated perception. It bears a kinship to the
row-and-column format of the bearing record
log, except in this case, the organization of rows
and columns is dynamically created by the ac-
tions of motors and the motion they produce in
time.
 529/1540

Like the bearings of landmarks, the depth of wa-

ter is converted to a digital representation in the
act of reading the depth scales on the recording
paper. This number is then propagated via the
phone circuit to the bearing record log where it
appears alongside the bearings of landmarks as a
number in a column. The depth of water indic-
ated by a number on the chart at or near the loc-
ation of the fix is corrected for tidal height and
compared with the observed depth. The correc-
tion and comparison operations are carried out
using
 Page 128

arithmetic operations. If the computed fix posi-

tion consists of a small triangle and the depth of
water at the fix point agrees with the measured
depth of water, the fix is taken to be correct. If
the charted and measured depths disagree, there
is reason to believe that one of them is in error.
This process of creating and comparing inde-
pendently generated constraints is a very general
procedure for error detection in this domain and
in many others.
 531/1540

It is often useful to consider the alternatives to

any representational scheme. Before the advent
of the echo sounder, the simplest way to meas-
ure depths was to lower a heavy weight on a
line. Of course, if the water was deep this pro-
cedure consumed time and energy. Matthew
Maury came up with an ingenious and surpris-
ing solution to the problem of measuring great
depths. He "made deep sea soundings by secur-
ing a cannon shot to a ball of strong twine. The
heavy weight caused the twine to run out rap-
idly, and when the bottom was reached, the
twine was cut and the depth deduced from the
amount remaining on the ball." (Bowditch 1977)
If one knows the length of the line and volume
of the ball it is wound onto, one can measure the
diameter of the ball and compute what fraction
of the volume of the ball, and therefore what
fraction of the length, was pulled off by the can-
non shot. Alternatively, the ball of twine may be
weighed before and after the line has paid out.
 532/1540

Knowing the weight of a given length of line

one can easily compute how much line was con-
sumed by the sounding.

Stepping Inside the Cognitive System

 533/1540

The basic computations of navigation could be

characterized at the computational, representa-
tional/algorithmic, and implementational levels
entirely in terms of observable representations.
On this view of cognitive systems, communica-
tion among the actors is seen as a process in-
ternal to the cognitive system. Computational
media, such as diagrams and charts, are seen as
representations internal to the system, and the
computations carried out upon them are more
processes internal to the system. Because the
cognitive activity is distributed across a social
network, many of these internal processes and
internal communications are directly observ-
able. If a cognitive psychologist could get inside
a human mind, he or she would want to look at
the nature of the representation of knowledge,
the nature and kind of communication among
processes, and
Page 129
 535/1540

the organization of the information-processing

apparatus. We might imagine, in such a fantasy,
that at some level of detail underlying processes
(the mechanics of synaptic junctions, for ex-
ample) would still be obscured. But if we could
directly examine the transformations of know-
ledge representations we might not care about
the layers that remain invisible. Any cognitive
psychologist would be happy enough to be able
to look directly at the content of the cognitive
system. With systems of socially distributed
cognition we can step inside the cognitive sys-
tem, and while some underlying processes (in-
side people's heads) remain obscured, a great
deal of the internal organization and operation
of the system is directly observable. On this
view, it might be possible to go quite far with a
cognitive science that is neither mentalistic (re-
maining agnostic on the issue of representations
"in the head") nor behavioristic (remaining com-
mitted to the analysis of information processing
 536/1540

and the transformation of representations "inside

the cognitive system").

Levels of Analysis and Hierarchy of Task

Reduction
 537/1540

As we have seen, the position-fixing task is im-

plemented in the manipulation of external rep-
resentations and tools. We can follow the trail of
representations quite a long way in some cases,
but from time to time the stream of representa-
tional state disappears inside the individual act-
ors and is lost to direct observation. Thus, while
such an analysis may tell us quite a lot about the
cognitive properties of the navigation system, it
does not, by itself, tell us much about the nature
of the processes and representations internal to
practitioners of navigation. The problem of indi-
vidual human cognition is not solved by this
analysis, but neither is it simply put off. The de-
scription of transformations of representational
state in the previous sections is both a descrip-
tion of how the system processes information
and a specification of cognitive tasks facing the
individual members of the navigation team. It is,
in fact, a better cognitive task specification than
can be had by simply thinking in terms of
 538/1540

procedural descriptions. And the task specifica-

tion is detailed enough in some cases to put con-
straints on the kinds of representations and pro-
cesses that the individuals must be using.

Thus far, I have given a computational descrip-

tion of navigation and have examined the rep-
resentational foundations for navigation
Page 130
 540/1540

and the algorithms by which the required com-

putations are accomplished. In the last section I
began to explore the implementational level of
description which specifies the "details of how
the algorithm and representation are realized
physically" (Marr 1982). The discussion of the
propagation of representational state from the
alidade to the chart was perhaps the most de-
tailed of these discussions. Here I brought the
description down to the level of implementa-
tional operations, such as aligning the arm of the
hoey with the appropriate point on the hoey
scale. Now, this operation is both an implement-
ation of the computation at the system level and
a cognitive task facing the individual plotter. As
such, one may ask of this computation how its
inputs and outputs are represented and what al-
gorithms are used to transform inputs into out-
puts. One might imagine a story like the follow-
ing (we will consider this analysis in more detail
later): The computation is to align the hoey
 541/1540

index with a particular value on the scale. By

observing the performance of this task, and es-
pecially by noting errors that are made, we may
place some constraints on the representations
that are used to perform the task. A key com-
ponent of the task is knowing the direction in
which scale values increase. This representa-
tional element may be used by an algorithm that
finds the target value by doing hill climbing on
the values with dynamic gain adjustment. That
is, if the index is currently far below the target
value it can be moved a large step toward the
target. When the index is near the target value, it
should be moved in smaller steps. Finally, we
might want to say how such representations and
algorithms are actually implemented in the
minds of the plotters. At this point, however, we
overreach the terms of our analysis. The details
of these internal processes cannot be directly ob-
served and must remain objects of speculation.
Notice that, although some of the
 542/1540

representations are internal, they are still all cul-

tural in the sense that they are the residua of a
process enacted by a community of practice
rather than idiosyncratic inventions of their indi-
vidual users. In the analysis that follows, I will
use culture as a resource in order to more pre-
cisely define the nature of the tasks that are ac-
tually engaged by the individual members of the
navigation team.

A Cognitive Account of a Navigator's Work

The computations of navigation are not platonic

ideals; they are real physical activities under-
taken by individuals manipulating
Page 131
 544/1540

real physical objects. Even though many of

them are symbolic activities and some of the
symbols are clearly represented inside the heads
of the practitioners, we must never forget that
symbols always have some physical realization
or that the nature of the physical form of sym-
bols constrains the kinds of operations to which
they can be subjected. In the previous section, I
described the major computations of the naviga-
tion task in terms of the propagation of repres-
entational state across a set of physical devices
and discussed the physical activities of the
members of the navigation team as they manipu-
late the devices that do the computation. That
discussion was both an analysis of the system-
level operation of the navigation system and a
specification of the tasks facing the individual
members of the navigation team. This task spe-
cification permits construction of the computa-
tional level of description for the individuals.
The present section presents the cognitive
 545/1540

requirements of performance of the navigation

task. What are the people in the setting doing?
Here I intend to finally engage what was for a
decade the central question of cognitive anthro-
pology: "What do they have to know in order to
do what they do?" Or, perhaps a better question,
"How do they go about knowing what they
know?"
 546/1540

Identifying the directly observable external

physical representational media involved in the
navigation task was easy. Even the task of de-
scribing their internal structures and the mech-
anisms of coordination among them was relat-
ively straightforward. With respect to structures
that are internal to the actors, we are, alas, much
more in the dark. It is possible to give functional
specifications to the structures that must be
present, but we cannot directly observe their in-
ternal organization, nor can we specify the
mechanisms of coordination by which represent-
ational state is propagated. These things are
simply beyond the reach of contemporary cog-
nitive science. In what follows, I will attempt to
push the theory of computation by the propaga-
tion of representational state as far as is possible
into the heads of the practitioners of navigation.
I will assume that a principal role of the indi-
viduals in this setting is providing the internal
 547/1540

structures that are required to get the external

structures into coordination with one another.

Because cognitive science has historically had

difficulty modeling the behaviors of the sensory
transducers that connect minds to the "outer en-
vironment," much more attention has been paid
to "processes that can go on inside the human
head without
Page 132
 549/1540

interaction with its environment" (Simon and

Kaplan 1989: 39). Furthermore, "deep thinking
has proved easier to understand and simulate
than hand-eye coordination" (ibid.) Unfortu-
nately, in order to get the cognitive game started
in a mind that is profoundly disconnected from
its environment, it is necessary to invent internal
representations of a good deal of the environ-
ment that is outside the head. This requirement
is simply not present in a mind that is in con-
stant interaction with its environment. The
mainstream thinking of cognitive science in the
past thirty years leads us to expect to have to
represent the world internally in order to interact
with it. This theory of "disembodied cognition''
(Norman, 1990) has created systematic distor-
tions in our understandings of the nature of cog-
nition. As we have seen, a good deal of the com-
putation performed by a navigation team is ac-
complished by processes such as hand-eye co-
ordination (see also Latour 1986). The task of
 550/1540

navigation requires internal representation of

much less of the environment than traditional
cognitive science would have led us to expect.

In the following, I will attempt to posit the min-

imum internal structure required to get the task
done. I choose the minimum because I would go
beyond this point with trepidation. In a study of
the conduct of science, Bruno Latour (1987) la-
ments the lack of studies of the forms and in-
scriptions in which scientific and technical
knowledge are concentrated. He suggests that
one might conclude from this that such studies
are not possible. Latour continues:
 551/1540

I draw a different conclusion; almost no one has

had the courage to do a careful anthropological
study offormalism. The reason for this lack of
nerve is quite simple; a priori, before the study
has even started, it is towards the mind and its
cognitive abilities that one looks for an explana-
tion of forms. Any study of mathematics, calcu-
lations, theories, and forms in general should do
quite the contrary; first look at how the observ-
ers move in space and time, how the mobility,
stability and combinability of inscriptions are
enhanced, how the networks are extended, how
all the informations are tied together in a cas-
cade of re-representation, and if, by some ex-
traordinary chance, there is something still un-
accounted for, then, and only then, look for spe-
cial cognitive abilities. (246-247)

I do not know whether I have fulfilled the terms

of Latour's instructions. But although I have de-
scribed the history, the use, the
552/1540
 Page 133

combination, and the re-representation of the

forms, something remains unaccounted for. Per-
haps it is not much, but it is something. As we
shall see, it is not special cognitive abilities.
Indeed, the cognitive abilities that navigation
practitioners employ in their use of the forms
and inscriptions are very mundane onesabilities
that are found in a thousand other task settings.
 554/1540

The fix cycle is truly a cycle of activity, with no

unambiguous beginning or end. Each step de-
pends on a previous step and feeds subsequent
steps. Of course, every real navigation perform-
ance must have had a first fix cycle, which must
have begun somewhere, but where in the cycle
the first round begins depends in uninteresting
ways on the circumstances of getting the activity
going. If the ship has just pulled away from the
dock, the fix cycle will begin with an estimated
position somewhere in the vicinity of the dock.
If the ship has been at sea and has just arrived in
coastal waters, the cycle may begin with a set of
observations of landmarks. For analytic con-
venience, we will begin with the symbol that
represents the projected position of the ship at
the time the fix observations will be made. This
is a position plotted on the chart.

Timing the Cycle

 555/1540

The fix cycle repeats at a specified interval. The

default interval for Sea and Anchor Detail is 3
minutes. That is, the entire cycle of activity
from mark signal to mark signal is 3 minutes.
The fix interval may be changed to another
value, such as 1, 2, or 6 minutes, on the basis of
needs of the ship. If the ship is in circumstances
in which it may quickly get into trouble, the in-
terval should be short. As the rate at which the
ship can approach danger decreases, the period
of the fix cycle can be increased. There are no
hard and fast rules for establishing the interval.
(The reasons for choosing 3 minutes as the de-
fault fix interval will be discussed later in this
chapter.)
 556/1540

With a specified fix interval, timing the fix re-

quires reading a timepiece. The implementation
at this level is up to the crew. Some procedures
specify the ship's clock as the source of the time
reference. The bearing recorder aboard the
Palau used his own wristwatch instead of the
ship's clock for two reasons. First, timing the
cycle requires knowing the time, computing the
next fix time, and
Page 134
 558/1540

vigilance to the clock. By removing his watch

and placing it at the top of the bearing record
log, the recorder made the time reference much
more convenient. This ship's clock was mounted
on the bulkhead at the back of the bridge and re-
quired the bearing recorder to look away from
the chart table to see the time. Second, the bear-
ing recorder's watch had a digital display. Fix
times have to be recorded in the bearing record
log, and it was easier to read and copy the digit-
al representation of the time than the analog rep-
resentation presented by the ship's clock. Addi-
tional cues that help the bearing recorder re-
member when to monitor the time reference in-
clude the pace of activities. For example, on a
3-minute fix interval, if no problems arise, the
plotter will have completed plotting the position
of the ship and will have projected future posi-
tions before it is necessary to choose the next
landmarks and prepare to begin the next cycle.
Other cues include the explicit remindings of
 559/1540

others. The plotter may say "Isn't it about time

for another round?" This illustrates the begin-
nings of a social distribution of cognitive labor
in which remembering is jointly undertaken.

Timing the fix cycle is a cognitive task that can-

not be reliably performed by the bearing record-
er without the aid of a mechanical timepiece.
The task of the bearing recorder is to coordinate
his actions with the behavior of the timepiece
and so permit the other members of the naviga-
tion team to coordinate their actions with his.

Identifying Landmarks
 560/1540

The task of identifying a landmark involves the

simultaneous coordination of many elements of
structure. In Sea and Anchor Detail, the choice
of a landmark is represented to the pelorus oper-
ator in the form of a spoken string of words:
"Point Loma Light," for example. The pelorus
operator must somehow get from this name to
both a description of the appearance of the land-
mark and a sense of where the landmark is in
the surrounding space. This must be represented
in some sort of memory. Although that memory
may not be the sort of storehouse of information
that many researchers assume, I think there is no
way to dismiss the fact that some internal rep-
resentation capable of producing a partial de-
scription of the landmark from the name of the
landmark must exist and must be attributed as a
structure internal to the human operator. Fur-
 Page 135

thermore, the appearance of the landmark may

depend on the pelorus operator's vantage, so a
single description will probably not suffice. The
sense of where the landmark is in the environ-
ment may guide the search of the surroundings,
which in turn changes the contents of the visual
field. Ultimately the identification of the land-
mark must arise from the coordination of the ex-
pectation about the appearance of the landmark
with the contents of the visual field.
 562/1540

Even though we do not know the mechanisms

by which these things are actually implemented
in the mind (and I take it to be a virtue of this
approach that we do not make any premature
commitment to such mechanisms), it is still use-
ful to speak of the propagation of representa-
tional state and the superposition of representa-
tional state in this description. Perhaps the ele-
ments of the description of the appearance of the
landmark are represented as images, or perhaps
as symbolic structures. Since this question has
not yet been satisfactorily settled in the cognit-
ive science community as a whole, we can
hardly expect to resolve it here on the basis of
observational data. But even without making a
commitment to a particular kind of representa-
tion and algorithm, we can observe the nature of
the task and provide a set of computationallevel
constraints for theories of cognition that aspire
to account for what people do "in the wild."
 563/1540

The identification of the landmark is a highly in-

teractive process, and it is likely that important
kinds of learning take place in every perform-
ance of this task. Whatever way it is implemen-
ted, it may be that all these representations are
simultaneously in coordination with one anoth-
er. That is, the representation of the landmark's
name, the expectations about the appearance of
the landmark, and the visual scene are all mutu-
ally constraining one another when the pelorus
operator fixates on the red-and-white tower and
declares "There is Point Loma Light." This is
another superposition of internal and external
structure on a single representational medium.
Scanning the currently visible scene may im-
prove the mental map and the representation of
the landmark itself and other recognized objects
in the scene. The successful visual search may
improve the description of the landmark stored
in memory and the association of the landmark
description to the landmark name.
 564/1540

The task is slightly different when the ship is not

in restricted waters and the entire fix cycle is
performed by a single watchstander. In that
case, the problem of identifying landmarks may
be
 Page 136

one of direct reconciliation of the chart and the

world. (We will investigate the cognitive con-
sequences of some of these differences in the
next chapter.) In some sense, the problem is
easier in Standard Steaming Watch because the
pelorus operator is also the plotter and the re-
corder. The navigator thus has access to the
chart itself as a representation of the landmark.
The chart is a much richer representation both of
the appearance of the landmark and of its rela-
tion to other objects in the world than is
provided by the spoken name of the landmark.
Still, the task is not without difficulties.
 566/1540

Consider the following situation. While the

Palau was steaming eastward, southwest of San
Diego Harbor, a quartermaster attempted to
identify the Coronado Islands, which lay about 7
miles south of the ship. The three islands were
clearly visible out the window of the pilothouse
just above the chart table. Of the three islands
on the chart, the leftmost island was labeled
"North Coronado" and the rightmost one was
labeled "South Coronado." Because the quarter-
master was looking to the south, however, North
Coronado was on the right and South Coronado
was on the left in the world (the reverse of their
positions on the chart relative to him). By map-
ping the spatial structure of the chart directly
onto the visible world, the quartermaster man-
aged to mistake North and South Coronado for
each other. Clearly he had relied on the spurious
but well organized spatial correspondences
between his perspective on the chart and his
view of the world. This example highlights two
 567/1540

points. First, human minds are good at finding

and projecting structural regularities. Second,
since this sort of misidentification is seldom
made by an experienced navigator, we must
wonder what internal structures are required to
do this task correctly.

Aiming the Alidade

Having identified a landmark in the world, the

pelorus operator must then aim the alidade at the
landmark. This process brings two external
structures into coordination with each other.
There is no need for an internal representation
of the hairline; it is built into the sighting tele-
scope, and the operator has the experience of it
in the visual field. Perhaps there is not even a
need to maintain an internal representation of
the description of the landmark once the land-
mark is in view, although one suspects that its
description and name may remain active during
the aiming operation. The coor-
568/1540
 Page 137

dination between the hairline and the landmark

is accomplished by reducing the distance
between the hairline and the landmark until they
are co-located in the visual field. This procedure
might be implemented in many ways.

Reading Bearings

Reading bearings in the alidade may be one of

the most complex cognitive skills required of
the quartermasters. Even though it is complex, it
takes place in a predictable world, and eventu-
ally it becomes an overlearned skill. In this sub-
section I will treat it in what may seem excruci-
ating detail. I do so because activities of this
kind are nearly ubiquitous not only in navigation
but in many of the everyday activities of inhab-
itants of the modern world and because, as far as
I know, no one has taken seriously the question
of just what is required, cognitively, to perform
them.
 570/1540

Reading the bearing requires the coordination of

at least four elements of structure:

the experience of the scale and the hairline

82866990447203f57400372e62d765b5.gif

a82866990447203f57400372e62d765b5.gif
knowledge of how to read digits

either a memory for the direction in which scale

values increase (clockwise by convention on the
circular scale, and to the right as seen through
82866990447203f57400372e62d765b5.gif
the viewfinder of the alidade), or a procedure for
establishing this direction

a82866990447203f57400372e62d765b5.gif
sequence of number names from zero to nine.

There are many ways to bring these elements in-

to coordination with one another in the bearing-
reading task. Let us first consider a fairly simple
(although not always efficient) method, and then
consider other possibilities.
 571/1540

The first two digits of any three-digit bearing

can be read directly from the scale itself. These
are the first two digits of the name of the nearest
labeled major tick to the left of the hairline. This
requires the scale and the hairline, the know-
ledge of how to read digits, and the knowledge
of the direction of increasing values.

One must, of course, know how to read the di-

gits in order to make use of the labels. This
knowledge is selectively brought into coordina-
tion with the labels on the major tick marks at
10° intervals on the scale. Many compasses have
only these first two digits of the labels for the
major ticks. This saves space on small instru-
ments, and the last digit (which is always 0)
may be left off, because
 Page 138

it is not required by the procedure used to read

the scale. Even though this is an overlearned
skill, it sometimes fails. We cannot know from
strictly observational data what the actual
sources of error are, but we can get some idea of
the kinds of considerations involved by looking
at the diagnoses of errors applied by other mem-
bers of the team. During my observations, the
following explanations for errors by pelorus op-
erators were offered by the bearing recorder or
the plotter:
 573/1540

The similar appearance of the printed represent-

ation of two digits on the face of the gyrocom-
pass card was offered to explain a report of 167
for an actual bearing that was reconstructed as
107. The 6 and the 0 have similar shapes. The
confusion of shapes may be facilitated by the
blurriness of vision caused by tears in the eyes,
which a pelorus operator sometimes gets from
working in the wind on the wing bridge.

An off-by-a-century error, e.g., 324 for 224.

This could be produced by a number of cognit-
ive mechanisms, including a data-driven error
(Norman 1981) involving increased activation
on the second digit. The crewmen offered no
specific hypothesis concerning the reason for
such an error, but they found it a plausible error
to have been made.
 574/1540

A repetition/substitution error, e.g., 119 for 199.

This one is probably due to a shift of activation
(Norman 1981).

A digit transposition, e.g., 235 for 325. Digit

transpositions are quite common in environ-
ments like this (Wickens and Flach 1988).

The occurrence of these errors, even if they are

rare, puts some constraints on the kinds of struc-
tures that must be involved in the performance
of the task.
 575/1540

Once the first two digits of the bearing have

been established, it remains to establish the last.
In the simplest case this is done by counting.
Counting is the coordination of an internally
generated sequence of number tags with a parti-
tioning of perceived unitary objects (Gelman
and Gallistel 1978). The knowledge of the direc-
tion of increasing values is required in order to
know that it is the first two digits of the label to
the left rather than the label to the right that will
be elements of the bearing name. The import-
ance of the knowledge of the direction of in-
creasing values can be inferred from the occur-
rence of errors in which a bearing n minor ticks
to the left of a labeled tick is reported as being n
units larger than the name of the major tick. For
example, a bearing of 257 (three ticks to
Page 139
 577/1540

the left of 260) will be reported as 263 if the dir-

ection of increasing scale value is inverted.
Knowledge of this direction can be remembered,
or it may be computed. If computed, the scale
increase direction is a representational state that
results from the coordination of the experience
of the scale with a procedure for finding the dir-
ection of increase. The latter may involve a
comparison of the magnitudes of the labeled
major ticks adjacent to the hairline. The know-
ledge of the direction of increasing scale values
is required in order to orient the partitioning
activity. The counting is accomplished by co-
ordinating the shift in attention from one tick to
the next (in the proper direction) with the trans-
ition from one number label to the next. It is not
necessary to remember all the correspondences
generated by this process; it is only necessary
that the correspondence between the number
name and the tick mark nearest the hairline be
produced. I believe we can assume that, for the
 578/1540

quartermasters, the number sequence runs fairly

automatically. No crew member ever attributed
a bearing-reporting error to the inability of the
pelorus operator to count from one to nine.

The use of the scale requires the abilities to pro-

duce the appropriate name (e.g., "1 3 6") for any
given tick mark and to locate the appropriate
tick mark when given any name between "0 0 1"
and "3 6 0." To do this the pelorus operator
needs the following:

a schema for the scale, with ticks (some of

which are labeled) in an ordered sequence

the abstraction of number sequence at least from

zero to nine

the ability to decompose a number representa-

tion into a decade (between 01 and 35)
 579/1540

the ability to use the number sequence locally

within a decade to determine the last digit of the
bearing

the ability to reassemble the decade and the last

digit into a whole number.

It is not necessary to posit an abstract internal

representation of the scale. Instead, the pelorus
operator works with the interpreted representa-
tion of this scale. This representation is caused
by attending to the scale and involves coordina-
tion with pre-stored structures that allow the
scale to be seen as a scale rather than as
something else. What is stored certainly need
not be an image of the scale.

If we asked a pelorus operator, while he was not

actually using the instrument, to tell us how the
scale-reading task is done, he
 Page 140

might make use of an imagined internal repres-

entation of the scale and imagine the operations
performed upon it. This would be a very differ-
ent task from reading a scale that was present.
An internal representation may be acquired
through continued interaction with the scale it-
self. What we know about internal representa-
tions of external structures, though, leads me to
believe that such a representation would be
schematic at best (Nickerson and Adams 1979;
Reisberg 1987).

Reporting and Remembering Bearings

 581/1540

Viewing language as one of the structured rep-

resentations produced and coordinated in the
performance of a task highlights language's
information-bearing properties. In cognitive sci-
ence, language is usually thought of primarily as
a human computational capacity that must be
understood in terms of the processing that indi-
viduals must do to produce or interpret it. Look-
ing at the role of language in the operation of a
system of socially distributed cognition leads us
to wonder about the properties of language as a
structured representational medium.
 582/1540

Traditional information theory fails us when we

approach spoken language. When a pelorus op-
erator reported the bearing of a landmark, how
much information was passed? The number he
reported is one of 360 possible full-degree bear-
ings. Does that mean that a bearing report car-
ries log2360 = 8.492 bits of information? Or,
since there are 1000 numbers that can be con-
structed of three digits, should we say that each
three-digit bearing carries log21000= 9.967 bits
of information? The problem is that the agree-
ment between the sender and the receiver con-
cerning the universe of messages and the ways
in which they will be coded is very weakly spe-
cified. The information-theoretic measures giv-
en above are irrelevant. What counts is what it
takes to understand what has been said, and un-
derstanding of language is poorly modeled by
classical information theory. Even in this highly
rationalized and predictable setting, there is no
previously agreed upon specifiable universe of
 583/1540

possible messages, and the ways of encoding

and decoding the messages are themselves ne-
gotiable at the time of communication. Rather
than attempt to force information theory onto
natural language, let us instead look at the prob-
lem of language understanding from the per-
spective of coordination of structured represent-
ational media. Utterances themselves are states
in structured
Page 141
 585/1540

representational media which we understand by

bringing them into coordination with both ex-
ternal and internal structured representational
media. Depending on the nature and the modal-
ity of the language expression, a great deal of
information may be entailed by what looks like
the transmission of a trivial message. The mes-
sage may be garbled and require partial recon-
struction by the hearer. The impact of the mes-
sage on the receiver depends on what the receiv-
er knows. For example, consider a case in which
a bearing of 059° is reported. To a novice, this
may be only a string of digits that is to be writ-
ten down in a book. To the experienced navigat-
or, this means a direction that is a little to the
east of northeast. When a knowledgeable navig-
ator hears or sees this bearing, he may know
which direction he is currently facing and may
actually feel the direction indicated by the bear-
ing as a physical sensation. For example, a nav-
igator facing west may hear "059" and
 586/1540

experience a sense of the direction to the right of

directly behind. This must involve the coordina-
tion of the cardinal direction schema with the
bodily frame of reference. This is quite similar
to what the Micronesian navigator does. The
differences in interpretation of such simple
verbal strings are easy to observe in the actual
interactions among members of the navigation
team. At one point, the bearing to Hotel del
Coronado was reported as 003 degrees. The
bearing recorder simply recorded and relayed
the reported bearing as a string of digits, but the
plotter, without plotting it, responded: "It better
not be. If it is we're pulling into Tijuana right
away!" In an interview, this same plotter, a
quartermaster chief, once described the ability to
feel bearings as directions in the local space
defined by bodily orientation as being able to
"think like a compass" and said it was
something he tried to teach all his men to do.
The bearing recorder's willingness to simply
 587/1540

record and report the impossible bearing is evid-

ence that he was not "thinking like a compass''at
least, not with respect to that bearing. This bear-
ing meant something more to the plotter than it
did to the bearing recorder because the plotter
brought it into coordination with a structured
representational mediumhis sense of directions
in local space.

Recording Bearings and Depths

The spoken representations of bearing and depth

must be transformed into written representations
for entry into the bearing record log. This part of
the task is relatively straightforward for
 Page 142

competent members of a literate culture. The

bearings and depths must be inserted in the ap-
propriate places in the table, and they must be
legible. One potential problem here is the ambi-
guity of symbols. A handwritten 2, for example,
may be indistinguishable from a handwritten Z.

Setting the State of the Hoey

 589/1540

Once the plotter has either read or overheard the

bearing to a landmark, it must be remembered
until it has been represented in the structure of
the hoey and the hoey configuration has been
locked. The task of locating the position on the
hoey scale that corresponds to the name of the
bearing is very similar to the task of reading the
bearing from the gyrocompass scale. As was
noted above, the hoey and the gyrocompass
scales are the interfaces between the digital and
the analog representations of the bearings.
Whereas the gyrocompass scale and the alidade
hairline are involved in an analog-to-digital con-
version, the hoey's arm and hairline index are
used to perform a digital-to-analog conversion.

Plotting and Evaluating the Fix

 590/1540

Once the hoey has been configured with the rep-

resentation of the bearing, the plotter must re-
member which landmark is to be associated with
the bearing. Sometimes the identity of the land-
mark is evident from the angle and the expected
location of the fix. There is a reconstructive
memory process here that may rely on the sim-
ultaneous coordination of the memory for the
landmarks chosen on the current round, the
physical shape of the configured hoey, and the
positional constraints provided by the arrange-
ment of symbols on the chart. The functional
system that realizes this memory clearly tran-
scends the bounds of the skull and the skin of
the individual plotter. If we were to characterize
this memory retrieval as a heuristic search, we
would have to say that the search (for the appro-
priate landmark to go with the bearing) is con-
ducted in the space of the chart itself by success-
ive positioning and repositioning of the hoey un-
til a fit between chart, hoey, and projected fix
 591/1540

position is found. The navigation system thus

remembers which landmark goes with the cur-
rent bearing, and most of the structure and pro-
cess of the memory function is external to the
human actor. The
 Page 143

heuristic part of the search lies in the plotter's

choices of ways to position the hoey to better
satisfy one or another of the spatial constraints
of the problem as they are represented in the
physical structure of the hoey and the chart. La-
tour (1986) calls this "thinking with the eyes and
hands."

TRANSLATION WITH PRESERVATION OF

DIRECTIONAL RELATION
 593/1540

On the Mercator projection, the straight edges of

rulers make straight lines which denote sets of
locations all bearing the same directional rela-
tionship to one another. A frequent component
of plotting tasks is the establishment of such a
directional relationship with respect to a direc-
tional referent and a given location. Thus far I
have discussed the use of the hoey, because that
was the tool of choice for the plotter aboard the
Palau. There are, however, several other tools
designed to do this same task, and each has
slightly different properties. There are two ma-
jor classes of these tools: those that have direc-
tional gradations (a protractor of some sort) built
in and need only to be aligned with a directional
referent, and those that simply translate direc-
tion. The tools in the latter class, parallel rulers
and pairs of triangles, rely on a compass rose
printed on the chart for the degree gradations.
Compass roses appear on some charts with both
magnetic and true orientation. On other charts
 594/1540

they are printed only with true orientation, and

on most non- Mercator-projection charts they do
not appear at all (since on most non-Mercator
projections directional relationships are not pre-
served). If a chart is designed for use with the
simple direction translating tools that rely on the
compass rose and on the translation of direction
from the rose to courses, or from relations
between points to the roses, several roses are
frequently printed so that for any particular op-
eration a compass rose will be nearby on the
chart. The farther one has to move a line of dir-
ection with parallel ruler or triangles, the greater
the probability of error.

Hoey
 595/1540

The hoey is also known as a one-armed pro-

tractor. Directional reference is established by
aligning the base of the protractor with any one
of the latitude lines on the chart. The latitude
lines provide a true east-west referent for the
direction scale of the protractor.

Consider the task of bringing a representational

state of the hoey into coordination with the
structure of the chart. One has to simultaneously
get the edge of the rule lined up with the symbol
of the
 Page 144

landmark on the chart and also get the base of

the hoey lined up with the directional frame of
the chart. The task has three degrees of freedom.
There are two degrees of freedom in getting the
base of the hoey aligned with the directional
frame of the chart (one rotational, the other
either vertical or horizontal depending on
whether the hoey is aligned with a parallel of
latitude or a meridian of longitude). The third
degree of freedom is in getting the edge of the
rule over the landmark symbol. It is difficult to
satisfy all of these at once, because one cannot
attend to all three at the same time (the hoey
base should be far from the landmark symbol in
order to present the longest possible line of posi-
tion), and a change in one tends to change the
others. There is a simple mechanical technique
for making this coordination easier to do. It re-
duces the problem from three degrees of free-
dom to two.
 597/1540

The technique is to place the point of a pencil on

the symbol of the landmark on the chart, bring
the edge of the rule up against the pencil point,
and then, keeping the edge in contact with the
pencil point, move the base of the hoey until it is
aligned with the directional frame of the chart.
This reduces the problem to one with two de-
grees of freedom, and permits the plotter to at-
tend visually to the rotational and lateral con-
straints while guaranteeing the satisfaction of
the landmark symbol position constraint with
gentle pressure on the hoey arm. In producing
the coordination between the hoey and the chart,
the task performer can transform the task to an
easier one by achieving coordination with an in-
ternal artifact: the knowledge of this technique.
When this skill is well learned it probably be-
comes an automatic motor skill, and experi-
enced plotters may find it difficult to describe
how it is actually performed.
 598/1540

Parallel Ruler

The parallel ruler is a pair of straightedges at-

tached to each other by a pair of diagonally
mounted bars. While one of the rulers is aligned
with the desired direction, the other ruler can be
moved away from the first, remaining parallel to
it. By alternately holding one ruler down on the
chart and moving the other, one can move a dir-
ection line anywhere on the chart. Parallel rulers
are awkward to use. Not only does their use re-
quire physical coordination to walk a line across
the chart; sometimes some planning must be
done to determine which sequence of walking
moves is required to get from the printed com-
pass rose to the desired point on the chart.
 Page 145

PMP
 600/1540

The standard plotting machine or parallel mo-

tion protractor (PMP) is also known as a draft-
ing machine. When using the PMP it is neces-
sary to tape the charts down to the table, be-
cause the directional reference is established via
the base of the PMP (which is bolted to the
table). If the orientation of the chart with respect
to the PMP changed, the directional referent
would be lost. Because the direction can be
locked into the PMP's arm, which is then free to
move in two dimensions while preserving the
selected direction, it is easy to use this to pass
the given direction line through any selected
point. The special technique required to get the
hoey into coordination with the chart is not
needed with this tool. The arm of the PMP is at-
tached to a platform that has two concentric
compass roses. These inner and outer roses can
be independently set to establish direction relat-
ive to any arbitrary reference. This can be useful
if it becomes necessary to plot bearings relative
 601/1540

to the ship's head rather than relative to true

north.

The hoey is perhaps a bit more difficult to use

than the PMP, because the 360° scale is folded
over into two scales of 180°° each. Efficient use
of the hoey requires additional strategies to get
the correct line of position. Because a single arm
position on the hoey represents both a bearing
and its reciprocal, the plotter must be able to de-
termine which bearing is to the landmark and
which is from it.

With the straightedge of the plotting tool in the

correct position on the chart, the plotter may
draw the line of position. Experienced plotters
almost never draw a complete line of position
extending from the landmark symbol to the pos-
ition of the ship. Instead they draw a short line
segment in the vicinity of the expected fix. The
judgement of what constitutes "the vicinity" of
the fix may take many factors into account.
 602/1540

Evaluating the Fix

Once three or more lines of position have been

drawn, the navigator should evaluate the fix. Is
it of good quality? Can it be trusted? Should
something different be done in order to improve
the quality of future fixes? The primary evid-
ence concerning the quality of the fix is the size
of the triangle formed by the three lines of posi-
tion. If the three lines do not intersect in exactly
the same
 Page 146

point, the ship's position is uncertain. There are

complex arguments about how to place a ship's
position relative to various shapes of triangles
(Bowditch 1977), but most navigators simply
assume that the ship is in the center of the tri-
angle and place a dot there as the fix point.
 604/1540

The displacement between the ship's projected

position for the fix time and the actual fix posi-
tion is a source of information about the quality
of the information used in constructing the
previous dead-reckoning position. If the ship has
not traveled as far as was projected during the
fix interval, it must not be traveling as fast as
expected. The change in speed may be due to a
change in the speed of the ship through the wa-
ter or to a change in the current through which
the ship is moving. Any information of this sort
derived from the comparison of the fix position
and the projected position must be remembered
in order to contribute to future projections.

Extending the Dead Reckoning Track

 605/1540

After the fix has been plotted and evaluated, the

dead-reckoning (DR) track of the ship should be
projected into the future for at least two fix in-
tervals. This requires the plotter to determine the
ship's heading and construct a track line from
the current fix position in the direction of the
ship's heading. The heading is available in the
form of a written number in the deck log. Again
the hoey is used to construct a line of position.
In this case, it is a line extending from the fix
position in the direction of the current heading.
Along this line, the plotter now predicts where
the ship will be at the end of each of the next
two fix intervals. To do this, the plotter must
know how fast the ship is traveling.
 606/1540

Mercator-projection navigation charts are not

normally printed with distance scales. (This is
because the measurement of distance is approx-
imate on the Mercator projection. The amount
of error depends on the distance measured and
the magnitude of its projection onto the north-
south dimension.) Instead they are printed with
latitude and longitude scales along the borders.
One minute of latitude corresponds to one naut-
ical mile. However, the length on the chart of a
minute of latitude increases as one moves away
from the equator. A reasonably accurate estim-
ate of distance can be had by using the latitude
scale at the mid-latitude of the segment to be
measured.
 Page 147

Measurements of distance on the charts are

made with dividers. These tools simply span a
given extent of space and permit that span to be
transferred to another part of space. To measure
the distance between two points on a chart, one
could span the distance with dividers and move
that span to a scale to read its magnitude in the
units of the scale. Like the hoey, the dividers are
a tool for capturing representational state and
moving it to another medium without distortion.

FOUR WAYS TO DO DISTANCE-RATE-

TIME PROBLEMS
 608/1540

The way a problem is represented can change

what is required of the problem solver. Suppose
the plotter has just plotted a fix and needs to
compute the ship's speed in nautical miles per
hour on the basis of the distance the ship has
moved in the interval of time that elapsed
between the current fix and the previous one. In
particular, suppose the two fix positions are
1500 yards apart and that 3 minutes have
elapsed between the fix observations. There are
at least four different ways to represent this
problem. Each representational condition re-
quires a different organization of cognitive
processes.

Condition 1 The task performer has the follow-

ing resources: paper and pencil, knowledge of
algebra, knowledge of arithmetic, knowledge
that there are 2000 yards in a nautical mile and
60 minutes in an hour, and knowledge that dis-
tance equals rate multiplied by time (D=RT).
 609/1540

Condition 2 The task performer has the same re-

sources as in condition 1, except that instead of
a paper and pencil the task performer has a four-
function pocket calculator.

Condition 3 The task performer has either a

three-scale nomogram of the sort shown in fig-
ure 3.3 or a nautical slide rule of the sort shown
in figure 3.4, and the knowledge required to op-
erate whichever tool is present.

Condition 4 The task performer has no material

implements at all, but knows how to use what
navigators call the "three-minute rule."

It is impossible to specify in advance exactly

how any particular person will actually do this
task under any of these conditions, but if the
person uses the resources in the ways they are
intended to be used it is not difficult to determ-
ine what is likely to be involved.
 610/1540

In condition 1, the task performer will first have

to use the knowledge of algebra to manipulate
the formula D=RT to the form
Figure 3.3 A three-scale no
 612/1540

Figure 3.4 A nautical slid

 Page 149

R=DIT so that rate can be solved for directly

from the given values of D and T. Then, the dis-
tance in yards will have to be converted to the
equivalent number of miles using the knowledge
of the number of yards in a mile and the know-
ledge of arithmetic. The time in minutes will
have to be converted to the equivalent number
of hours using the knowledge of the number of
minutes in an hour and, again, arithmetic. The
distance measure must be divided by the time
measure (arithmetic again) to get the rate. Of
course, these things can be done in a different
order (for example, the division could come be-
fore either of the unit conversions, or between
them), but in any case all these things must be
done at some point in order to solve the
problem.
 614/1540

The reader may want to try this as an exercise

just to get a feel for the sort of work that is in-
volved. I believe that this problem would tax the
abilities of many navigation practitioners in the
Navy, not because the arithmetic is difficult but
because the applications of the arithmetic opera-
tions must be planned so that the elements of the
solution fit together to produce the desired solu-
tion. One may be perfectly capable of doing
every one of the component subtasks in this
problem but fail completely for lack of ability to
organize and coordinate the various parts of the
solution.
 615/1540

In the calculator version, the procedures for do-

ing the arithmetic operations of division and
multiplication are restructured so that, instead of
constructing a pattern of symbols on a piece of
paper and decomposing the problem to a set of
operations on single digit arithmetic arguments,
one keys values into a calculator and pushes op-
erator buttons. Also, depending on the order in
which the steps are taken, it may be necessary to
remember a previous result and enter it into a
later operation after other operations have inter-
vened. I think this version of the task would also
tax the abilities of many navigation practition-
ers. The calculator makes the easy part of the
problem easier to do. The difficult part is decid-
ing how to coordinate the arithmetic operations
with each other, and the calculator provides no
support for that part of the task.
 616/1540

The paper-and-pencil condition and the calculat-

or condition are alike in that they utilize com-
pletely general computational routines. The
knowledge of the equation for distance, rate, and
time and the knowledge of the constants re-
quired for the unit conversions are specific to
the task, but they provide little help in structur-
ing the actions of the task performer. Because of
this, the
 Page 150

procedures for doing the computation are com-

plex. When we write them out at even the shal-
low level of detail given above, we find that
they contain many steps. If we actually got
down to counting each symbol written on the
paper or each key-press on the calculator as a
step (not an unusually detailed level for a cog-
nitive analysis), we would find that they each
contain many tens of steps.
 618/1540

Now consider the cognition required of the task

performer in condition 3. To use the nomogram,
one finds the value of the time on the time scale
and makes a mark there. One finds the value of
the distance on the distance scale and makes a
mark there. Then one draws a line through those
two marks with a straightedge and reads the
value of speed, in the desired units, where the
drawn line intersects the speed scale. The fact
that these scales are already constructed in terms
of the units set by the problem clearly gives this
condition a substantial advantage over the first
two conditions. This is a very common problem,
and the nomogram is designed specifically to
make its solution easy. The use of the nautical
slide rule is very similar to the use of the nomo-
gram. It, like the nomogram, is a medium in
which multiplication and division are represen-
ted as alignments of logarithmic scales. One
aligns the distance index with the desired dis-
tance on the distance scale (this could be yards
 619/1540

or miles; both are represented, side by side) and

aligns the elapsed-time index with the desired
time on the time scale (either minutes or hours;
both are present, side by side); the speed index
will then point to the speed in knots on the
speed scale.
 620/1540

Having the scales in the units set by the problem

eliminates the need to convert one kind of unit
into another. More important, knowledge of al-
gebra is not required for this condition of the
task. The nomogram and the slide rule transform
the task from one of computational planning
(figuring out what to divide by what) to one of
simple manipulation of external devices. In the
first two conditions, all that stands between the
task performer and the nonsensical expressions
"R = DT" and "R = T/D" is a knowledge of the
syntax of algebraic transformations. When one
is using the nomogram or the slide rule, the
structure of the artifacts themselves obviate or
lock out such relations among the terms. The re-
lations D=RT, R=D/T, and T=DIR are built into
the structure of the nomogram and slide rule.
The task performer has no need to know any-
thing about these relations, either implicitly or
explicitly. The correct relationships are built in-
to the tool; the task performer
621/1540
 Page 151

simply aligns any two scales to constrain the

value of the third. Even more important, the in-
correct relations are ''built out"it is not possible
to produce those relations with these tools.

On the nomogram, the time and speed scales

flank the distance scale. A line drawn between
any point on the speed scale and any point on
the time scale intersects the distance scale at a
point that is the averaged sum of the logarithm
of the time and the logarithm of the speed. Since
sums of logarithms are products, the physical
construction of the nomogram constrains the re-
lationships among the terms to be of the correct
type. Similarly, the slide rule is constructed so
that the distance reading is the angular sum of
the logarithm of the speed and the logarithm of
the time.
 623/1540

The task performer still needs to know

something, but the knowledge that is invoked to
solve the problem with these tools is less com-
plicated and less general than the knowledge re-
quired with the paper-and-pencil or the calculat-
or version. A good deal of what needs to be
done can be inferred from the structure of the ar-
tifacts, which constrain the organization of ac-
tion of the task performer by completely elimin-
ating the possibility of certain syntactically in-
correct relationships among the terms of the
computation. One may be more reluctant to say
that the answer was actually computed by the
task performer in condition 3 than in condition 1
or condition 2. It seems that much of the compu-
tation was done by the tool, or by its designer.
The person somehow could succeed while doing
less because the tool did more. But before we go
that far, let us consider the task in condition 4.
 624/1540

Where condition 3 utilized specialized external

artifacts, condition 4 utilizes a specialized in-
ternal artifact. Since 3 minutes is 1/20 of an
hour and 100 yards is 1/20 of a mile, the number
of hundreds of yards (twentieths of a mile) a
ship travels in 3 minutes (1/20 of an hour) is its
speed in nautical miles per hour. Thus, a ship
that travels 1500 yards in 3 minutes has a speed
of 15 nautical miles per hour. In order to "see"
the answer to the problem posed, the navigator
need only imagine the number that represents
the distance traveled in yards, 1500, with the
last two digits removed: 15. The distance
between the fix positions in the chart is spanned
with the dividers and transferred to the yard
scale. There, with one tip of the divider on 0, the
other falls on the scale at a tick mark labeled
1500. The representation in which the answer is
obvious is simply one in which the navigator
looks at the yard-scale label and
625/1540
 Page 152

ignores the two trailing zeros. A complex com-

putation is realized by a simple strategy of situ-
ated seeing in a carefully constructed
environment.
 627/1540

Experienced navigators actually apply this rule

in an even more compressed form. The three-
minute rule changes the distance itself into
speed. The distance I span with the dividers can
be interpreted directly as a speed rather than as a
distance, and the "yard" scale on the chart
(marked in hundreds of yards) is now not a site
of distance measurement that is converted to
speed; rather, the conversion has worked its way
upstream computationally, and a structure that is
sometimes read as the "yard" scale is now read
as a speed scale. An experienced navigator has
no need to imagine the distance in yards in order
to use the three-minute rule to compute speed.
The extent of space spanned by the dividers may
be a representation of a distance as it comes off
the chart, but this same extent of space becomes
a representation of speed as the dividers ap-
proach the scale (which is now being read as a
speed scale). When the dividers touch down on
the scale, the space they span represents what
 628/1540

the scale represents, and the scale represents

speed in knots.
 629/1540

When used in this way, the three-minute rule

utilizes the same interpretation of space as speed
that we saw in the use of the chip log line. If the
conditions of measurement can be constructed
in the right way, a distance traveled in a fixed
period of time can be read directly as a speed.
This seems to be a re-evolution of a conceptual
solution in a new technical medium. That is,
when charts and compasses became accurate
enough to measure and plot positions, the care-
ful juggling of units of distance, time and speed
to get simple solutions was re-created. In the
case of the chip log, the unit of distance between
the knots in the rope was constructed (41> feet
2/3) to yield a speed reading of that distance for
a given unit (30 seconds) of time. In the case of
the three-minute rule, the time span was manip-
ulated to fit the distance-speed relationship. This
is a nice example of the evolution of representa-
tions in the ecology of ideas.
 630/1540

The application of the three-minute rule is very

neat, but it will not be very useful if the condi-
tions under which it can be applied occur only
infrequently. In fact, this is not an unusual prob-
lem for a navigator. In the above discussion of
the task of setting the fix interval, it was noted
that 3 minutes is the default fix interval for Sea
and Anchor Detail. Three minutes is the default
precisely because this rule is so easy to use. The
navigation team is capable of per-
 Page 153

forming the fix cycle on 2-minute, or even

1-minute intervals. Three minutes is by far the
most common interval, not because it meets the
needs of the ship better than the other intervals,
but because it meets them well enough, and it
makes this computation so easy to do.
 632/1540

The nautical slide rule and the nomogram are

normally used only when the ship is away from
land and the fix intervals are much longer than 3
minutes. When the cycle is performed on inter-
vals of 1 or 2 minutes, speed is normally com-
puted by conversion to the 3-minute standard.
For example, if the ship travels 800 yards in 2
minutes, it would travel 1200 yards in 3
minutes, so its speed is 12 knots. And regardless
of the speed of the ship, as long as both the
speed and the fix interval are constant there is
no need to actually recompute the ship's speed
to project its position for the next fix. The dis-
tance traveled during the next interval will be
the same as that covered in the last interval, so it
can simply be spanned with dividers and laid on
the projected track line, without the distance or
the rate ever represented as a number.
 633/1540

All the methods for computing speed from dis-

tance and time have the same description at the
computational level. Each of them, however,
represents the inputs in different ways and ap-
plies different algorithms to those representa-
tions in order to produce the output. Each of
them implements the algorithms in operations
applied to physical entities. All involve the co-
ordination of representational structure that is
inside and outside the task performer, but each
calls on a particular collection of internal struc-
tures to be coordinated with external structure in
a particular way.

What we see here is a set of functional systems,

each of which is capable of making the mapping
from inputs to outputs but each of which organ-
izes a different set of representational media in
relation to one another.
 634/1540

These are low-level functional systems. We will

see later how they are embedded in larger func-
tional systems to construct the activity of the
navigation team.

What are these tools contributing to the compu-

tations? It has now become commonplace to
speak of technology, especially information pro-
cessing technology, as an amplifier of cognitive
abilities. Cole and Griffin (1980) show,
however, that the appearance of amplification is
an artifact of a commonly assumed but mistaken
perspective. When we concentrate on the
product of the cognitive work, cultural technolo-
gies, from writing and mathematics to the
Page 154
 636/1540

tools we have considered here, appear to ampli-

fy the cognitive powers of their users. Using
these tools, people can certainly do things they
could not do without them. When we shift our
focus to the process by which cognitive work is
accomplished, however, we see something quite
different. Every complex cognitive performance
requires the application of a number of compon-
ent cognitive abilities. Computing speed from
distance and time with a calculator involves
many component subtasks: remembering a sym-
bolic expression, transforming the expression,
determining which quantities correspond to
which terms of the expression, mapping the ex-
pression to operations on the calculator, finding
particular calculator keys, pressing the keys, and
so on. The application of these abilities must be
"organized" in the sense that the work done by
each component ability must be coordinated
with that done by others. If we now consider do-
ing the same task with the nomogram or with
 637/1540

the three-minute rule, we see that a different set

of abilities is enlisted in the task. None of the
component cognitive abilities has been ampli-
fied by the use of any of the tools. Rather, each
tool presents the task to the user as a different
sort of cognitive problem requiring a different
set of cognitive abilities or a different organiza-
tion of the same set of abilities.
 638/1540

The tools provide two things simultaneously.

First and most apparent, they are representation-
al media in which the computation is achieved
by the propagation of representational state. Se-
cond, they provide constraints on the organiza-
tion of action. This is most apparent in the way
that the nautical slide rule precludes the execu-
tion of operations that violate the syntax of the
computational description. The physical struc-
ture of the slide rule is not only the medium of
computation. By constraining the representa-
tional states that can be produced to ones that
are syntactically correct, it provides the user
with guidance as to the composition of the func-
tional system in which it will participate. In this
sense, these mediating technologies do not stand
between the user and the task. Rather, they stand
with the user as resources used in the regulation
of behavior in such a way that the propagation
of representational state that implements the
computation can take place.
 639/1540

There are two important things to notice about

the computational technology of the piloting
task. First, these tools and techniques permit the
task performer to avoid algebraic reasoning and
arithmetic. Those activities are replaced by
aligning indices with numbers on scales, or ima-
gining numerical representations and making
Page 155
 641/1540

simple transformations of them. Rather than

amplify the cognitive abilities of the task per-
formers or act as intelligent agents in interaction
with them, these tools transform the task the
person has to do by representing it in a domain
where the answer or the path to the solution is
apparent. Second, the existence of such a wide
variety of specialized tools and techniques is
evidence of a good deal of cultural elaboration
directed toward avoiding algebraic reasoning
and arithmetic. In fact, there are more methods
than I have presented here. The problem could
also have been solved by looking up the speed
in a table of distances, rates, and times. The
kinds of cognitive tasks that people face in the
wild cannot be inferred from the computational
requirements alone. The specific implementa-
tions of the tasks determine the kinds of cognit-
ive processes that the performer will have to or-
ganize in order to do the task. The implementa-
tions are, in turn, part of a cultural process that
 642/1540

tends to collect representations that permit tasks

to be performed by means of simple cognitive
processes.
 643/1540

Perhaps this should also give us a new meaning

for the term "expert system." Clearly, a good
deal of the expertise in the system is in the arti-
facts (both the external implements and the in-
ternal strategies)not in the sense that the artifacts
are themselves intelligent or expert agents, or
because the act of getting into coordination with
the artifacts constitutes an expert performance
by the person; rather, the system of person-in-
interaction-with-technology exhibits expertise.
These tools permit the people using them to do
the tasks that need to be done while doing the
kinds of things the people are good at: recogniz-
ing patterns, modeling simple dynamics of the
world, and manipulating objects in the environ-
ment (Rumelhart, Smolensky, McClelland, and
Hinton 1986). At this end of the technological
spectrum, at least, the computational power of
the system composed of person and technology
is not determined primarily by the information-
processing capacity that is internal to the
 644/1540

technological device, but by the role the techno-

logy plays in the composition of a cognitive
functional system.

Choosing Landmarks

The choice of landmarks to shoot for a fix is a

complex judgement that may take into account
many constraints. It is desirable to have an even
angular dispersion among the landmarks. Three
landmarks equally spaced at 120° intervals
would be ideal. However, rarely
 Page 156

are three such landmarks available from the pos-

ition of the ship at the time of the fix. It is essen-
tial to avoid landmarks that are too nearly in the
same or opposite direction from the ship. A con-
stant error in one of two lines of position that in-
tersect at a 30° angle will produce a position er-
ror twice as great as the same error will produce
in lines that intersect at 90°. It is useful to get
one bearing nearly ahead or astern, because that
will produce the best information about the rela-
tion of the ship to its desired track. Similarly, a
bearing near the beam provides information
about the position of the ship along its track and
is sometimes called a "speed line." Additional
constraints are imposed by the physical layout
of the ship. The port pelorus operator cannot see
any landmark located at an angle of less than
10°° off the bow, because a large mirror blocks
the view. All these constraints may contribute to
the choice of landmarks to shoot.
 646/1540

Navigators often use their arms or fingers to as-

sess the relationships among provisional lines of
position in the space of the chart or in the sur-
rounding space. I have seen a plotter stand on
the bridge and extend each arm toward a land-
mark, assessing the angle of intersection of his
arms. More frequently, the potential lines of po-
sition are constructed in gestures over the sur-
face of the chart. A plotter may point on the
chart to the symbol depicting the location of a
prospective landmark and then sweep his finger
across the chart to the projected location of the
ship at the time of the next fix. By constructing
several such provisional lines of position in the
air, a suitable set is chosen. The plotter must be
remembering or re-imagining the earlier lines as
others are added for consideration, since the
judgement cannot be based on the properties of
any line alone but must be based on the relations
among them. I suspect that the gestures help to
create and maintain these representations
 647/1540

through time. The memory of the trajectory of

the finger decays with time, but it seems to en-
dure long enough that several of these can be su-
perimposed on one another and on the perceptu-
al experience of the chart. Such a composite im-
age permits the navigators to envision the direc-
tional relationships among the provisional lines
of position. The provisional lines of position are
not drawn on the chart, of course, because they
would add a great deal of potentially confusing
and possibly dangerous clutter to the chart. A
different representational technology might per-
mit the temporary construction and evaluation
of provisional lines of position.
 Page 157

Pipelining Activities
 649/1540

Humans are opportunistic information pro-

cessors. An account of the pelorus operator's
activity that says "read the number and then re-
port it" is shaped by a powerful pedagogical
simplification. It is a very understandable ac-
count, but not a very accurate one. In fact, the
activities of aiming the alidade, reading the
bearing, and reporting the bearing are often in-
terleaved. Standing on the wing, one can some-
times see the pelorus operator swing the alidade
to the approximate direction of the landmark, re-
port the first two digits of the bearing, and then,
with a slight pause in the report, turn the alidade
a bit more and pronounce the last digit of the
bearing. In extreme cases even the locating of
the landmark may be interleaved with the sub-
sequent activities. Thus, one might observe a re-
port punctuated by short pauses in which non-
report work is being done: "Hotel del (short
pause to swing alidade), 0 4 (short pause to
count to last digit) 3." In such a case many
 650/1540

representational structures may be simultan-

eously in coordination.
 651/1540

If the unit of analysis is defined by the media

that are actually in coordination, we should
change the unit of analysis as different aspects
of the task are considered. When the pelorus op-
erator reads the bearing, the visual scene outside
the pelorus is no longer a salient part of the co-
ordination, and the mental structures that encode
the description of the landmark are no longer
needed. The coordination event that accom-
plishes the reading of the bearing spans a differ-
ent set of structures; a different set of media are
brought into coordination. Again, the normally
assumed boundaries of the individual are not the
boundaries of the unit described by steep gradi-
ents in the density of interaction among media.
Now the unit includes the degree scale, the hair-
line, and perhaps the structures required to read
the digits printed on the scale and count the tick
marks lying between the labeled tick mark and
the hairline. The bearing taker is still the task
performer, but now different aspects of the
 652/1540

bearing taker's knowledge and different struc-

tures in the environment are brought into co-
ordination. The active functional system thus
changes as the task changes. A sequence of
tasks will involve a sequence of functional sys-
tems, each composed of a set of representational
media.

Sometimes the reading is co-articulated with re-

porting and recording the bearing. The bearing
recorder may have already recorded the first di-
gits of the bearing before the bearing taker
Page 158
 654/1540

has read the last digit. In that case, the unit of

coordination includes the activities of two per-
sons. The two actors could be said to be in co-
ordination with one another in a profound sense
(more on this later). The representational state
of the pelorus is propagated onto the representa-
tional state of the bearing log by the coordinated
actions of two people. But what is in coordina-
tion is more than the two people or the know-
ledge structures of the two people. The function-
al system includes these components and the ar-
tifacts themselves at the same time. When the
bearing is recorded as it is being observed, the
chain of coordination may include the name of
the landmark, partial descriptions of the land-
mark, the visual experience of the landmark, the
hairline of the alidade, the gyrocompass scale,
the knowledge and skills involved in reading the
bearing, the spoken sounds on the phone circuit,
the knowledge and skills involved in
 655/1540

interpreting the spoken bearing, and the digits

written in the bearing record log.
 656/1540

It is tempting to offer the sequential account as

the nominal account and then explore instances
in which the subsequences overlap. Doing this
makes the exposition clearer, but it does viol-
ence to the phenomena. In fact, the maximally
co-articulated case is typical. The elements can
be strung out sequentially only by a deliberate
process that involves other structures. If report-
ing and recording are co-articulated with the
reading of the bearing, for example, neither the
bearing taker nor the recorder need remember
the bearing. If this same task is strung out serial-
lyread, report, recordthen the memories of both
must be brought into play. The serial sequential
account is the one that appears in the written
procedures that describe the task. These ration-
alized versions are easier to think about, under-
stand, and promulgate than a description of how
the system actually works, but it would be both
difficult and inefficient to do the job the way it
is described in the written procedures.
 657/1540

When we turn to the coordination events and see

all the media that are simultaneously in coordin-
ation (some inside the actor, some outside), we
get a different sense of the units in the system.
When a bearing taker sights a landmark, we can
imagine that there is coordination between the
landmark name and the memory for landmark
descriptions that assumes a state of description
of the desired landmark, and then that is co-
ordinated with some aspects of the visual scene
containing the landmark itself, and then the hair-
line of the pelorus is superimposed upon that.
All these things
 Page 159

are in coordination with one another at once.

This coordination produces the representational
state in the pelorus that is then read in the indic-
ated bearing. Here we have a coordination of
two external structures (visual scene and hair-
line) with an internal structure (landmark de-
scription) that is the result of the coordination of
two internal structures (landmark name and
memory for landmark descriptions).

Constructing the Task Setting

 659/1540

The activities described in the previous section

take place in a carefully organized task setting.
One cannot performed the computations without
constructing the setting; thus, in some sense,
constructing the setting is part of the computa-
tion. In the most straightforward sense, the set-
ting is constructed by means of preparation pro-
cedures that are performed before navigation is
begun. The Watch Standing Procedures specify
the following steps in preparation for Sea and
Anchor Detail:

8. Preparations for Sea and Anchor Detail. Prior

to Sea and Anchor Detail, the Assistant to the
Navigator will ensure that:

a. All approach and Harbor charts are laid

82866990447203f57400372e62d765b5.gif
out with:

(1) Track with courses and distances

82866990447203f57400372e62d765b5.gif
labeled
 660/1540

(2) Turn bearings taking into account the

tactical characteristics of the ship (utilize
82866990447203f57400372e62d765b5.gif
15 degrees rudder and 10 knots of speed)

(3) Danger bearings/ranges where ever

necessary, especially if the ship must
82866990447203f57400372e62d765b5.gif
head straight at a shoal

(4) Outline in bright indelible marker all

hazards to navigation and all soundings
82866990447203f57400372e62d765b5.gif
of thirty feet (5 fathoms) or less

(5) A convenient yard scale for

82866990447203f57400372e62d765b5.gif quick use

(6) If anchoring, lay out the anchorage as

recommended in the Officer of the
82866990447203f57400372e62d765b5.gif
Deck's Manual

b. All pertinent publications are corrected,

and marked, the information reviewed by the
82866990447203f57400372e62d765b5.gif
Navigator, and plotter.

c. Tides and currents graphed and posted.

82866990447203f57400372e62d765b5.gif
 661/1540

d. If possible, the gyro error is determined

82866990447203f57400372e62d765b5.gif
within one hour of stationing the detail.

e. Qualified personnel are assigned to each

82866990447203f57400372e62d765b5.gif
position.

Getting the Right Charts onto the Table in the

Right Order

While a ship is entering a harbor, several charts

may be required. The change from one chart to
another must be accomplished quickly. It is im-
portant to get a new fix on the new chart as soon
as possible. It would not do to be digging in the
chart collection looking for the right chart when
navigation needed to be done. A typical
 Page 160

warship carries more than 5000 charts, so the

search could be very time consuming. Further-
more, the charts are stored forward, near the
auxiliary conning space. Therefore, all the
charts that will be needed for a particular detail
are assembled ahead of time and arranged on the
chart table, one over the other, in the order in
which they will be needed. This makes it pos-
sible to change charts quickly by simply pulling
the top chart off and exposing the next chart.
Extra copies are stored in the chart table so that
they can be brought up quickly if the charts in
use are inadvertently destroyed or are collected
as evidence in the case of an incident.

Computation of Entry (Exit) Track

 663/1540

The task of navigating a ship into a harbor can

be greatly simplified by planning the ship's track
ahead of time. This consists of plotting on the
chart the courses to be steered, the lengths of the
legs, the locations of turns, the identification of
turn bearings, and the landmarks to be used to
determine turn bearings. Of course, constructing
this track requires considerable effort. It also re-
quires the determination of the tactical charac-
teristics of the ship. Recall that ships do not turn
on a dime. After the rudder is put over, a ship
will move forward and to the side some distance
before it reaches its new desired heading. The
amount it moves forward is called advance; the
amount it moves to the side is called transfer. It
is possible to compute these values, but it is
much simpler to look them up in an advance and
transfer table (figure 3.5). These tables change
the nature of the computations involved in
preplotting the track, which in turn changes the
nature of the computations done during Sea and
 664/1540

Anchor Detail. The advance and transfer tables

make it easier to plot the projected track, and the
projected track makes it easier to provide sup-
port to the officer of the deck. We can say more
than that these tasks have been made easier. By
specifying the changes in the cognitive require-
ments of the tasks that must be performed, we
can say how they have been made easier. The
structure of these artifacts is such that the re-
quired computations can be achieved by simpler
cognitive processes in interaction with these ar-
tifacts than by the method they replace.

These features of the planned route of the ship

do not change from one entry of the harbor to
another, so they are often plotted on the chart in
ink. They thus become permanent features of
the chart. Since the exit track is different from
the entry track (ships keep to
 Page 161

the right of a channel), a chart is usually made

up for entry and another for exit, at least for the
home harbor. Yard scales showing distances pri-
or to the turns are plotted directly on the ship's
projected track. All this could be done while the
ship is entering port, but doing it ahead of time
changes the cognitive requirements of the tasks
that are to be done. One of the most important
tasks of the navigation team is to provide advice
to the OOD on the progress of the ship. Having
the chart properly made up ahead of time makes
this task much easier. For example, if the de-
sired track has been plotted already, displace-
ment left or right of track can be measured dir-
ectly. The information regarding the next course
is ready at hand and need only be read off the
chart after the position has been plotted. On a
chart made up in this way, the number of yards
to the next turn need not be measured; it is
available by simple inspection.
 666/1540

Further customizations to the chart include

drawing in the critical depth contours (areas of
water that are shallower than the draft of the
ship), plotting bearings of landmarks that will
keep the ship out of dangerous areas, and laying
out the approaches to an anchorage. As with the
projected tracks, these modifications to the
structure of the chart are specific to the ship and
support computations that could otherwise be
done on-line, but only at the cost of greater cog-
nitive demands during the actual performance of
the main task.

Updating the Chart

 667/1540

Charts must be kept updated. Channels silt in,

sandbars move, new buildings are constructed,
and lighthouses may be demolished and rebuilt.
Since it is expensive to publish a new edition of
a chart, periodic notices of changes are pub-
lished and distributed. These changes should be
incorporated in a ship's existing charts. Further-
more, the crew may choose to add landmarks to
a chart that are not yet on it or on a published
change notice. The crew of the Palau added a
new tower to their chart by establishing bearings
to the tower from several position fixes acquired
while exiting the harbor. Doing this meant
adding workload to one performance of the task
in order to make future performances less effort-
ful or more flexible.
 668/1540

Charts are thus customized in ways that trans-

form the nature of the work that can be done
with them and change the cognitive require-
ments of doing that work. The customizations to
the chart
670/1540
671/1540
 672/1540

Standard Tactical Diameter, 15000 Yards - Standa

Angle of Turn Advance Transfer Angle of Turn

15° 500 38 105°

30° 680 100 120°

45° 827 207 135°

60° 940 347 150°

75° 1007 513 165°

90° 1020 687 180°

Figure 3.5 A table of advance and transfer. These descripti
characteristics are determined empirically and are used to pla
stricted waters.
(From Maloney 1985.)
 Page 163

are bits of structure that will become compon-

ents of cognitive functional systems that operate
during the performance of the navigation task.
The computations are distributed in time, and
these customizations to the chart are examples
of moving computation out of the high tempo
activities of Sea and Anchor Detail by doing the
computation ahead of time.

Computation of the Compass Deviation Table

 674/1540

Although the magnetic compass is not normally

used as a primary navigation instrument, when it
is called into service it becomes very important.
Every magnetic compass has small errors that
are a consequence of the electromagnetic envir-
onment of the compass. These are corrected
when possible. Often some minor errors remain.
In order to remove these errors, a deviation table
is constructed. The error of the compass on
headings at 10° intervals is observed empirically
and entered into a table. The entries in this table
can be used later to compensate for errors in
compass readings. The deviation table must be
constructed ahead of time when a reliable direc-
tional referent is available.

The Compilation and Posting of the Tide and

Current Graphs
 675/1540

As each fix is plotted, the navigation team must

verify that the depth of water observed under the
ship corresponds to the depth of water shown on
the chart for the location of the fix. This com-
parison would be trivial if there was no tidal
variation. The actions of the tide cause the water
to be shallower at some times and deeper at oth-
ers. In some places tidal ranges of 15 feet are
common. Depths reported on the chart are given
with respect to the tidal datum of the chart, a
reference with respect to which tidal heights are
also reported. If the reported tide is 0 feet, then
the depths reported on the chart should be accur-
ate. If the reported tide is +4 feet, then the water
will be 4 feet deeper than the depths shown on
the chart. Predictions of tidal movement are
provided in tables.

The tides are one of the messier aspects of nav-

igation. In the open ocean, tidal fluctuations are
relatively regular. Where tides
676/1540
Page 164
 678/1540

interact with land (and especially in harbors,

where they interact with complex basin shapes),
the regularities become quite complex. When
planning maneuvers in a harbor, one must com-
pute a general tidal amplitude and phase for the
tidal movement. This may be a good description
of the tidal movement at one point; however, as
the tide surges into and out of the harbor, the
peak tide occurs at different times and with dif-
ferent amplitudes at different points in the har-
bor. The navigator must be able to compute the
times and heights of the tides at all points along
the planned route. Additional tables are pub-
lished showing the corrections of phase and
amplitude at selected points in the harbor. Figur-
ing the tides on a basin with a complex shape,
such as San Francisco Bay, can be a real night-
mare. In preparing to enter or exit a harbor, the
navigation team will construct a set of graphs of
the height of the tide at a sample point (or sever-
al sample points) in the harbor along the
 679/1540

planned track. These graphs provide corrections

to the calculations of the expected depth of wa-
ter under the ship for comparison to the depth
observed by the fathometer operator.

The crew also produces a current graph showing

the direction and velocity of the tidal current at
selected locations. These currents can be quite
strong in bays and harbors. They affect the
speed made good over the ground (a standard
nautical term), and this can be a very important
consideration affecting rate calculations and
steering commands.

Cashing In the Precomputation Precomputa-

tions Redistribute Cognitive Workload across
Time
 680/1540

Many of the elements of the preparation for Sea

and Anchor Detail involve the performance of
the parts of the anticipation computations. It is
easy to see that one of the effects of doing as
much as possible ahead of time is a reduction in
the amount of work that has to be done in the
high-tempo phases of the Sea and Anchor De-
tail. Because the task is event driven, and since
the quartermasters do not have the option of
quitting or starting over, this is an important
strategy for keeping the workload within the ca-
pacity of the navigation team. Stacking the
charts on the table in the order in which they
will be needed is an example of this sort of re-
distribution of effort across time.
 Page 165

Precomputations Transform the Tasks

Performed
 682/1540

Many of the elements of the preparation do

more than simply move some the computational
activity out of the time-limited performance of
the main task. They create new structures that
change the cognitive nature of the tasks that
must be done in the time-limited performance of
the main task. The pre-plotted entry track trans-
forms the task of generating advice for the of-
ficer to the deck. The ''convenient yard scale for
quick use," in concert with a pair of dividers and
the three-minute rule, becomes an element in a
functional system that performs distance-rate-
time calculations via perceptual inferences. Tide
graphs, turn bearings, and danger ranges have
similar roles to play in the on-line computations.
The activity of the fix cycle collects all these
constraints together in a medium where their
combined effects interact. The precomputations
are saved representational structures that trans-
form the nature of the task performance. They
aren't just doing part of the task ahead of time,
 683/1540

they are doing things ahead of time that make

the task easier to do. Thus, in the distribution of
cognitive effort across time, the integral of ef-
fort over time is not the same in the case of do-
ing part of the task ahead of time as it is in do-
ing a restructuring precomputation ahead of
time. In the latter case, the total effort may actu-
ally be less.
 684/1540

Of course, the construction of the chart itself

could also be described in exactly this way. It is
a structure that is constructed ahead of time and
changes the nature of the computation that is to
be done in the period of high-tempo activity.
The Micronesian navigator's imagery of the
voyage is very much like a chart in that it is a
precomputed structure onto which observations
can be mapped so that the answer to the pressing
question is literally staring the navigator in the
face. A great deal of prior experience is distilled
in the structure of the chart and in the Microne-
sian navigator's knowledge of the stars and seas.
In both cases, the structure that is the center-
piece of the computational functional system is
not something that could have been created by
the navigator alone. In both cases, there is a gen-
eral framework onto which specific observations
that are local in time and space are projected.
 685/1540

Precomputations Capture Task-Invariant Prop-

erties on Multiple Time Scales

Each of these precomputations is a way of

building local invariants into the structure of the
tools that are used in the performance of
Page 166
 687/1540

the navigation task. The invariants are tempor-

ally local in the sense that none of them lasts
forever. They are local with respect to the per-
sistence of the invariant they encode. Structure,
both in the world and in representations of the
world, endures over variable periods of time. In
general, where structure in the world is invari-
ant, the computation can be made more efficient
by building representations of the invariants into
the representations of the world. The chart, as
published, encodes invariants of the navigation
environment. These are relatively long-lasting,
and so is the chart when properly cared for. The
publication of notices to mariners and the updat-
ing of the chart in accordance with them are
concessions to the fact that the invariants of the
chart may be longer-lived than those of the
world it represents. Many of the customizations
to the chart are based on the physical character-
istics and handling properties of the shipthe loc-
ations of turn bearings and the danger depth
 688/1540

contours, for example. These are invariants that

could not be encoded on the chart as published,
since the chart must serve the needs of many
ships. These items are, however, invariant so
long as the chart is used aboard this ship, so the
crew adds them indelibly to the chart. Shorter-
lived invariants, such as those pertaining to any
particular transit of the waters depicted on the
chart, are marked in pencil or some other eras-
able medium. Some ships even add an acetate
overlay over the track lines on frequently used
charts so that accumulated erasures do not des-
troy the underlying features. Invariants that are
specific to a particular fix are captured in the
structure of malleable media such as the hoey, in
which state is preserved only as long as the
locking screw is twisted down tight.
 689/1540

Different tools may permit different sorts of in-

variants to be captured. Consider the difference
between the parallel motion protractor (PMP)
and the hoey in the plotting of magnetic bear-
ings. With the PMP, one can add magnetic vari-
ation into the structure of the device itself by ro-
tating and locking the outer compass rose. With
the PMP so configured, magnetic bearings can
be plotted and "automatically" corrected for
magnetic variation. This is useful because mag-
netic variation is an approximate invariant over
local space (within tens of miles) and local time
(within tens of years). The cost of capturing this
invariant with the PMP is that it depends on
maintaining the orientation of the charts with re-
spect to the base of the instrument. The charts
must therefore be taped to the chart table. It is
not possible to build the invariant of magnetic
variation
 Page 167

into the structure of the hoey. Plotting magnetic

bearings with the hoey requires additional arith-
metic operations to correct each line of position,
because the hoey captures only the more local
invariant of the bearing itself. The lock on either
the hoey or the PMP provides a way to "write"
or "save" the current value in the structure of the
device.
 691/1540

All these precomputational activities are in-

stances of a wider class of computational phe-
nomena called modularity. They are modular in
the sense that they remove from local computa-
tions any aspects that are invariant across the
spatial and temporal extent of the computation.
Thus, in addition to redistributing computation
in time and transforming the character of com-
putations, these precomputational elements
eliminate redundant computation. One must
know the relationship of the observed fix posi-
tion to the desired position, but one does not re-
construct the projected track at every fix or even
on every transit of the waters depicted. Since the
desired track is invariant across all entries of
this particular harbor, it need be constructed
only once if it is constructed in a medium that
endures. The relationship of charts to plotting
procedures thus implies a particular modularity
of the task. This will become clearer in the
 692/1540

consideration of the adaptation of the system to

change in chapter 8.
 693/1540

The chart table and its environs are preconstruc-

ted by the crew to serve the purposes of the task.
The prepared chart is a layered set of representa-
tions, each of the layers matching the temporal
or spatial extent of invariants that it represents.
This permits the final computation to be per-
formed by adding a final layer of short-lived
representations to a representational medium in
which many layers of longer-lived invariants
have already been superposed upon one another.
With three quick strokes of a pencil, the plotter
places the lines of position in coordination with
one another and in coordination with a deeply
layered representation of the world and the
ship's relationship to it. There, instantly, is the
position of the ship in relation to the land, to the
previous positions, to the desired track, to pos-
sible future positions, to the depth of the water,
to the next turn, to the latitude and longitude
grid, and more.
 694/1540

Precomputations Are a Window on a Cultural

Process

The computation of the present fix relies on the

most recent setting of the hoey, which was done
a few seconds ago. The present
Page 168
 696/1540

computation also involves the projection of the

dead-reckoning position, a piece of work that
was done just a few tens of seconds ago; on the
tide graphs that were constructed a few hours
ago; on the changes to the chart that were plot-
ted a few days ago; on the projected track and
the turning bearings, which were laid down
when this chart was "dressed" a few weeks ago;
on the placement of the symbols on this chart,
which was done upon the publication of the new
chart issue a few years ago; on the nature of the
plotting tools, which were designed a few dec-
ades ago; on the mathematics of the projection
of the chart, which was worked out a few cen-
turies ago; and on the organization of the
sexagesimal number system, which was de-
veloped a few millenia ago. It may seem silly to
tie this moment's navigation activity to a number
system developed by the Babylonians. Notice,
however, that when we get down to looking at
the details of the practice we see that the most
 697/1540

mundane facts about the representations chosen

have consequences for the sorts of things that
are easy to do and those that are difficult to do.
This was one of the lessons of the comparison
with Micronesian navigation. These details
present certain opportunities for error while pre-
cluding others. This system is part of a cognitive
ecology in which the various representational
technologies constitute one another's functional
environments. Since the presence of any one
technology may change the computational po-
tentials of the others, it is not possible, in prin-
ciple, to exclude any contributing element from
the list of precomputations simply on the basis
of its antiquity.
 698/1540

Given that each of these elements makes as

large a contribution to the computation as any
other, we may wonder where we should bound
the computation in time. It will not do to say
that the current computation is bounded by this
second, for clearly it spans many seconds. Per-
haps we should use the navigators' partitioning
of the world into meaningful pieces and say that
the current computation is the fix and is tempor-
ally bounded by the fix interval. That too is a
fiction, because the fixes are elements of a fix
cycle and any starting point is as arbitrary as any
other. We may attempt to put temporal bounds
on the computation that we observe now, today,
in any way we like, but we will not understand
that computation until we follow its history back
and see how structure has been accumulated
over centuries in the organization of the material
and ideational means in which the computation
is actually implemented.
 699/1540

This is a truly cultural effect. This collection

through time of partial solutions to frequently
encountered problems is what cul-
Page 169
 701/1540

ture does for us. Simon (1981) offered a parable

as a way of emphasizing the importance of the
environment for cognition. He argued that, as
we watch the complicated movements of an ant
on a beach, we may be tempted to attribute to
the ant some complicated program for construct-
ing the path taken. In fact, Simon says, that tra-
jectory tells us more about the beach than about
the ant. I would like to extend the parable to a
beach with a community of ants and a history.
Rather than watch a single ant for a few
minutes, as psychologists are wont to do, let us
be anthropologists and move in and watch a
community of ants over weeks and months. Let
us assume that we arrive just after a storm, when
the beach is a tabula rasa for the ants. Genera-
tions of ants comb the beach. They leave behind
them short-lived chemical trails, and where they
go they inadvertently move grains of sand as
they pass. Over months, paths to likely food
sources develop as they are visited again and
 702/1540

again by ants following first the short-lived

chemical trails of their fellows and later the
longer-lived roads produced by a history of
heavy ant traffic. After months of watching, we
decide to follow a particular ant on an outing.
We may be impressed by how cleverly it visits
every high-likelihood food location. This ant
seems to work so much more efficiently than
did its ancestors of weeks ago. Is this a smart
ant? Is it perhaps smarter than its ancestors? No,
it is just the same dumb sort of ant, reacting to
its environment in the same ways its ancestors
did. But the environment is not the same. It is a
cultural environment. Generations of ants have
left their marks on the beach, and now a dumb
ant has been made to appear smart through its
simple interaction with the residua of the history
of its ancestor's actions.
 703/1540

Simon was obviously right: in watching the ant,

we learn more about the beach than about what
is inside the ant. And in watching people think-
ing in the wild, we may be learning more about
their environment for thinking than about what
is inside them. Having realized this, we should
not pack up and leave the beach, concluding that
we cannot learn about cognition here. The envir-
onments of human thinking are not "natural" en-
vironments. They are artificial through and
through. Humans create their cognitive powers
by creating the environments in which they ex-
ercise those powers. At present, so few of us
have taken the time to study these environments
seriously as organizers of cognitive activity that
we have little sense of their role in the construc-
tion of thought.
 Page 170

Where Is the Computation?

Relating Mental Activity to Computation

Ship navigation involves lots of numbers. Num-

bers have to be processed in order to find out
where the ship is and especially to determine
where it will be. It is easy to assume that navig-
ators must be good at arithmetic. When I looked
closely at the practice of navigation, however, I
found the navigators engaged in very few arith-
metic tasks. How can that be?
 705/1540

It must be evident by now that the computations

performed by the navigation system are not
equivalent to the cognitive tasks facing the indi-
vidual members of the navigation team. It is
possible to describe the computations performed
by the navigation team without recourse to the
cognitive abilities or activities of the individual
members of the team. I have done so above in
the description of the propagation of representa-
tional state across a set of structured representa-
tional media. The navigation system combines
one-dimensional constraints to fix a ship's posi-
tion. The members of the navigation team read
scales and translate spoken representations into
written ones. The navigation system computes
distance from rate and time, while the members
of the team imagine four-digit numbers as two-
digit numbers. The computations that are per-
formed by the navigation system are a side ef-
fect of the cognitive activity of the members of
the navigation team. The tools of the trade both
 706/1540

define the tasks that are faced by the navigators

and, in their operation, actually carry out the
computations. As we have seen, the very same
computation can be implemented many ways,
each implementation placing vastly different
cognitive demands on the task performer.

I argued above that the naive notion of these

tools as amplifiers of cognitive activity was mis-
taken. Is a written procedure an amplifier of
memory? Not if the task performer never knew
the procedure. Then, and always, the functional
system that performs the task is a constellation
of structured representational media that are
brought into coordination with one another.
These tools permit us to transform difficult tasks
into ones that can be done by pattern matching,
by the manipulation of simple physical systems,
or by mental simulations of manipulations of
simple physical systems. These tools are useful
precisely because the cognitive processes re-
quired to
707/1540
 Page 171

manipulate them are not the computational pro-

cesses accomplished by their manipulation. The
computational constraints of the problem have
been built into the physical structure of the
tools.
 709/1540

The slide rule is one of the best examples of this

principle. Logarithms map multiplication and
division onto addition and subtraction. The log-
arithmic scale maps logarithmic magnitudes
onto physical space. The slide rule spatially jux-
taposes logarithmic scales and implements addi-
tion and subtraction of stretches of space that
represent logarithmic magnitudes. In this way,
multiplication and division are implemented as
simple additions and subtractions of spatial dis-
placements. The tasks facing the tool user are in
the domain of scale-alignment operations, but
the computations achieved are in the domain of
mathematics. The chart is a slightly subtler ex-
ample of the same principle. Consider the
preplotting of danger bearings. Once this has
been done, the determination of whether the
ship is standing into danger is made by simply
seeing on which side of the line the position of
the ship lies. In this case, a conceptual
 710/1540

judgement is implemented as a simple perceptu-

al inference.

These tools thus implement computation as

simple manipulation of physical objects and im-
plement conceptual judgements as perceptual in-
ferences. But perhaps this refinement will be
lacking from the next generation of tools. By
failing to understand the source of the computa-
tional power in our interactions with simple "un-
intelligent" physical devices, we position
ourselves well to squander opportunities with
so-called intelligent computers. The synergy of
psychology and artificial intelligence may lead
us to attempt to create more and more intelligent
artificial agents rather than more powerful task-
transforming representations.
 711/1540

In thinking about the slide rule, one may also re-

flect on doing multiplication by mental arith-
metic. There are many ways to do mental multi-
plication. In Western culture a person faced with
the need for mental arithmetic typically does
some version of imagining place-value arithmet-
ic. The person applying the algorithms of place-
value arithmetic to mental images of numbers is
using an internal rather than an external artifact.
These internal artifacts are cultural programs
(D'Andrade 1981, 1989). Internalizing the arti-
factual structure imposes new cognitive tasks, to
be sure, and the properties of the individual pro-
cessors put limits on the sorts of
 Page 172

artifactual structures that can be successfully

manipulated internally. Faced with a mental
multiplication problem, even those who are fa-
miliar with the operation of a slide rule don't try
to imagine how they would manipulate a slide
rule to solve the problem. This is because what
makes the slide rule work is the precision of the
placement of the tick marks that represent num-
bers. The medium of mental imagery is poorly
suited for preserving such precise spatial rela-
tionships, especially when one set of them must
be moved relative to another.
 713/1540

What was said above about the difference

between the processes realized by the manipula-
tion of the artifact system and the cognitive pro-
cesses required to perform the manipulation is
as true for internalized representations as it is of
external ones. Internal representations are cap-
able of transforming tasks too. And we should
expect that they would be. If the cognitive sys-
tem acquires new capabilities by combining rep-
resentations in new functional constellations,
then it is just as likely that an internal represent-
ation will give rise to a a new constellation as it
is that an external one will. Doing mental place-
value arithmetic imposes a particular set of re-
quirements on the task performer. Doing the
same problems with some other systemthe
Trachtenberg system of speed arithmetic (Cutler
and McShane 1975), for example; or the
Chinese base-17 system (Taylor 1984), which
uses no carry bitsimposes different require-
ments. In each case, the task is accomplished by
 714/1540

the operation of a functional system composed

of a number of representations that are brought
into coordination.

We are all cognitive bricoleursopportunistic as-

semblers of functional systems composed of in-
ternal and external structures. In developing this
argument I have been careful not to define a
class, such as cognitive artifacts, of designed ex-
ternal tools for thinking.
 715/1540

The problem with that view is that it makes it

difficult to see the role of internal artifacts, and
difficult to see the power of the sort of situated
seeing that is present in the Micronesian
navigator's images of the stars. The stars are not
artifacts. They are a natural rather than a human-
made phenomenon, yet they do have a structure
which, in interaction with the right kinds of in-
ternal artifacts (strategies for "seeing"), becomes
one of the most important structured representa-
tional media of the Micronesian system. The
more or less random sprinkling of stars in the
heavens is an important component of the Mi-
cronesian system. In a sky with an absolutely
 Page 173

uniform distribution of stars, navigation by the

stars would be impossible: information is differ-
ence, and there would be no differences to be
seen as informative.

If we ascribe to individual minds in isolation the

properties of systems that are actually composed
of individuals manipulating systems of cultural
artifacts, then we have attributed to individual
minds a process that they do not necessarily
have, and we have failed to ask about the pro-
cesses they actually must have in order to ma-
nipulate the artifacts. This sort of attribution is a
serious but frequently committed error.

Knowing Why Things Work

A well-known student of navigation laments the

effects of this accumulation of structure as
follows:
 717/1540

Even today, of course, since the ultimate sources

of time-keeping and position-finding are the
heavenly bodies, the sailor must look up at the
sky. But so long and so far has the chain of ex-
pertsprofessional astronomers, mathematicians,
almanac-makers, instrument-makers and so
forth-separated the ordinary man from the first-
hand observation that he has ceased to think
beyond the actual clock, time-signal, map calen-
dar, or whatever it may be that "tells" him what
he wishes to know. (Taylor 1971)

Frake (1985: 268) makes a similar point about

modern knowledge of the tides:

[Modern tidal theory] is far beyond the reach of

the modern navigator. Sailors today have no
need to understand tidal theory at any level.
They merely consult their tide tables anew for
each voyage.
 718/1540

The Mercator-projection chart is a specialized

analog computer, and the properties of the chart
that make its use possible are profoundly math-
ematical in nature. But those parts of the compu-
tation were performed by cartographers and
need not be a direct concern of the chart's users.
The cartographer has already done part of the
computation for every navigator who uses his
chart. The computation has been distributed
over time as well as across social space. The
navigator doesn't have to know how the chart
was made and doesn't need to know about the
properties of the Mercator projection that give
special computational meaning to straight lines.
The
 Page 174

device is actually more powerful if the user does

not have to know how or why it works, because
it is thereby available to a much larger com-
munity of users. The computational abilities of
the mind of the navigator penetrate only the
shallows of the computational problems of nav-
igation. In the day-to-day practice of navigation,
the deeper problems are either transformed by
some representational artifice into shallow ones
or not addressed at all.
 Page 175

4 The Organization of Team

Performances
Having presented an account of the performance
of the component tasks of the fix cycle in
chapter 3, here I will address the ways in which
those component tasks can be coordinated to
form a larger computational system. In Sea and
Anchor Detail, this requires getting the activities
of a number of team members into coordination.
Thus, in this chapter I consider not only how the
tools are used but also how the members of the
navigation team use the tools together. The unit
of cognitive analysis in this chapter will be the
navigation team rather than the individual
watchstander.
 721/1540

In anthropology there is scarcely a more import-

ant concept than the division of labor. In terms
of the energy budget of a human group and the
efficiency with which a group exploits its phys-
ical environment, social organizational factors
often produce group properties that differ con-
siderably from the properties of individuals. For
example, Karl Wittfogel (1957, cited in Roberts
1964), writing about the advent of hydraulic
farming and Oriental despotism, says:

A large quantity of water can be channeled and

kept within bounds only by the use of mass
labor; and this mass labor must be coordinated,
disciplined, and led. Thus a number of farmers
eager to conquer arid lowlands and plains are
forced to invoke the organizational devices
whichon the basis of premachine technologyof-
fer the one chance of success; they must work in
cooperation with theirfellows and subordinate
themselves to a directing authority.
 722/1540

Thus, a particular kind of social organization

permits individuals to combine their efforts in
ways that produce results (in this case a techno-
logical system called hydraulic farming), that
could not be produced by any individual farmer
working alone. This kind of effect is ubiquitous
in modern life, but it is largely invisible. The
skeptical reader may wish to look around right
now and see whether there is anything in the
current environment that was not either pro-
duced or delivered to its present location by the
cooperative efforts of individuals working in so-
cially organized groups.
 Page 176

The only thing I can find in my environment

that meets this test is a striped pebble that I
found at the beach and carried home to decorate
my desk. Of course, the very idea of bringing
home a pretty pebble to decorate a desk is itself
a cultural rather than a personal invention.
Every other thing I can see from my chair not
only is the product of coordinated group rather
than individual activity, but is necessarily the
product of group rather than individual activity.
 724/1540

All divisions of labor, whether the labor is phys-

ical or cognitive in nature, require distributed
cognition in order to coordinate the activities of
the participants. Even a simple system of two
men driving a spike with hammers requires
some cognition on the part of each to coordinate
his own activities with those of the other. When
the labor that is distributed is cognitive labor,
the system involves the distribution of two kinds
of cognitive labor: the cognition that is the task
and the cognition that governs the coordination
of the elements of the task. In such a case, the
group performing the cognitive task may have
cognitive properties that differ from the cognit-
ive properties of any individual.
 725/1540

In view of the importance of social organization

and the division of labor as transformers of hu-
man capacities, it is something of a surprise that
the division of cognitive labor has played such a
very minor role in cognitive anthropology.
There have been few investigations of the many
ways in which the cognitive properties of human
groups may depend on the social organization of
individual cognitive capabilities. Over the years
there has been some interest in the way that the
knowledge of a society is distributed across its
members. Schwartz's (1978) ''distributional
model of culture" was one of the best worked
out of such approaches. In recent years there has
been increasing interest in intracultural variabil-
ity, the question of the distribution of know-
ledge within a society (Romney, Weller, and
Batchelder 1986; Boster 1985, 1990). For the
most part, this recent work has addressed the
question of the reliability and representativeness
of individual anthropological informants and has
 726/1540

not been oriented toward the question of the

properties of the group that result from one or
another distribution of knowledge among its
members.

The notion that a culture or a society, as a

group, might have some cognitive properties
differing from those of the individual members
of the culture has been around since the turn of
the century, most conspicuously in the writings
of the French sociologist Emile Durkheim and
his followers and largely in the form of pro-
Page 177
 728/1540

grammatic assertions that it is true. This is an in-

teresting general assertion, but can it be demon-
strated that any particular sort of cognitive prop-
erty could be manifested differently at the indi-
vidual and group levels? Making a move in that
direction, Roberts (1964) suggested that a cul-
tural group can be seen as a kind of widely dis-
tributed memory. Such a memory is clearly
more robust than the memory of any individual
and undoubtedly has a much greater capacity
than any individual memory has. Roberts even
speculated on how retrieval from the cultural
memory might be different from individual
memory retrieval and how a variety of social or-
ganizational devices might be required for the
continued support of memory retrieval functions
in increasingly complex cultures. Roberts ex-
plored these issues in a comparison of four
American Indian tribes, holding that information
retrieval (what Roberts called scanning) at the
tribal level among the Mandan was more
 729/1540

efficient than among the Chiricahua because

"the small geographical area occupied by the
tribe, the concentrated settlement pattern, the
frequent visiting, the ceremonial linkages, made
even informal mechanisms (of retrieval) more
efficient" (ibid.: 448). Roberts also noted that
the tribal-level information-retrieval processes
of the Cheyenne had properties that were differ-
ent from those of the Mandan or Chiricahua. He
linked the properties to particular features of so-
cial organization: "If the membership of a coun-
cil represents kin and other interest groups in the
tribe, each member makes available to the coun-
cil as a whole the informational resources of the
groups he represents.... Councils have usually
been viewed as decisionmaking bodies without
proper emphasis on their function as informa-
tion retrieval units." (ibid.: 449)
 730/1540

In the sentences cited above, Roberts attributes

the differences in retrieval efficiency at the
group level to the size of the group, the pattern
of interactions among individuals, the pattern of
interaction through time, and the distribution of
knowledge. Thus, it seems important to come to
an understanding of the ways in which the cog-
nitive properties of groups may differ from
those of individuals. In the comparison of the
physical accomplishments of pre- and post-hy-
draulic agriculture societies it is obvious that the
differences in physical accomplishment are due
to differences in the social organization of phys-
ical labor rather than to differences in the phys-
ical strength of the members of the two societ-
ies. Similarly, if groups can have cognitive
properties that are significantly different from
those of the individuals in them, then differ-
ences in the
 Page 178

cognitive accomplishments of any two groups

might depend entirely on differences in the so-
cial organization of distributed cognition and
not at all on differences in the cognitive proper-
ties of individuals in the two groups. This theme
is the topic of the next two chapters.

Sea and Anchor Detail

What role does social organization play in the

cognition of the navigation team during Sea and
Anchor Detail? In chapter 1 we saw that the
ship's documents specify a normative division of
labor for this task. The specified roles were lis-
ted as follows:

3. The Sea and Anchor Piloting Detail will con-

82866990447203f57400372e62d765b5.gif
sist of:

a. The Navigator
82866990447203f57400372e62d765b5.gif
 732/1540

b. The Assistant to the Navigator

82866990447203f57400372e62d765b5.gif

c. Navigation Plotter
82866990447203f57400372e62d765b5.gif

d. Navigation Bearing Recorder/Timer

82866990447203f57400372e62d765b5.gif

e. Starboard Pelorus Operator

82866990447203f57400372e62d765b5.gif

f. Port Pelorus Operator

82866990447203f57400372e62d765b5.gif

g. Restricted Maneuvering Helmsman

82866990447203f57400372e62d765b5.gif

h. Quartermaster of the Watch

82866990447203f57400372e62d765b5.gif

i. Restricted Maneuvering Helmsman in

82866990447203f57400372e62d765b5.gif
After Steering

j. Fathometer Operator
82866990447203f57400372e62d765b5.gif

(When sufficient Quartermasters are avail-

able, each of the positions except Navigator,
82866990447203f57400372e62d765b5.gif
will be filled by a Quartermaster.)
 733/1540

The procedures to be followed and the duties of

each member of the navigation team are also
given in the watch standing manual. In consider-
ing these procedures and the division of labor
they imply, it will become apparent that the
written procedures are not used by the members
of the navigation team as structuring resources
during the performance of the task, nor do they
describe the actual tasks performed. Further-
more, if a system was actually constructed to
perform as specified in the procedures as writ-
ten, it would not work. Still, the normative pro-
cedures are a good starting point and provide a
stable framework within which the properties of
the system can be described. In the following
paragraphs, the elements of the task as specified
in the Watch Standing Procedures are inter-
spersed with text discussing the roles of the pro-
cedural elements in the constitution of the nav-
igation team as a cognitive system.
 734/1540

4. While operating in Restricted Waters, the fol-

82866990447203f57400372e62d765b5.gif
lowing procedures will be adhered to:

a. Fixes will be taken at least every three

minutes (Periodicity may be increased by
82866990447203f57400372e62d765b5.gif
the Navigator)
 Page 179

The default fix interval is 3 minutes because this

permits the simplification of certain computa-
tions. This interval can be made shorter by the
navigator if more resolution is required. The fix
interval is a parameter that controls the rate of
sampling the environment.

b. A fix will be obtained immediately fol-

82866990447203f57400372e62d765b5.gif
lowing each turn
 736/1540

Because of the nature of the position-fixing and

position-projecting computations, a ship's course
will be made to approximate a series of straight
segments punctuated by turns. This is entirely a
consequence of the way courses are steered and
positions are computed. With a different compu-
tational technology (the satellitebased Global
Positioning System, for example), it would be
possible to have ship's track consist of smooth
curves. There are two problems with fixes taken
while the ship is turning. The first is that it is
difficult to make accurate observations from a
turning platform. Even though the true bearing
of the landmark may change little while the
pelorus operator is aiming at it, if the ship is
turning then the relative bearing of the landmark
will be changing at whatever the rate of turn is.
This may leave the pelorus operator "chasing"
the landmark with the telescopic sight of the al-
idade. The second problem with fixes taken
while the ship is turning is that even if they can
 737/1540

be made accurately, they are a poor basis for the

projection of the ship's position in the future. It
is impossible to know the exact shape of the
ship's track while the ship is making the turn, so
a position fix in a turn does not permit an accur-
ate projection of where the ship will be at the
next fix time. For these reasons, fixes are not
normally taken in turns. As soon as the ship
steadies on a new course, however, it is desir-
able to take a fix from which future positions on
the new course leg can be projected as straight
lines.

c. Each set of bearings and/or ranges will be

accompanied by a sounding, which will be
82866990447203f57400372e62d765b5.gif
compared against plotted position

This element of the procedure establishes a

cross-check among the representations gener-
ated in the fix cycle. The role of comparison of
representations in error detection will be dis-
cussed in chapter 6.
 738/1540

d. The Fathometer Log will be maintained

82866990447203f57400372e62d765b5.gif
by the Fathometer operator

The quartermaster of the watch (QMOW) nor-

mally keeps all the logs. In Sea and Anchor De-
tail, however, the fathometer log is kept by the
fathometer operator. This is one of many shifts
in the
 Page 180

distribution of cognitive labor brought about by

Sea and Anchor Detail.

e. The Magnetic Compass Checkbook will

be securred on stationing the Sea and An-
chor Detail. Checking headings for each
82866990447203f57400372e62d765b5.gif
course will be entered in the Deck Log by
the QMOW

f. The Ship's Position Log will be securred

on stationing the Sea and Anchor Detail,
82866990447203f57400372e62d765b5.gif
with annotation to that effect
 740/1540

The magnetic compass checkbook is used in

Standard Steaming Watch to keep track of the
behavior of the magnetic compasses. This and
the ship's position log are secured (put away)
during Sea and Anchor Detail because the in-
formation that would normally go into them is
being generated in much more detail and recor-
ded elsewhere (in the bearing log and the deck
log).

g. If sufficient Quartermasters are available,

the Assistant to the Navigator will not tie
himself down to the plot. He will instead su-
82866990447203f57400372e62d765b5.gif
pervise the entire team with emphasis on the
plot, the recorder, and the bearing takers.
 741/1540

The Assistant to the Navigator, the Quartermas-

ter Chief, would like to be able to supervise all
the activities of the navigation team. When I
first went aboard the Palau, the team was oper-
ating in this configuration. Unfortunately, the
quartermasters were not well enough trained to
keep up with the workload, and the chief had to
step into the role of plotter. It was not even al-
ways possible to fill all the positions with
quartermasters. During some of my observation
periods at sea, sailors of the signalman rating
served as pelorus operators and fathometer oper-
ators. The effects of personnel availability on
this aspect of team composition is one of the dif-
ferences that was observed between ships. Dur-
ing Colleen Seifert's observations aboard anoth-
er ship, for example, the Assistant to the Navig-
ator had a completely supervisory role and eval-
uated the fixes as they were produced. This ele-
ment of the procedure concerns the distribution
of access to information in the navigation team
 742/1540

and can be seen as a specification of one aspect

of the computational architecture of the naviga-
tion team.

h. Periodically, every third or fourth fix, the

information passed from secondary plot in
82866990447203f57400372e62d765b5.gif
CIC will be plotted on the Primary plot for
comparison purposes.

There is a clear tradeoff here between the costs

of constructing the redundant plot and the bene-
fits of increased error detection that it provides.
In view of the nature of the representations used,
the information from Combat Information
Center (CIC) cannot be passed to the bridge in
the form of plotted positions. Rather, it is
Page 181
 744/1540

passed in a format that requires additional pro-

cessing by the bridge team. The information
could be passed as raw data (bearings and/or
ranges of landmarks), as latitude and longitude
coordinates, or in terms that locate the ship rel-
ative to the precomputed track. The latter format
is most frequently observed. "Combat holds us
30 yards right of track, 600 yards to the turn" is
a typical CIC status report. Given such a report,
the plotter might "eyeball" the position and say,
"I'll buy that.'' This apparently offhand comment
represents the outcome of a computation. The
coordinates passed by CIC fit the structure that
is available to the plotter on the chart. Remem-
ber that the distances to the turns are marked in
hundreds of yards. Locating a point on the chart
that represents a position 30 yards right of the
planned track and 600 yards prior to the next
turn is therefore relatively easy. Economical en-
coding of position in relation to the planned
track is possible only if both the bridge
 745/1540

navigation team and the Combat Information

Center have the same track plotted on their
charts. The report is about the position of the
ship, but it assumes shared representations of
the framework with respect to which the posi-
tion is reported.

This redundant processing by the plotter

provides another opportunity for the detection of
error through the comparison of independently
computed representations. The navigation pro-
cess generates many representations from
sources of data that are reasonably independent.
The positions plotted in the CIC, for example,
are based on radar returns rather than on visual
bearings. The comparison of such representa-
tions is a very general theme in the practice of
navigation. The measures listed here are simply
specifications for Sea and Anchor Detail of pro-
cedural strategies that are followed in all naviga-
tion. In Standard Steaming Watch for example,
the following instruction holds:
 746/1540

h. During prolonged periods in which no nav

aids are available (1 hour or more), both the
DRAI [Dead Reckoning Analyzer Instru-
ment] and Nav Sat DRs will be recorded in
the Ship's Position Log, and plotted as es-
82866990447203f57400372e62d765b5.gif
timated positions for comparison against the
hand DR. If unexplainable differences de-
velop, the Assistant to the Navigator will be
called immediately.

We see here again the emphasis on the compar-

ison and correlation of representations from dif-
ferent sources. The chart is the "common
ground" on which all of these representations
can be compared.

i. The Navigator will act as overall coordin-

ator of the Bridge Party during Sea and An-
82866990447203f57400372e62d765b5.gif
chor Detail.
 Page 182

This is another element in the organization of

the computational architecture of the navigation
team. The navigator is given the authority to re-
configure the navigation team as he sees fit.

j. If the ship goes into a condition of Re-

duced Visibility during Sea and Anchor De-
tail, the senior Quartermaster will man the
82866990447203f57400372e62d765b5.gif
LN-66 Radar on a time sharing basis with
the OOD.
 748/1540

The ship never went into a condition of reduced

visibility while I was aboard. As reported in
chapter 1, the assistant to the navigator claimed
that he would not abide by the procedural spe-
cification of the relationship between the CIC
and the bridge in reduced-visibility situations. It
must be remembered that there is a surface-
search radar unit on the bridge that can be used
for observing radar bearings and ranges of land-
marks (the LN-66 mentioned above). The
Assistant to the Navigator should attempt to
generate his own navigation data using that
device. However, the competition for the use of
the radar between the navigation team and the
officer of the deck is likely to be intense in such
a setting. In reduced-visibility situations, it
might be impossible to even see past the edge of
the flight deck. The officer of the deck will want
to adjust the radar to be most effective in detect-
ing and tracking other ships. The quartermaster
using the radar will want to adjust it to measure
 749/1540

the bearings and ranges of landmarks. These

two uses conflict with each other.

The procedures given above describe the pro-

cedures and the division of labor mandated for
Sea and Anchor Detail. These have a variety of
cognitive consequences at the system level, in-
cluding changes in the organization of the per-
ceptual apparatus of the system to meet anticip-
ated changes in environmental conditions,
robust error-detection procedures grounded in
the comparisons of multiple representations of
the same situation, increase in work capacity
provided by distributing cognitive labor across
social space, and self-reflection provided by su-
pervisory functions.

The duties of the individual members of the Sea

and Anchor Detail team are further specified as
follows:

5. The Navigation Plotter will:

82866990447203f57400372e62d765b5.gif
 750/1540

a Plot each fix.

82866990447203f57400372e62d765b5.gif

b. Plot periodic fixes from CIC

82866990447203f57400372e62d765b5.gif

c. Maintain a constant DR ahead for a min-

82866990447203f57400372e62d765b5.gif
imum of two fix intervals.

d. Provide the following information to the

82866990447203f57400372e62d765b5.gif
Navigator/OOD

(1) Present position with respect

82866990447203f57400372e62d765b5.gif to track

(2) Present SOG (Speed over the

82866990447203f57400372e62d765b5.gif
ground.)

(3) Distance to the next turn, and time at

82866990447203f57400372e62d765b5.gif
present SOG
 Page 183

(4) Turn Bearing for the next turn

82866990447203f57400372e62d765b5.gif

(5) Set and Drift when determined (ap-

82866990447203f57400372e62d765b5.gif
proximately every third fix)

(6) Nearest shoal water forward of the

82866990447203f57400372e62d765b5.gif
beam

(7) If anchoring, distance and bearing to

the drop point, slow point, stop point, or
82866990447203f57400372e62d765b5.gif
back point, whichever is next.
 752/1540

The first three duties of the restricted-maneuver-

ing plotter are straightforward. The items of in-
formation listed in paragraph d above are not ac-
tually reported on every fix, or even on the
stated intervals where specified. Rather, they are
provided when they are thought to be of use to
the OOD. This requires the plotter to know
something about the nature of the work being
done by the OOD, so that he can anticipate the
OOD's information needs and provide the right
information at the right time. As noted in
chapter 3, the determination of the relation of
the ship to the intended track is greatly simpli-
fied by the precomputation of the track.
 753/1540

The ship's speed over the ground may be very

different from its speed through the water be-
cause the water itself may be moving. The speed
over the ground is of concern to the officer of
the deck, because it is the rate at which the ship
is moving relative to land. The motion of the
ship over the ground is the vector sum of the
motion of the ship through the water and the
motion of the water with respect to land. In har-
bors, tidal effects may produce very strong cur-
rents that can augment or diminish the speed of
the ship over the ground. (Racing yacht tacti-
cians on San Francisco Bay sometimes joke that
the anchor is the fastest piece of equipment on
the boat. In truly adverse current conditions, a
boat at anchor with zero speed over the ground
may be doing better than a boat sailing fast
against a current so strong that its speed over the
ground is negative or in the wrong direction.)
The direction of the movement of the water over
the ground is called the set, and the speed of the
 754/1540

water over the ground is called the drift. This is

useful information for the officer of the deck be-
cause it affects both the speed over the ground
and the handling characteristics of the ship.

6. The navigation Recorder/Timer will:

82866990447203f57400372e62d765b5.gif

a. Time each fix to three minutes, or the

82866990447203f57400372e62d765b5.gif
Navigator's instructions

b. Keep the Bearing Record Book in accord-

82866990447203f57400372e62d765b5.gif
ance with the instructions posted therein

c. Inform the Pelorus Operators of nav aids

82866990447203f57400372e62d765b5.gif
to be used

d. Speak out continuous bearings when

82866990447203f57400372e62d765b5.gif
ordered to do so

e. Obtain soundings from the Fathometer

82866990447203f57400372e62d765b5.gif
Operator

f. Record data for at least three LOPs

82866990447203f57400372e62d765b5.gif
 755/1540

The navigation recorder/timer provides temporal

and informational coordination among the other
elements of the navigation
 Page 184

team. His timing signals and instructions on the

navigation aids to be used control the behavior
of the pelorus operators. His entries in the bear-
ing record log are the system's first permanent
representation of its relationship to a landmark.
The structure of the bearing record log in stand-
ard form (OpNav Form 3530/2) is a resource for
the organization of action. Its columns and rows
are preprinted with labels. Entries must be made
in ink, and no erasures are permitted: "If an er-
ror is made, the recorder must draw a line
through the entry that is in error and enter the
correct information leaving both entries legible."
(Maloney 1985)

7. The Restricted Maneuvering Fathometer

82866990447203f57400372e62d765b5.gif
Operator will:

a. Take soundings and send them to the

82866990447203f57400372e62d765b5.gif
bridge on request
 757/1540

b. Record the time and sounding every time

82866990447203f57400372e62d765b5.gif
a sounding is sent to the bridge.

The fathometer operator makes a redundant re-

cording of the soundings in the sounding log.

8. The Bridge Wing Pelorus Operators shall:

82866990447203f57400372e62d765b5.gif

a. Acquaint themselves with all available in-

formation on the nav aids to be utilized prior
82866990447203f57400372e62d765b5.gif
to the Sea and Anchor Detail

b. Clothe themselves for comfort.

82866990447203f57400372e62d765b5.gif

c. Checkout the operation of their Alidade as

soon as they reach the Bridge, reporting any
82866990447203f57400372e62d765b5.gif
discrepancy immediately to the Leading
Quartermaster.

d. Maintain sound powered phone commu-

82866990447203f57400372e62d765b5.gif
nications with the Recorder.
 758/1540

e. Take and report bearings to the objects

ordered by the Recorder and when ordered
82866990447203f57400372e62d765b5.gif
by the Recorder.
 759/1540

The pelorus operators must acquaint themselves

with the navigation aids so that they will be able
to find them when directed to shoot bearings to
them. Aboard some ships, the aids are given let-
ter identifiers and are referred to over the
phones in that way. This lettering scheme is an
example of a feature that benefits one part of the
organization while putting costs elsewhere
(Grudin 1988). On an entry to an unfamiliar har-
bor, the landmarks may be labeled in alphabetic-
al order from the harbor entrance to the pier.
This simplifies the work at the plotting table be-
cause it imposes a coherent ordering on the
landmarks. It makes the work of the pelorus op-
erators much more difficult, however, because
they must master a set of arbitrary names for the
landmarks. Some quartermasters have remarked
that this can be a real problem going into a for-
eign port.
 760/1540

Once on duty, the pelorus operators are expec-

ted to stay at their posts for the duration of Sea
and Anchor Detail. Since the peloruses are loc-
ated outside the skin of the ship, it is important
that pelorus operators dress appropriately from
the beginning so that they do
 Page 185

not become uncomfortable while exposed to the

elements. These prescriptions for the pelorus op-
erators cover several aspects of their contribu-
tion to the piloting system. They are required to
prepare to do the job, avoid an anticipated fail-
ure due to discomfort, test sensors with enough
time to make repairs, maintain connection to the
rest of the system, and operate as instructed.

9. The QMOW shall:

82866990447203f57400372e62d765b5.gif

a. Maintain the deck log

82866990447203f57400372e62d765b5.gif

b. Maintain the Gyro Behavior Log

82866990447203f57400372e62d765b5.gif

c. Maintain a copy of the Pac Fleet Organiz-

82866990447203f57400372e62d765b5.gif
ation Manual

d. Maintain a copy of the Rules of the Road

82866990447203f57400372e62d765b5.gif
for immediate reference.
 762/1540

This is a contraction of the normal duties of the

QMOW. The deck log and the gyro behavior log
are repositories of memories.

Paragraphs 5-9 of the Watch Standing Proced-

ures lay out the allocation of jobs to the mem-
bers of the team and the interlocking system of
functions they perform. Since the work of the
team is a computation, we can treat this as a
computational system and treat the social organ-
ization of the team as a computational
architecture.

Social Organization as Computational

Architecture
 763/1540

In a paper titled "Natural and Social System

Metaphors for Distributed Problem Solving,"
Chandresekaran (1981) discussed properties of
distributed problem-solving systems.
Chandresekaran took social systems as a base
domain for the metaphorical organization of dis-
tributed computer systems. Of course, the com-
putational properties of the computer systems
that are built on the social metaphors may also
be computational properties of the social sys-
tems themselves. Thus, although it is not cus-
tomary to speak of the computational properties
of social institutions, the navigation team in Sea
and Anchor Detail can be seen as a computa-
tional machine. In this section I explore this
metaphor, looking at the ways in which aspects
of the behavior of the system can be interpreted
in a computational framework. This seems to
me a much more solidly grounded application of
the computational metaphor to a cognitive sys-
tem than the application of this metaphor to the
 764/1540

workings of an individual mind. See chapters 7

and 9 for further discussion.

When computational tasks are socially distrib-

uted, there are two layers of organization to the
activity: the computational organization, as
defined by the computational dependencies
among the
 Page 186

various parts of the computation, and the social

organization, which structures the interactions
among the participants to the computation.

Activity Score

In order to examine the properties of the per-

formance of the navigation team, it is useful to
have a representation of the activity that makes
clear the relations among the activities of the
various members of the team.
 766/1540

Figure 4.1 is an activity score for a typical posi-

tion fix. The purpose of the activity score is to
show the temporal pattern of activity across the
representational media that are involved in the
fix cycle. Along the left axis of the figure are the
names of the media, the sensors at the bottom
and the "higher-level" processing media (such as
the chart) at the top. Across the bottom is a time
scale marked off in 2-second intervals. Each fill
pattern in the score denotes the activities that in-
volve the coordination of representations of a
single landmark bearing. The first event shown,
at the extreme left of the diagram, is the "stand
by to mark" signal that brought the bearing re-
corder and the two pelorus operators into
coordination.
 767/1540

As indicated in the Watch Standing Procedures,

the pelorus operators should aim their alidades
at their landmarks when they receive the "stand
by to mark" signal from the recorder. This is in-
dicated in the activity score by the regions that
show simultaneous coordination of each pelorus
operator with his alidade and an element of the
world of landmarks (beginning at time 59). In
this case, the starboard pelorus operator had the
landmark called Dive Tower and the port pelor-
us operator had Point Loma. Immediately before
giving the "mark" signal, the recorder and the
plotter were discussing which landmarks to use
on this fix. This caused the recorder to be late
giving the "stand by to mark" signal. The inter-
val between the "stand by to mark" and "mark"
signals was less than 3 seconds, rather than the
usual 10 seconds. The team is on a 2minute fix
interval at this point. This is one of the reasons
that the recorder rushed the mark signal. The fix
is accompanied by a sounding that is provided
 768/1540

by the fathometer operator via the phone circuit

just after the "mark" signal. The fathometer op-
erator, who expects to have a 10-second window
between the "stand by to mark" and "mark" sig-
nals in which to report the depth, began to read
the depth while the recorder was giving the
"mark" signal. He
770/1540
771/1540
 772/1540

Figure 4.1 An activity score for a fix cycl

among coordinated activities o
 Page 188

deferred to the recorder and then repeated the

depth report immediately after the mark signal.

The "mark" signal also seemed to take the pelor-

us operators by surprise. They were probably
still locating their landmarks and aiming their
alidades when it arrived. About 3½ seconds
after the "mark" signal, the starboard pelorus op-
erator asked "Did you say 'mark'?" The recorder
answered by simply repeating the "mark" signal
(at time 06). Any additional explanation offered
by the recorder at this point would only have
delayed the fix even more. The starboard pelor-
us operator read the bearing of the Dive Tower
and reported it (at time 08).
 774/1540

While recording the bearing in the log, the bear-

ing recorder read the bearing aloud for the plot-
ter to hear (at time 10). Meanwhile, the port
pelorus operator aligned Point Loma in the
sights of his alidade and reported the bearing of
Point Loma just as the bearing recorder was
reading the Dive Tower bearing to the plotter (at
time 11).

Upon hearing the bearing from the recorder, the

plotter said "OK" and aligned the scale of the
hoey to reproduce the bearing (at time 11-13).
He then aligned the hoey with the chart and
plotted the line of position for Dive Tower (time
13-18).
 775/1540

Just after the plotter applied the hoey to the

chart to plot the LOP for Dive Tower, the re-
corder read aloud the bearing to Point Loma.
Since the plotter had already aligned and locked
the hoey, he was no longer dealing with the
bearing to Dive Tower as a number. Hearing the
spoken bearing to Point Loma probably in-
terfered little with the task of getting the aligned
hoey into coordination with the directional
frame of the chart. If the plotter had still been
aligning the hoey scale when the new bearing
was spoken, we might have expected some de-
structive interference between the two tasks.
 776/1540

While the bearing recorder was repeating the

bearing to Point Loma, the starboard pelorus op-
erator reported the bearing to Hotel del Coron-
ado ("Hotel del"). Again, the overlap between
the speaking and listening tasks did not cause
destructive interference. The pelorus operator
pronounced the name of the new landmark
while the recorder was speaking the name of the
previous landmark. The numbers did not
overlap.

When the plotter had finished plotting the LOP

for Dive Tower, he began scaling the hoey for
Point Loma. He may have been able to attend at
least partially to the spoken report of the bearing
while he was placing the hoey on the chart a few
seconds earlier. However,
 Page 189

after fiddling with the hoey for 2 seconds he

looked into the bearing log and read the bearing
of Point Loma (time 21-24). He then returned to
the scaling task (time 24-26) and then applied
the hoey to the chart and plotted the LOP (time
26-35).

Meanwhile, the bearing recorder was instructing

the starboard pelorus operator on a change of
landmark that the recorder and plotter had de-
cided upon in the seconds leading up to the cur-
rent fix.
 778/1540

The bearing recorder completed his instructions

to the starboard pelorus operator while the plot-
ter was still plotting the Point Loma LOP. When
the recorder saw that the plotter had finished
plotting Point Loma, he read the bearing of
Hotel del aloud from the bearing log (time 35).
The plotter had already turned to the log and
read the bearing there (time 36) before setting
the bearing into the state of the hoey (time
37-40) and plotting the LOP (time 40-44).

After plotting the third LOP, the plotter marked

and labeled the fix (time 44-47). He then went
on to extend the dead reckoning positions (time
48-60).

This example illustrates a number of interesting

properties of the team performance of the fix
cycle.

Parallel Activities
 779/1540

Perhaps the most obvious property is that the

activities of the members of the team take place
in parallel. For example, at time 11 the port
pelorus operator was reporting the bearing of
Point Loma to the recorder, who was at that mo-
ment reporting the bearing of the Dive Tower to
the plotter. At the same moment, the starboard
bearing taker was aiming his alidade at the
Hotel del landmark.
 780/1540

This is a clear example of the simultaneous co-

ordination of many media in a functional system
that transcends the boundaries of the individual
actors. In chapter 3, identifying the landmark,
aiming the alidade, reading the bearing, and re-
porting and remembering the bearing were each
described as processes in which a set of mutu-
ally constraining media are placed in coordina-
tion by the pelorus operator. In chapter 3 we
also saw how recording the bearings involved
the construction of a complex functional system
by the bearing recorder. Now we see that these
two functional systems were assembled into a
larger functional system in the coordination of
the activities of the two crew members. Here
two team members, the pelorus operator and the
recorder, worked together on a single problem.
 Page 190

This example also demonstrates simultaneous

activity within a single individual. The bearing
recorder was reading one bearing and listening
to another at the same time. The overlap of
activity is such that there is no destructive inter-
ference between the two tasks, although if the
timing was even a few tenths of a second differ-
ent there could be. The recorder's words and the
port wing bearing taker's words overlap like
this:

Recorder: 0 6 5 Dive Tower

Port Wing: Point Loma 3 1 9.

Bottom-Up and Top-Down Processes

 782/1540

The propagation of the bearings from the alid-

ades to the chart is a "bottom-up" information
process. The representation of the relationship
of the ship to the world is transformed into sym-
bolic form and moved across a set of media until
it arrives at the chart. In an idealization of the
fix cycle, information flows bottom-up, from
sensors to central representation, in the first part
of the cycle; it flows top-down, from the central
decision makers to the sensors, in the latter part
of the cycle, when the pelorus operators are in-
structed to shoot particular landmarks. The gen-
eral trend is apparent in figure 4.1 in the form of
the upward slope of activity regions from left to
right. The top-down activities in the fix cycle
are more diverse. As soon as the bearings had
been reported, the recorder instructed the star-
board pelorus operator to shift to a new
landmark.
 783/1540

The "stand by to mark" and "mark" signals are

also top-down messages. This example illus-
trates some of the potential complexities of the
flow of information in the system. Because of a
disruption of the recorder's activities, the expec-
ted mark signal came at an unexpected time.

Other top-down messages guide the sensors to

targets in the world. For example, the recorder
instructed the pelorus operator: "Shoot the end
part that is away from you." On another occa-
sion, the plotter asked the recorder: "Give me a
quick line ahead, then back to the Aero
Beacon."

Some top-down messages request information

about sensor capability. Before a fix, the record-
er asked a pelorus operator "Can you still see
Bravo pier?" During a fix, when a pelorus oper-
ator failed to report, the recorder asked: "Are
you still there? What happened, man?"
784/1540
Page 191
 786/1540

Top-down messages also give the sensors feed-

back on their performance. When the LOPs yiel-
ded a tight triangle on the chart, the recorder
told the pelorus operators ''Excellent fix, guys."
When the bearing reports were coming in too
slowly, he chided "Let's pick up the marking,
man." The top-down signals observed in the op-
eration of the navigation team in Sea and An-
chor Detail also included recalibrations of the
senses. The plotter instructed the recorder as fol-
lows: "The fixes are getting open; tell them to
mark their heads." To mark heads means to have
the two pelorus operators simultaneously report
the true heading of the ship with respect to the
gyroscope. Any difference between the reported
"heads" is an indication of a failure of the gyro-
repeater system. This comment from the plotter
is simultaneously a complaint about the quality
of the information received, an indication of a
hypothesis concerning the source of the degrad-
ation of the data (that the repeaters are not
 787/1540

aligned), and an instruction to perform a proced-

ure that will provide a test of the hypothesis.

Human Interfaces

The creation of human and organizational inter-

faces to tasks is ubiquitous. Roy D'Andrade
(personal communication) has pointed out that
in the academic world we appoint discussion
leaders to act as interfaces for the rest of us to
some particular reading. The discussion leader
in the meeting has a rank, a responsibility, and
certain privileges that are bestowed by the rest
of the group. Aboard a ship, the quartermaster
chief has the authority to make one of his men
an interface to a particular task. From a func-
tional perspective, the navigation team is the
conning officer's interface to the navigation
problem. The team provides mediation of such a
complex form that it is barely recognizable as
mediation.
 788/1540

DAEMONS

A commonly created sort of interface to a task is

what in computer science is called a daemon. A
daemon is an agent that monitors a world wait-
ing for certain specified conditions. When the
trigger conditions exist, the daemon takes a spe-
cified action.

Setting a depth threshold detector

During an approach to an antenna-calibration

buoy near the shore, Chief Richards assigned
Smith to the fathometer with instructions to re-
port when the depth of water under the ship
shoaled to less
 Page 192

than 20 fathoms. This is an example of the so-

cial construction of an information-processing
mechanism. In this case, the chief had recon-
figured the navigation team to create within it a
daemon to detect a particular condition. The re-
configuration involved the construction of a
short strand of representational state. Smith was
stationed at the fathometer, concentrating on the
relationship of the marks indicating the depth of
water to the labeled 20-fathom line on the echo-
sounder graph paper. His job was to detect a
certain analog relationship (depth indication
above the 20-fathom line) and transform that to
a symbolic signal to the plotter.

Continuous bearings
 790/1540

Continuous bearing reporting is a nice case of

setting up a somewhat more complicated
information-processing structure that detects a
single very specific condition. On the open sea,
turns are made at specified times or when the
ship is reckoned to have reached a specified po-
sition. In restricted waters more precision is re-
quired. For this purpose, turn bearings are con-
structed (see chapter 3).
 791/1540

When the ship is approaching the turn bearing,

the plotter will ask the recorder to have the
pelorus operator on the appropriate side observe
the landmark on which the turn bearing is based
and give continuous readings of its bearing.
These are not recorded in the bearing record log,
but are relayed verbally to the plotter. By align-
ing the plotting tool with the landmark and the
spoken bearings, the plotter can move the rep-
resented position of the ship along the course
line in a relatively continuous fashion. Since the
track is marked in 100-yard increments, it is
then easy for the plotter to determine and call
out the distance to the next turn.
 792/1540

An excerpt from the transcript of the moments

leading up to a turn looks as follows. (The turn
bearing is 192° on North Island Tower. The CIC
COMM is a phone talker who relays informa-
tion between the chart table and the plotting
table in CIC. See figure 4.2 for the position of
the ship during this event.)

Plotter: OK. What course is he on? OK, how

about continuous bearings...

Recorder: Continuous bearings on North Island

Tower.

Plotter: ... on the Tower.

Recorder: 229...228...227...226...225...224...

Plotter: Five hundred yards to the turn. Next

course will be 2 5 1.
Figure 4.2 Moving the hoey arm in coordination with continu
reports. The turn bearing for this turn is 192° to North Island
 794/1540

plotter, having aligned the hoey with the chart, holds the base
moves the arm of the hoey to each of the bearings as it is rep
permits him to read the distance to the turn directly off th

CICComm: Navigation holds 500 yards to the turn.

OOD: Very well.

Plotter: Ah ... see if I try 8 knots, 8 and a half mayb

CIC Comm: Combat holds 420 yards to the turn.

Recorder: 221...220...219... 218...21 7...216...215...

Plotter: Three hundred yards to the turn. Next cours

Recorder: 213

CICComm: Navigation holds 300 yards.

Recorder: 212...211...210...2 09...208...

Plotter: Two hundred yards to the turn.

Recorder: 207...206...205...204...203...202...201
 795/1540

Plotter: One hundred yards to the turn.

 Page 194

CIC Comm: Navigation holds 100 yards to the

turn.

Recorder: 2 0 0 ...

Plotter: One hundred yards to the turn.

Recorder: 199...198...197... 196...195...194...193

Plotter: Recommend coming left 2 5 1.

Recorder: 192...

OOD: Left 15 degrees rudder. Steer course 2 5

1.

CICComm: Left 15 degrees rudder, change

course 2 5 1.

Helm: Change course 2 5 1, aye sir. Left 15 de-

grees. Course 2 5 1.

OOD: Very well.

 797/1540

The recorder also had access to the chart and

knew which landmark the bearings would be
taken on before the plotter had finished telling
him to begin continuous bearings. Every bearing
spoken out by the recorder was spoken out just
the instant before by the pelorus operator. The
trajectory of representational state in this case
flows without interruption from the relationship
between the ship and the world to the state of
the pelorus to the bearing spoken by the pelorus
operator to the bearing spoken by the recorder to
the state of the hoey on the chart to the advice
announced by the plotter for the OOD to begin
the turn. In making a turn in Sea and Anchor
Detail, the system hooks up four people and a
suite of technology into a tightly coupled func-
tional system. It is a temporary structure that
brings media and processes into coordination in
order to track the ship's relationship to its envir-
onment on a finer time scale than in normal op-
erations. This entire functional system is
 798/1540

organized around the detection of a single con-

dition: the arrival of the ship at the turn point.

BUFFERS

The bearing recorder and bearing record log are

information buffers. They enable the pelorus op-
erators, whose job is to make the observations as
nearly simultaneously as is possible, and the
plotter, whose job is to get the lines of position
onto the chart, to operate asynchronously. There
is a great deal of variation in the pace of the
work done by the members of the navigation
team. The buffering activity of the bearing re-
corder introduces slack into the system so that
the temporal constraints of the pelorus operators
do not interfere with the temporal constraints of
the plotter. The bearing re-
 Page 195

cord log is also a special kind of filter that

passes the bearings without passing the temporal
characteristics of their production. In this way, it
inhibits the propagation of some kinds of repres-
entational state. Without this buffering, the re-
ports of the pelorus operators might interfere
with the plotting activity of the plotter, or data
might be lost because both sender and recipient
were unable to attend to a message at the same
time.
 800/1540

The bridge team is connected to other parts of

the ship by soundpowered phone circuits like
the one used by the pelorus operators and the
bearing recorder. These lines provide the bridge
team with communication links to the foc'sle, to
after-steering, to the combat information center,
to the signal bridge (where lookouts are posted),
and to other locations on the ship. There is a
person called a phone talker posted at each end
of each of these phone lines. The numerous
phone talkers around the ship are also informa-
tion buffers. Each pair of them permits commu-
nication to take place when the sender and the
receiver are not overloaded. For example, rather
than simply blurt out whatever message has ar-
rived, a bridge phone talker can wait for a pause
in the OOD's work to pass a message to him.
The phone talker can hold the message until an
opportunity to insert it into the activity on the
bridge has arrived. Someone sending a message
to the bridge from another part of the ship
 801/1540

cannot know when would be an appropriate mo-

ment to interject the message. The phone talker
is a sophisticated buffer who uses his knowledge
of conversational turn taking to decide when to
forward a message.

Buffering contributes to what Perrow (1984) has

called "loose coupling" of the system. The buf-
fering prevents the uncontrolled propagation of
effects from one part of the system to another.
Buffering provides protection against destruct-
ive interference between processes running in
parallel.

Communication and Memory

 802/1540

In any implementation of the fix cycle, repres-

entational state needs to propagate physically
from the pelorus to the bearing record log and
then to the chart. In Standard Steaming Watch,
the states are propagated from the pelorus to the
memory of the watchstander and then transpor-
ted from the pelorus to the bearing record log. In
Sea and Anchor Detail, the state is also propag-
ated from the pelorus
 Page 196

to the mind of the pelorus operator, but it is then

propagated by way of communications techno-
logy to the mind of the bearing recorder. It is
then propagated to the bearing record log. Thus,
the work that is done by individual memory in
the solo condition is replaced in the group con-
dition by interpersonal communication. Perhaps
this should come as no surprise. If we think of
individual memory as communication with the
self over time (Lantz and Stefflre 1964), then
the replacement of intrapersonal communication
by interpersonal communication is an expected
consequence of the move from individual to
team performance of a task.
 804/1540

For example, the quartermaster in Standard

Steaming Watch will have to remember which
landmarks have been chosen while moving from
the chart table to the peloruses on the wings.
More challenging, the quartermaster will have to
remember or record the observations as they are
made, so that they can be recalled and recorded
in the bearing log. The chart and the bearing log
are located in the pilothouse, while the peloruses
are located on the wings. Between the time
when the first observation is made and the time
when the bearings are recorded in the bearing
record log, the quartermaster will have to make
two other observations and then return to the
chart table. Some watchstanders rely on spoken
rehearsals of the bearings to remember them. In
this case only the numbers are usually re-
hearsed, not the numbers and the names. The as-
signment of names to the numbers can be made
at the chart, and the position of the lines on the
chart can help the quartermaster remember
 805/1540

which landmark has the bearing being plotted.

The problems with this are that the subsequent
bearings may interfere with the earlier ones; that
the talk in the pilothouse itself often is filled
with unrelated numbers, so it too may interfere;
and that if the chart is being used to disambigu-
ate the assignment of landmarks to remembered
bearing numbers, the power of error checking at
the chart is sharply diminished. Other quarter-
masters jot the bearings down on a sheet of pa-
per or on one hand as each is observed. If the
landmarks are off to the port side of the ship and
the weather is cold, the quartermaster may have
to walk aft to gain access to the port pelorus via
the passageway at the back of the island because
the captain will not permit the door behind his
chair to be opened. This means that the bearing
will have to be remembered longer, which intro-
duces additional cognitive requirements. This is
an example of the way that the cognitive
 806/1540

requirements of real-world task performance

may be driven by unexpected factors.
 Page 197

Task Allocation and Equipment Layout

 808/1540

The arrangement of equipment in a workplace

might seem to be a topic for traditional, noncog-
nitive ergonomics. However, it has an interpret-
ation in terms of the construction of systems of
socially distributed cognition. The interaction of
the properties of the senses with the physical
layout of the task environment defines possibil-
ities for the distribution of access to informa-
tion. For example, the location of the fathometer
in the charthouse, away from the bridge, makes
a distribution of labor necessary in order to meet
the time requirements of Sea and Anchor Detail.
It is simply not possible for a single watchstand-
er to make the required observations in the allot-
ted time, given the physical locations of the
equipment. In fact, the computational con-
sequences of the locations of equipment may in-
teract in unexpected ways with other aspects of
the ship's operation. In Standard Steaming
Watch a single quartermaster may be respons-
ible for all navigation activities. While making
 809/1540

the bearing observations for the fix, the QMOW

must go out on the wings. The starboard pelorus
is within 10 feet of the chart table, and it is eas-
ily accessed through a nearby door. The port
pelorus is about 30 feet from the chart table, just
outside a door on the port side of the bridge. On
this particular ship, however, if the captain is on
the bridge, taking a bearing with the port pelorus
can involve an absence from the chart table of
up to a minute. The reason is that, as was men-
tioned above, the captain likes to keep the door
immediately behind his chair closed while he is
on the bridge. In order to get to the port pelorus,
the QM must go aft to a doorway at the back of
the port wing and walk forward on the wing to
reach the pelorus. Upon returning to the pilot-
house, the QM should then go to the helm and
leehelm stations to see if any changes in course
or speed have been ordered in his absence.
 810/1540

Chief Richards says it would be nice to have in-

strument repeaters at the chart table. A speed log
repeater would be especially useful. In Sea and
Anchor Detail, the QMOW should be able to
keep track of speed and heading changes by at-
tending to the commands issued by the conning
officer to the helmsman and the leehelmsman.
In practice, however, it is not always possible to
do this. Speed and heading are also available to
the QMOW on instruments, but the instruments
are not conveniently located. Their placement
requires the QMOW to leave the vicinity of the
chart table to acquire this information. The mas-
ter gyrocompass is located in the steering bin-
acle, and the speed display is forward on the
port side of the
 Page 198

bridge, in front of the captain's chair. These

facts have nontrivial consequences for the
information-processing properties of the naviga-
tion team.

Sequential Control of Action

Russian legend has it that Prince Potemkin once

organized a band in which each musician had a
horn, but each horn could only sound one note.
To play a piece, "the players had to be ex-
tremely skillful in order to preserve the syn-
chronic performance of all the instruments and
weave their own note into the melody at the
right time" (Kann 1978: 52). Playing in
Potemkin's horn band was apparently an enorm-
ously difficult coordination task. Sequential
control was achieved by having every musician
know the plan of the entire piece and also know
the place of every instance of his own note with-
in the piece.
 812/1540

A procedure is sequentially unconstrained if the

execution of any enabled operation will never
disable any other enabled but as yet unexecuted
operation. A task that has no sequential con-
straints can be accomplished by a "swarm of
ants" strategy. In such a scheme, there is no
communication between the active agents other
than their effects on a shared environment. Each
agent simply mills about taking actions only
when he encounters situations on which he can
act.

A procedure is sequentially constrained if the

execution of any enabled operation will disable
any other enabled but as yet unexecuted opera-
tion. Where there are sequential constraints, it is
necessary to have some control over the se-
quence of actions.
 813/1540

The performance of a sequentially constrained

procedure may require planning or backtrack-
ing. For example, getting dressed is sequentially
constrained because at the moment in which one
has neither shoes nor socks on, putting on shoes
disables the operation of putting on socks. The
sequence of operations for orthodox dressing
contains a sequential constraint on the donning
of socks and shoes.

Whether a task is sequentially constrained may

depend on the representation of the task as well
as on the formal properties of the task. Zhang
(1992) has recently shown that it is possible to
change the sequential constraints of isomorphs
of the Tower of Hanoi problem by embodying
some of the sequence-constraining "rules" in the
physical instantiation of the problem. For ex-
ample, in one
 Page 199

version of the puzzle, placing a smaller disk on

a larger disk is a violation of a sequence-con-
straining rule. In Zhang's coffee cup isomorph,
this same move would be executed by placing a
small coffee cup in a larger onean act that
causes coffee to spill. These are ways to build
the sequential constraints into the behavior of
the game tokens and thereby reduce the require-
ment for memory of sequence-constraining rules
in support of planning and backtracking.
 815/1540

One general technique for turning sequentially

constrained tasks into sequentially uncon-
strained tasks is to manipulate the enablement
conditions of various operations. A simple rule
is "suppress the enablement of any operation
that could disable another already enabled oper-
ation." This can be done through interlocks. In
many automobiles, for example, the starter mo-
tor will not turn unless the transmission is in
park or neutral. This is the mechanical enforce-
ment of a sequential constraint on the engine-
starting procedure.

THE NAVIGATION TEAM AS A

PRODUCTION SYSTEM
 816/1540

Sequentially unconstrained procedures are eas-

ily distributed or can be solved by very loosely
interconnected systems. Tasks that have sequen-
tial constraints require some coordination
among the actions to be taken. There are many
ways to achieve this coordination. Specifying
the overall pattern of behavior in a script, a
score, or an overall plan is an obvious solution
to the sequencing problem. Since there are many
sequential constraints among the actions of the
fix cycle, one might assume that the fix cycle
unfolds according to a stored plan or description
of the sequence of actions involved.
 817/1540

In fact, it is possible for the team to organize its

behavior in an appropriate sequence without
there being a global script or plan anywhere in
the system. Each crew member only needs to
know what to do when certain conditions are
produced in the environment. An examination of
the descriptions of the duties of the members of
the navigation team shows that many of the spe-
cified duties are given in the form "Do X when
Y." Here are some examples from the
procedures:

a. Take soundings and send them to the bridge

on request.

b. Record the time and sounding every time a

sounding is sent to the bridge.

e. Take and report bearings to the objects

ordered by the Recorder and when ordered by
the Recorder.
818/1540
 Page 200

These and other instructions suggest that the

navigation team could be modeled by a set of
agents, each of whom can perceive the environ-
ment and can act on the environment when cer-
tain triggering conditions appear there. An inter-
locking set of partial procedures can produce the
overall observed pattern without there being a
representation of that overall pattern anywhere
in the system.
 820/1540

Each participant knows how to coordinate his

activities with the technologies and persons he
interacts with. The pelorus operators know to
negotiate the order of report with each other and
to take and report bearings when given the
"mark" signal. The recorder knows to say
"Stand by to mark" before the mark, then to say
"Mark," and then to attend to and record the
bearings. The plotter knows to plot the recorded
bearings to get a position, and then to project the
dead-reckoning positions and choose new land-
marks. The plotter's duties may cover a longer
procedural stretch than those of any other mem-
ber of the group, but even they do not come
close to completing the cycle. The whole cycle
is something that emerges from the interactions
of the individuals with one another and with the
tools of the space. The structure of the activities
of the group is determined by a set of local com-
putations rather than by the implementation of a
global plan. In the distributed situation, a set of
 821/1540

concurrent socio-computational dependencies is

set up. These dependencies shape the pattern of
behavior of the group. The existence of the plot-
ter waiting for bearings is how the system re-
members what to do with the recorded bearings.
These concurrent dependencies are not present
in the solo performance case.
 822/1540

When the nature of the problem is seen as co-

ordination among persons and devices, much of
the organization of behavior is removed from
the performer and is given over to the structure
of the object or system with which one is co-
ordinating. This is what it means to coordinate:
to set oneself up in such a way that constraints
on one's behavior are given by some other sys-
tem. This is easy to see in the use of the
recorder's wristwatch. Perhaps through some
complicated toe tapping or counting the recorder
could provide a regular meter for the perform-
ance of the rounds of fixes and dead-reckoning
projections, but that is unlikely. The only way
humans have found to get such tasks done well
is to introduce machines that can provide a tem-
poral meter and then coordinate the behavior of
the system with that meter. The system's co-
ordination with the meter of the watch is
provided by the recorder's coordinating with the
watch and the others' coordinating with the re-
823/1540
 Page 201

corder. The recorder's coordinating with the

watch requires him to maintain (1) vigilance to
the watch and (2) a test of when it is time to take
another round. For the second of these, he must
have (1) a procedure for determining (or a
memory of) when the next round should fall and
(2) a way of determining when that time has
been reached. Both parts of this task require
some cognition, to be sure, but no sophisticated
reasoning.
 825/1540

It should be noted, however, that the members

of the team may engage in considerably more
cognitive activity than the minimum required.
The various actors may have ideas about what a
particular task requires, and they may anticipate
particular sorts of failures on the basis of these
ideas. In the case of the recorder's maintaining
coordination with the watch, lack of vigilance
due to the appropriation of attention by other
tasks may cause the recorder to miss a mark.
The plotter, who shares the physical environ-
ment of the recorder, apparently sometimes par-
ticipates in that task redundantly and has been
observed to comment ''Isn't it about time for a
round?"
 826/1540

The coordination problem is more difficult for

the quartermaster standing watch alone in the
Standard Steaming Watch configuration. In that
case, the task performer must not only provide
coordination with each of the devices, but must
coordinate those activities with other activities.
In Sea and Anchor Detail, this latter sort of
higher-level coordination is accomplished via
the social coordination of the distributed situ-
ation. It is still ultimately provided by the hu-
man participants, but the cognitive load is not
only distributed; it is also lessened by distribu-
tion. Here is how: Consider the relation between
the pelorus operator and the recorder. The re-
corder coordinates his activities with the behavi-
or of his wristwatch. That is to say, he has del-
egated some aspect of the control of his own be-
havior to this external device. Now, the pelorus
operator coordinates his activity with the (tim-
ing) behavior of the recorder. He waits for the
"mark" signal to read his bearing. He has
 827/1540

delegated some aspect of the control of his own

behavior to the recorder. He has also delegated
some other aspects of his behavior to the device
with which he interacts. His behavior is nicely
and comfortably constrained by the two co-
ordination activities. He gets the "mark" signal
and invokes the coordination with the alidade-
which, in its relation to the world, is coordinat-
ing him (and, through him, the whole system)
with a particular aspect of the setting of the ship
in the surrounding world. He reads the bearing.
When he does so, the
 Page 202

recorder coordinates his (recording) behavior

with the pelorus operator. His other activities
are on hold while he attends to and (perhaps
simultaneously) records the bearing.
 829/1540

The problems of coordination in the solo per-

formance concern the control of the change in
roles and the meta-coordination required to
move through the sequence of steps required. It
is in the consideration of solo performances that
the apparent importance of executive function
emerges. This is to be expected because, from
the point of view of the individual, the task in
the solo performance is sequentially constrained
in a way that the modular tasks faced by the in-
dividual team members in the distributed per-
formance are not. There are certainly still se-
quential constraints in the distributed form, but
each individual is responsible for satisfying few-
er of these constraints than in Standard Steam-
ing Watch. In the place of an executive we find
a continual collision of interest and negotiation
of coordination status.
 830/1540

The quartermasters align themselves as a co-

ordinating structure that passes information
from one transforming device to another. The
people are the glue that sticks the hardware of
the system together. What is the relationship
between the position of the ship in the world and
the location of the fix on the chart? The formal
relationship is one of spatial correspondence.
The causal relationship is a tissue of human re-
lationships in which individual watchstanders
consent to have their behavior constrained by
others, who are themselves constrained by the
meaningful states of representational technolo-
gies. The sequential constraints of the proced-
ure, which are in part determined by the repres-
entation of the problem, constrain the universe
of social arrangements in which the procedure
can be performed. That is, they specify a co-
ordination task that must be solved by the social
organization of work.
 831/1540

SOCIAL STRUCTURE AND GOAL

STRUCTURE

The distribution of labor in Sea and Anchor De-

tail creates a distribution of attention to goals
such that the system is unlikely to halt before
completing a task. Imagine a problem described
by the goal tree shown in figure 4.3a. Individu-
als engaged in problems with deep goal trees
sometimes lose sight of higher-level goals and
halt after satisfying a lower-level goal. This is
the sort of problem faced by the solo watch-
stander in Standard Steaming Watch: Having
shot a bearing, what should I do next? Now,
suppose that rather than a single watchstander
we have a team, and we give each
 Page

Cognition p203

Figure 4.3 Goal hierarchy and distribution of responsibility f

goal satisfaction.
Responsibility for the satisfaction of the goals shown in pane
can be allocated to agents
(the enclosed shapes in panel b) in such a way that social de
pendencies provide control structure.
 833/1540

member of the team responsibility for a main goal

and for the subgoals required to achieve the main
goal. The areas of responsibility of the members of
the team are superimposed on the goal tree in figure
4.3b. Let the social contract between the agents be
such that a subordinate can halt only when his supe
or determines that the responsibilities of the subord
ate have been met. Agent A2, for example, can halt
only when agent A1 judges subgoal SG1 to be satis-
fied. Each agent is responsible only for a shallow se
tion of the goal hierarchy, so goal stack depth is no
problem for individual processors. Such a setup res
ults in computational control through a network of
cial relationships. When a problem has a deeply nes
ted goal structure, a social hierarchy can provide a
mechanism for distributing the attention to various
parts of the goal structure.
 834/1540

Social structure and problem representation both in

teract with goal structure in the implementations of
solutions to the problem of sequential control of ac-
tion. These things constrain the computational prop
erties of systems of socially distributed cognition an
cannot be excluded from an understanding of huma
cognition as it is manifested in such systems.

The fit between the computational dependencies an

the social organization is an important property of t
system. We might imagine a situation in which thos
of higher rank provide input to a lower-ranking ind
vidual, who integrates the information and makes d
cisions. That would be a very strange relationship
 Page 204

between the social organization and the compu-

tational dependencies. Gathering and providing
information for the support of a decision are
low-status jobs. Integrating information and
making decisions are high-status jobs. There
are, of course, exceptions, and some such rela-
tionships are more workable than others. In gen-
eral, however, the goals are in the hands of high-
er status individualsthose who control the goals
are, by cultural definition, of higher status.

Beam Bearings in Sea and Anchor Detail

 836/1540

The members of the navigation team normally

take for granted the accomplishment of sequen-
tial ordering of their own actions in coordination
with one another. The fact that the sequential or-
dering of action is no simple accomplishment,
especially when it is not built into the social or
material structures of the task, is highlighted in
the case of beam bearings. The beam of the ship
is those directions that are perpendicular to the
keel of the ship. Thus, the bearings of landmarks
that are off to either side of the ship, rather than
ahead or astern, are called beam bearings. The
navigation team must both decide which land-
marks to observe and determine an order in
which to observe them. To produce a high-qual-
ity fix, they must make their observations of the
landmarks as nearly simultaneously as is pos-
sible. The undesirable effects of delays between
the observations can be minimized by shooting
first the landmarks with which the angular rela-
tionship (bearing) is changing most quickly and
 837/1540

shooting last the landmarks whose bearings are

changing least rapidly.

Why Some Bearings Change More Rapidly Than

Others

The rate of change of a bearing depends two

things: (1) the component v of the relative velo-
city vector that is perpendicular to the line join-
ing the objects and (2) the distance between
them, d. Thus, dB/dt=v/d. For objects of equal
distance from the ship, the bearings of those ob-
jects off to the side of the ship, rather than those
that are ahead and astern, will be changing most
rapidly, because nearly all of the relative velo-
city of the object with respect to the ship will be
perpendicular to the line joining the object and
the ship. This is not the only consideration,
however, since for any given relative bearing
objects that are nearer will change in bearing
faster than
objects that are farther away. Distance and relative
fect the rate of change of bearings, and the total tim
observations affects the magnitude of the errors in t
tions. The quartermasters must therefore shoot the b
quickly and in the correct order.

Example of Effects of Beam Bearings

A ship with a speed over the ground of 10 knots wi

1000 yards in 3 minutes. Suppose a bearing 1000 y
the beam of the ship is taken 10 seconds late. What
this have on the plotted position? The ship will hav
yards in those 10 seconds, so the line of position de
that landmark's bearing will be 55 yards further dow
than it should have been (figure 4.4). All of the forw
of the ship is captured in the observation of a beam
this reason beam bearings are also called speed line
sider a bearing ahead or astern of the ship. The ship
same 55 yards in those 10 seconds, but the direction
ship to landmarks ahead and astern
 839/1540

Cognition p205

Figure 4.4 The effects of a late beam bearing. Heavy lines in

tions of the bearing
lines that are actually drawn on the chart to plot the fix. The d
 840/1540

depicts the location

of the ship at fix time. The light shape depicts the location o
seconds after the fix time.
 Page 206

changes little. The component of relative motion

between the objects that is perpendicular to the
line connecting the objects is small, so the rate
of change of the bearing is small, and the mag-
nitude of any error caused by delayed observa-
tions will also be small.

The Rule of Thumb

 842/1540

The requirement of this geometry is captured in

the rule "Shoot the beam bearings first." This
rule ignores the distance of the landmark from
the ship. In principle distance could be an im-
portant factor, but in practice it is not. This pro-
cedure is most critical when the ship is in a
channel, and in that case bearings on the beam
tend not to differ greatly in distance from the
ship. Furthermore, the measurements and calcu-
lations required to assess the effects of differing
distances of objects could be a more serious dis-
ruption of the fix procedure than the errors
caused by ignoring those effects. Thus, to shoot
the beam bearings first is a good rule of thumb
to use in sequencing the observations.

Configuring the Team

 843/1540

The application of the "shoot the beam bearing

first" rule to the Standard Steaming Watch situ-
ation is straightforward. The bearings must be
observed sequentially, and the beamiest bearing
should be observed first. In Sea and Anchor De-
tail, two pelorus operators work in parallel while
shooting the three bearings. One of the pelorus
operators will have two bearings to shoot; the
other will have just one bearing. How should the
pelorus operators sequence their actions in order
to produce the best fix?
 844/1540

Finding a procedure for performing this sequen-

tially constrained task turns out to be a nontrivi-
al problem for the crew. It must be kept in mind
that the port pelorus operator may not be able to
see the landmarks assigned to the starboard
pelorus operator and vice versa. Any bearing on
the beam of one side of the ship will not be vis-
ible to the pelorus operator on the opposite side,
so neither pelorus operator can see enough to
decide who has the beamiest bearing. The direc-
tional relationship between the bearings is easier
to imagine at the chart table, but determining the
shooting sequence there would impose an addi-
tional burden on an already busy bearing record-
er. Before we examine what the crew actually
does with this problem, it may be useful to in-
dicate the form of the
 Page 207

correct solution. The sequencing of observations

could have adverse consequences if the order in
which the elements of the procedure were ex-
ecuted delayed the observation of a landmark.
The bearing recorder is a limiting resource in
this procedure because he can attend to only one
bearing report at a time. Imagine that the star-
board pelorus operator has one landmark to
shoot, and that it is on the beam. The port pelor-
us operator has the other two landmarks, but
they are not so beamy as the one to starboard.
Figure 4.5 depicts this situation.
 846/1540

If the pelorus operator with the beamiest bearing

goes first, and going first is understood to mean
both shooting and reporting, then the sequence
of actions shown in figure 4.5a will result. (This
figure is intended only to show relative times of
completion.) The solution shown in figure 4.5.a
mimics the structure of the performance when it
is done by a single watchstander. It fails to take
advantage of the parallelism of activity that is
possible with two pelorus operators.
 847/1540

Both of the pelorus operators could observe a

bearing immediately upon hearing the "mark"
signal. If each pelorus operator observes the
"beamiest" of the assigned bearings immedi-
ately, and they still report the beam bearing first,
the sequence shown in figure 4.5b will result.
The port pelorus operator will have to wait
while the starboard pelorus operator reports the
bearing of the landmark on the beam because
the bearing recorder can only attend to one re-
port at a time. This is an improvement over the
previous solution because it packs the same
number of actions into a smaller period of time,
thus reducing the magnitude of the errors in pos-
ition caused by delays in making the observa-
tions. In particular, the first bearing observed by
the port pelorus operator is shot at the mark sig-
nal, rather than after the starboard pelorus oper-
ator has shot and reported a bearing, and the
second bearing observed by the port pelorus op-
erator now comes one action cycle earlier. This
 848/1540

implementation takes advantage of some of the

parallelism of activity that is possible with two
pelorus operators.

A further gain can be achieved by realizing that

the observation of the bearing and the report of
the bearing can be procedurally separated from
one another. The computational constraint is on
the sequence and times of the observations. The
beam bearings must be shot before less beamy
bearings, and the three observations must be
made as near in time to the mark signal as is
possible. There is no similar constraint on the
reporting of the observations. As far as the
 850/1540

Figure 4.5 Coordinating the actions of the pelorus operators.

one to starboard,
the pelorus operators can organize their actions in several po
of activity; shoot and
 851/1540

report actions linked; beamiest bearing reported first. (b) O

bearing shot at mark);
beamiest bearing reported first. (c) Overlapping observation
pelorus operator
with two landmarks reports fir
 Page 209

quality of the fix goes, as long as the three bear-

ings are accurate, they may be reported in any
order-beam first, beam second, or beam last. In
order to take full advantage of the parallelism of
action that is possible in the team configuration,
two rules are required: (1) Each pelorus operator
should shoot the beamiest of the landmarks as-
signed to him immediately at the mark signal,
and (2) the pelorus operator who has two bear-
ings to shoot and report should report first. The
application of these two rules results in the pat-
tern shown in figure 4.5c.

Instructions Concerning Beam Bearings

 853/1540

When explained with diagrams like those shown

in figure 4.5, the appropriate patterns of activity
are fairly obvious. When the members of the
navigation team attempt to organize their efforts
in the performance of the task, however, the ap-
plication to the group condition of the "rule of
thumb" that serves so well in the solo task per-
formance case is problematic at best. All of the
members of the team seem eventually to "know"
and understand the rule, but their attempts to use
the rule to coordinate their actions in time re-
peatedly fail. To see why, consider the instruc-
tions that are passed among the members of the
team concerning the need to take beam bearings
first. In the simplest case, the sequencing in-
struction may come from the people working at
the chart table.

EXAMPLE 1
 854/1540

The landmarks are Hotel del, Dive Tower, and

Point Loma. The ship is outbound from the har-
bor, west of the 1SD channel marker. The ship's
course is 270°, so 360°and 180° °are the beam
bearings. The starboard pelorus operator's name
happens to be Mark. Here is the beginning of
the round:

Recorder: Stand by to mark, Point Loma, Hotel

del, and Dive Tower.

Plotter Tell him to take Point Loma first. It's on

his beam.

Recorder: Take Point Loma first, Mark. Beam

bearing first, mark it.

SW: Point Loma 3 5 9.

 855/1540

In this example, an invocation of the rule is em-

bedded in the instructions from the bearing re-
corder to the pelorus operator concerning the or-
der in which the observations should made. The
example is unproblematic, but there is an oppor-
tunity here for the
 Figure 4.6 The situation of example 2.

pelorus operator to see that Point Loma is an examp

beam bearing and perhaps add to his knowledge of
ing of the expression. This is thus also an example
socialization.

EXAMPLE 2
 857/1540

Example 2 is a case in which the recorder chose tw

marks near the beam on the same side of the ship. A
sequence, the lines of position did not converge on
triangle. The plotter tried to explain the spread of th
the recorder. Ship's course: 324°. Beam bearings: 0
Landmarks: Pier 5 122°, Dive Tower 244°°, Hotel d
(See figure 4.6.)

Plotter. See, the reason is ... you get the spread, ah,
these (Hotel del and Dive Tower) are both close to

Recorder: Yeah.

Plotter: Right. They are both close to the beam. The

sp ... I mean, unless he can get 'em really, really fas
gonna split it even at 10 knots, you know. Ten knot
progressing along in between the time he reads thos

Recorder: Yeah.

Plotter: 'Cause they're both so, both so close to the b

Recorder: Yeah.
 858/1540

Plotter: That's the reason.

There are two potential problems with what the rec

done. The plotter explains one of them at length. Si
bearings for the port pelorus operator are near the b
matter which one the pelorus operator shoots first, t
will change while he is shooting the first. The secon
is that two bearings within
 Page 211

Cognition p211

Figure 4.7 The situation of example 3.

 860/1540

30° of each other intersect at a shallow angle, so

small errors in the observation move the point of
intersection a long ways.

EXAMPLE 3

A beam bearing (to the carrier pier) was the last

bearing observed. The plotter and the recorder
have discussed the effects of shooting beam
bearings last, and the recorder tells the pelorus
operators to remember to shoot the beam bear-
ing first. Course: 309°. Beam bearings: 039 and
219. Landmarks: Reuben E. Lee 328°°, Aero
Beacon 281°°, Carrier Pier 210°. (See figure
4.7.)

Recorder: (to plotter): That was a late bearing.

 861/1540

Plotter: Yeah, see, he's gotta remember that.

These late bearings on beam are doing, is why
you get these big open fixes. Your beam
bearing's gotta shoot first. Tell those guys to
watch what the hell they are doing.

Recorder: That's what I've been tellin' 'em.

Plotter: Yeah, but tell him, OK, over there he's

got to go to the beam one first.

Recorder: OK.

Plotter: Because that is the one that is changing

on him real fast.

Recorder: (to pelorus operators): OK guys. Re-

member to shoot the beam bearings first.
 Page 212

Notice that much less is communicated to the

pelorus operators in the recorder's last turn than
what passes between the plotter and the record-
er. The plotter and the recorder share the chart
as well, and it is a rich communicative resource.
When the plotter refers to ''these big open fixes"
he is pointing to the fix triangles on the chart.
The pelorus operators are separated from this
scene by the phone circuit. The instruction to
them contains only an admonition to shoot beam
bearings first.

The next example raises the possibility that the

pelorus operators do not know how to make
sense of what they are being told, and there is
nothing here to help the pelorus operators de-
termine how to put the advice into practice.

EXAMPLE 4
 863/1540

Example 4 is the most complete and complex

interaction concerning the strategies for sequen-
cing the activities of the pelorus operators. Be-
cause of the size and complexity of this ex-
ample, I will break it into segments punctuated
with commentary. It begins with a question
from the starboard pelorus operator (SW).

SW: I got two points. You want the fartherest

first, and then the closest?

Recorder: Oh, OK.

SW: I got two points, right?

Recorder: Whatever is closer to the beam. Shoot

the beam first. The one closer to sidewise. You
got two points, um.

SW: So I shoot the fartherest one first, then the

closest.
 864/1540

Recorder: If you got three, you shoot the one in

the middle, then forward, then aft. If you got
two, forward and aft.

SW: If you got ... (interrupted by port wing

pelorus operator)

Recorder: OK, just a reminder. You always

shoot the beam bearings first. If you got three of
them, you best shoot the ones ... (port pelorus
operator talking)

Recorder: If you got two of 'em, only shoot the

ones up forward and back.

Plotter: Huh?

Recorder: I was just trying to explain which

ones to shoot first. Beam bearings first.

Plotter: Beam, then forward and aft.

 Page 213

Recorder Forward and aft. If he's got two, he's

got to shoot forward first and then aft. (Recorder
talks to port pelorus operator on headset.)

The origins of the starboard pelorus operator's

ideas about shooting the most distant landmarks
first is unknown. The instruction from the re-
corder, "If you got two, (shoot) forward and
(then) aft," will lead to the wrong sequence if
the aft bearing is closer to the beam than the for-
ward one. The difficulty here is interpreting the
meaning of the rule of thumb across a wide
range of possible configurations of landmarks.
After hearing this exchange, the plotter went out
to the starboard wing to talk to the starboard
pelorus operator. (In the conversation that fol-
lowed, REFTRA refers to an upcoming inspec-
tion in which the crew's performance will be ob-
served by an evaluation team.)
 866/1540

Plotter: Remember, one of the things you guys

wanna do, the guys on the wings. Them
REFTRA guys'll watch for it no matter who's
out on the wings, and supposedly you're all inter
... any of you can be there. If the guy that sees ...
When a round is comin' up and he knows, he
says, see, you know, pretty clo ... you know
about, you can tell. The guy who knows he's going to have a beam
bearing. He gets through saying "OK on the next round I want you to
have ..." and that guy can see he's gonna have it on the beam, tell the oth-
er guy "I've got the >beam bearing." OK, so that . ..

SW: Beam? Where is the beam? Right here?

Plotter: Right here (demonstrates with his arms).

The beam is between here and here. The bearing
that's chas ... changing fastest. Right along side
of the ship. Even if it's out there, it's the bearing
...

SW: Uh huh
 867/1540

Plotter: ... that's changing the fastest. OK. That's

your speed line. That's the one that should come
first, and then the other guy can go ahead and
shoot forward or, or, you know, he can go and
shoot. But always the beam first. And if the guy
that's got the beam bearing, see, the other guy
can't see what you can see. Just like you can't
see his. If you got the beam bearing, say "Hey,
mine is the beam bearing." That way, he'll shut
up ...

SW: You'll never have ...

Plotter: ... and let you give your beam.

 Page 214

SW: Comin' into a channel though, you could

both have a bearing on the beam.

Plotter: True, you could, you could.

SW: You could, yeah, but not ...

Plotter: But not very often, 'cause he don't give

things that are right across from one another.
 869/1540

In this conversation it becomes evident that the

pelorus operator was not at all clear on the
meaning of "beam bearing." The plotter de-
scribes it to him, but also includes additional
features of the beam bearings (e.g., "changing
fastest" and ''speed line") which are conceptu-
ally salient to the plotter but are probably mean-
ingless to the pelorus operator, who is just trying
to figure out how to identify the beam bearing.
Notice also that the plotter links the observation
of the bearing to the report of the bearing in his
description of how the pelorus operators should
negotiate the sequence of their activities. This is
important because observation and report must
be uncoupled in order to produce a more effi-
cient procedure.

EXAMPLE 5
 870/1540

Later in the same entry, the ship was inbound.

Course: 345°. Beam bearings: 075°and 255°°.
Landmarks: port, Point Loma 335; starboard,
Dive Tower 045 and Hotel del 032. (See figure
4.8.) It is unclear what either participant takes
"go first" to mean in the following exchange.

SW: When I got two points and he's only got

one, shouldn't he let me go first?
Cognition p214
 871/1540

Figure 4.8 The situation of example 5.

 Page 215

Recorder: Nah, it doesn't matter really right

now.

SW: Doesn't matter?

(S in conversation with CIC seems to have no

time to pursue question from SW)
 873/1540

The answer to the starboard pelorus operator's

question should have been an unequivocal
"Yes." The recorder's response probably leaves
the starboard pelorus operator in some confu-
sion. Saying that it doesn't matter is a way for
the recorder to indicate that he does not wish or
does not have time to intervene in the negoti-
ation between bearing takers at this moment.
Unfortunately, this conversational move also
has a substantive interpretation: that the number
of landmarks one has is irrelevant to the order in
which the bearings are reported. In this case, the
starboard pelorus operator should observe Dive
Tower first (before observing Hotel del, because
Dive Tower is beamier) and should report first
(that is, before the port pelorus operator reports,
because the port pelorus operator has only one
bearing to report). The two senses of "first" are
different, as are the reasons for the two relative
orderings. In either case, however, the starboard
pelorus operator should have gone first.
 874/1540

EXAMPLE 6

Course: 35°. Beam bearings 083°and 263°°.

Landmarks: port, Point Loma 327; starboard,
Dive Tower 058, Hotel del 044. (See figure 4.9.)
In this sequence the recorder encourages the
linkage of shooting and reporting.

Figure 4.9 The situation of example 6.

875/1540
 Page

SW: John?

Recorder: Yo!

SW: Is the Dive Tower right on our beam?

Recorder: Say again?

SW: Dive Tower. Isn't it just about on our beam?

Recorder: Yeah, just about. (2 seconds) OK, Shade

PW: What?

Recorder: Steve's gonna be shooting the Dive Towe

first, so let him say, uh, let him say the bearing first

PW: You want Point Loma last, then?

Recorder: Yeah, that's fine.

 877/1540

In this example, the starboard pelorus operator's qu

tion about the beam status of the Dive Tower is inte
preted by the recorder as also being an indirect requ
to shoot and report that bearing first, perhaps in ac-
cordance with their previous discussion. The record
takes up the role of negotiating the sequence and se
to expect the port pelorus operator to share this inte
pretation of the starboard pelorus operator's request
Once again, the recorder's instructions explicitly co
bine shooting with reporting.

EXAMPLE 7

Time: 6 minutes later. Course 353°. Beam bearings:

083°and 263°...° Landmarks: port, Point Loma 275°;
starboard, Hotel del 066°°, Light Zulu 049°. (See fig
4.10.)
 878/1540

Figure 4.10 The situation of example 7.

 Page 217

Plotter: What did you take a bunch of beam

bearings for? Why ain't you shooting up there
(ahead) some place. Look what you did. You
shot three beam bearings. You shot three beam
bearings. You better tell 'em to shoot from up
ahead some place.

Recorder: OK. Drop Point Loma and pick up

Ballast Point, John.

PW: OK.

Failure to Uncouple Shoot and Report Actions

 880/1540

The rule "shoot the beam bearing first" works

fine in solo watchstanding. Yet the attempts of
the navigation team to use that rule to coordin-
ate their actions through time fail repeatedly.
Why is the application of this simple rule so dif-
ficult? Why is the team unable to use this rule to
organize its performance? The above examples
of instructions concerning the taking of beam
bearings provide some clues.
 881/1540

When the rule is invoked in Standard Steaming

Watch by a single quartermaster standing watch
alone, the beam bearing refers to the bearing in
the set of three that is nearest the beam of the
ship, and the sequence specifier "first" is estab-
lished with respect to the entire set of three bear-
ings. In the group version of the task, a pelorus
operator cannot always determine whether any
bearing he has been assigned is nearer the beam
than any bearing assigned to the other pelorus
operator. A pelorus operator stationed on one
wing of the ship cannot give either of these
words the meaning it has for a solo watchstand-
er. It is as though other words were missing
from the simple statement of the rule. A more
explicit version of the rule in the solo watch-
standing case would be "Of the set of three
bearings, shoot the beam bearing first." It is not
necessary to say these words in the solo watch-
standing context, because the entire set of three
bearings is the watchstander's responsibility.
 882/1540

Their presence in that context is not needed, and

their absence when the context has changed is
not noticed. If these words had been present, the
problems of giving the rule to the individual
members of the team may have been more ap-
parent. The statement of the rule implies or as-
sumes a perspective that takes in all three bear-
ings at once. That is the meaning of 'beamiest'
that we get from looking at the diagrams, it is
the meaning exchanged by the recorder and the
plotter when they are looking at the chart, and it
is the meaning that the plotter brings out onto
the wing when he explains to the
 Page 218

pelorus operator how to apply the rule. But the

perspective on which this meaning rests is not
available to the pelorus operators. Neither pelor-
us operator can see the entire set, and neither
can really know the relation of the bearings on
his side to those on the other. The pelorus oper-
ators need a meaning of 'beamiest' that they can
apply on the basis of what they can see, and they
cannot see all three bearings at once. Transport-
ing knowledge from the solo performance con-
text to the group performance context is very
problematic. It may require changes in the
meanings of words.
 884/1540

There is also considerable difficulty in interpret-

ing the meaning of the rule across a wide range
of possible configurations of bearings. The
pelorus operators have never stood watch alone,
and may not even know what the beam is. This
highlights the fact that the group performance
requires a particular distribution of knowledge.
 885/1540

Both the plotter and the recorder link the obser-

vation of the bearing to the act of reporting the
bearing in their descriptions of how the pelorus
operators should negotiate the sequence of their
actions. This has multiple causes. First, ob-
serving and recording are a unit in the solo ver-
sion of the task (which is the source of the rule).
Second, an explicit vocabulary is required to
sort out observing from reporting. It is not easy,
in the absence of a diagram, to describe what the
order of actions should be. It is still more diffi-
cult to negotiate this sequence without a prior
agreement about how observing can be de-
coupled from reporting. What is needed is a lan-
guage for the two aspects of the rule: The oper-
ator who has two bearings should report before
the operator who has only one, and he who has
two bearings should always shoot the beamier of
the bearings before the other. The rule that
comes from solo performance has no such
terms.
 886/1540

Here is an aspect of the organization of team

activity that is problematic for the team. The
members perceive it as being a problem, and
they apply themselves to it, but they come to no
satisfactory solution. Transporting the simple
rule of thumb from the solo watchstanding con-
figuration to the group configuration presents
unexpected difficulties. The words of the rule
themselves seem to change meanings when the
rule is moved to a new context, and new words
seem necessary to make distinctions in the new
context that were not needed and not made in
the old context. The conceptual linkage between
observing and reporting prevents the team from
exploiting possibilities of the socially distributed
system for
 Page 219

manipulating the temporal relations among ac-

tions. It is difficult to reason about a system as
complex as this from a position within it.
Quartermasters are not trained in the sorts of re-
flection on organization that are required to
solve problems like this.

Going Beyond the Job Description

 888/1540

One important aspect of the social distribution

of this task is that the knowledge required to
carry out the coordinating actions is not dis-
cretely contained inside the various individuals.
Rather, much of the knowledge is intersubject-
ively shared among the members of the naviga-
tion team. This permits the human component of
the system to act as a malleable and adaptable
coordinating tissue, the job of which is to see to
it that the proper coordinating activities are car-
ried out. In their communication and in their
joint actions, the members of the navigation
team superimpose themselves on the network of
material computational media. They provide the
connecting tissue that moves representational
state across the tools of the trade. In addition,
they dynamically reconfigure their activities in
response to changes in the task demands. This
amounts to a restructuring of functional systems
that transcends the individual team members.
The individual team members do their jobs by
 889/1540

constructing local functional systems that bring

media in their immediate environment into co-
ordination. They also must coordinate their
activities of helping one another achieve co-
ordination. The computation is implemented in
the coordination of representational states, and
the human participants coordinate their coordin-
ating actions with one another.

Shared Task Performances

 890/1540

Sometimes the coordination of actions occurs at

a very fine grain. One day during Standard
Steaming Watch, Silver and Smith were work-
ing the chart table together. They needed to use
the hoey determine the direction of a line
between two points. Smith placed the tip of his
pen on the one of points. Silver put the point of
his pencil down on the second point and pushed
the edge of the hoey arm up against the pencil
and pen points. Then, while Silver held the hoey
arm in place, Smith rotated the base of the pro-
tractor to align it with a latitude line and read
the bearing from the hoey
 Page 220

scale. This ad hoc division of labor was based

on a shared understanding of the microstructure
of the task. There was no verbal negotiation of
the parts of the task to be done by each man;
they simply created this coordination in the do-
ing of the task. The social skills required to
enter into shared task-performance relationships
probably develop fairly early in life.

Distributed Memory
 892/1540

Task-relevant information is present in many

representations in this system. Some of these
representations are in the minds of the parti-
cipants. During an exit from a harbor, the plotter
expected a 1000-yard interval between fixes. In-
stead he measured only 700 yards. This indic-
ated that the ship had slowed from 10 to about 7
knots. This is troubling, because it indicates a
discrepancy between the information used to
project the dead-reckoned position and the actu-
al observations. To resolve the discrepancy, the
plotter began by talking to himself, but quickly
addressed a question to the keeper of the deck
log.

Plotter: This is not showing no, a no goddamn ...

they show a twothirds, still got a two-thirds bell,
right?

Deck log: One-third bell.

Plotter: Why?
 893/1540

CIC talker: Got the pilot off.

Plotter: Oh, that's why. OK, he's [the pilot] get-

ting off right now, isn't he?

CIC talker: Yeah. He [the OOD] went back to 5

knots.

Plotter: That's what messed me up. They have

this goddamn 7-knot goddamn thing in here and
I'm trying to figure out why.
 894/1540

In this exchange, both the CIC talker and the

keeper of the deck log provide the plotter with
task-relevant representational state. The plotter
could have gotten the information about the one-
third bell (ahead one-third on the engine-order
telegraph) from the deck log itself, or from the
engine-order telegraph, or from the leehelms-
man as well as from the keeper of the deck log.
This bit of system state is redundantly represen-
ted in the memories of several participants, and
in written records. The information about the
departure of the harbor pilot was probably not
present in any record at this time, because as far
as the bridge crew knew the pilot was still on
board.
 Page 221

Still, the CIC talker knew about this and was

able to judge that it would be of use to the plot-
ter at this time.

In another instance, while simultaneously

watching a fishing boat that crossed close under
the bow of the ship and discussing the watch bill
for the remainder of the day, the plotter and re-
corder missed a fix time. This problem was
caught by the keeper of the deck log about 2
minutes late.

Deck log: Chief, you're going to have another

call. Missed at 3. Your round at 3.

Plotter: I'll get one here in a minute.

Recorder: Stand by to mark.

Plotter: Time is 5, yeah 5; we'll just kind of

space this one out.
 896/1540

Even though timing the fixes is not part of the

keeper of the deck log's job, he is a participant at
the chart table and in this case, happens to have
noticed that a scheduled fix was missed. This
sort of overlapping knowledge distribution is
characteristic of cooperative work and is an im-
portant source of the robustness of such systems
in the face of error and interruption

Recorder Cuing the Plotter

 897/1540

With only two landmarks visible, the team sub-

stituted a radar range on one of the visible land-
marks for the third line of position. After ad-
vising the pelorus operators to stand by for a
round, the recorder turned to the plotter and
said: "Get a range, Chief? (2 seconds) Mark it."
The plotter had apparently forgotten that he was
required to take the radar range of the landmark
as part of the position fixing operation. In this
case, an element of the sequential organization
of the plotter's activity was provided by the re-
corder. This is not strictly in accordance with
the normal division of labor. The plotter was
supposed to remember that the "stand by to
mark" signal was his cue to take the radar range.
He and the recorder had come to an agreement
about a nonstandard division of labor to meet
the needs of an unusual situation. We may spec-
ulate that the recorder's memory for the plotter's
role was in part cued by the fact that he had
labeled a column in the bearing record log as
 898/1540

"Range Pt. Loma." The need to fill the cell in

the bearing log at the intersection of this labeled
column with the row representing the current
time acts as a memory for the decision to take
the radar range and may remind the recorder of
the plan for getting the range.
 Page 222

Landmark Descriptions

Since the ideal manning requirements are sel-

dom met, it is often the case that the pelorus op-
erators are not entirely familiar with the land-
marks they will be required to observe. When
pelorus operators fail to locate a landmark, the
plotter and recorder may attempt to help them
out by providing verbal descriptions. In the fol-
lowing example, the port pelorus operator is un-
able to find a landmark. (This one, the Dive
Tower, is frequently a problem.) Notice also that
the plotter interjects additional sequencing ad-
vice at the beginning of the fix cycle. (The star-
board pelorus operator's name is Mark). Those
portions of the recorder's talk that are transmit-
ted over the phone circuit appear in boldface.

Recorder: Stand by to mark, Point Loma, Hotel

del, and Dive Tower.
 900/1540

Plotter: Tell him to take Point Loma first. It's on

his beam.

Recorder: Take Point Loma first, Mark. Beam

bearing first. Mark it.

SW: Point Loma, 3 5 9.

Recorder: 3 5 9 Point Loma.

Nav: Ten knots good speed?

Plotter: Yes. Anything you want. We're clear

now. Wherever you want to go.

PW: Hotel del 0 3 8.

Recorder: 0 3 8 Hotel del.

PW: I can't find the Dive Tower.

Recorder: Can't find the Dive Tower.

 901/1540

Plotter: Tell him it is about 8 degrees, 9 degrees

to the right of Hotel del (plotting)

Recorder: Nine degrees?

Plotter: Yeah.

Recorder: It's about 9 degrees to the right of

Hotel del. It's about 0 4 6.

The clue provided by the plotter in this case is

not a description of the landmark, but a location
relative to a previously reported landmark to-
ward which the pelorus operator should look.
This is evidence of how strong the expectations
of the position of the landmark are. Also notice
that the recorder transforms the description even
further, converting it by mental arithmetic to a
bearing at which the pelorus operator should
look for the landmark.
 Page 223

In this example we see something of the rela-

tionship of social structure and computational
structure. Given a computational procedure and
a social organization, there are better and worse
ways to distribute the computation across the so-
cial network. One way in which the distribution
of tasks can be better or worse concerns the rela-
tion between the amount of information that
must be passed between computational seg-
ments and the capacity of the medium of com-
munication between the participants responsible
for those segments. That is a computational ar-
gument for a particular decomposition of the
task. The observed decomposition works well in
many respects but is frustrating for certain
classes of problems.

Recorder Setting Up Plotting Tool for Plotter

 903/1540

In another instance, the plotter was called away

from the chart table just as the bearings of the
landmarks were reported. After recording the
bearings in the bearing record log, the recorder
reached across the chart and set the hoey arm to
the value of the first bearing. When the plotter
returned to the chart table, he looked in the log
for the bearing. When he looked at the hoey, he
noticed that it had already been set to the proper
bearing. He then simply aligned it with the chart
and plotted the line of position. Setting the hoey
was not in the recorder's job description, but it
was a way of pushing the representational state
a little further toward its end point.

Flexibility and Robustness

 904/1540

These examples also illustrate the robustness of

the system of distributed knowledge. If one hu-
man component fails for lack of knowledge, the
whole system does not grind to a halt. If the task
becomes difficult or communications break
down, the navigation team does not have the op-
tion of stopping work. The task is driven by
events and must be performed as long as the
ship is underway. In response to a breakdown,
the system adapts by changing the nominal divi-
sion of labor. It is the bearing taker's job to find
the landmarks, for example, but if he is unable
to do so, some other member of the team will
contribute whatever is required to ensure that
the landmarks are found and their bearings ob-
served. This robustness is made possible by the
redundant distribution of knowledge among the
members of the team, the access of members to
one
 Page 224

another's activities, and the fact that the indi-

vidual workloads are light enough to permit mu-
tual monitoring and occasional assistance. Both
the knowledge required to do the task and the
responsibility for keeping the system working
are distributed across the members of the navig-
ation team. We can think of the team as a sort of
flexible organic tissue that keeps the informa-
tion moving across the tools of the task. When
one part of this tissue is unable to move the re-
quired information, another part is recruited to
do it.

Performance as a Language of Social Interaction

 906/1540

The division of labor mandated in Sea and An-

chor Detail distributes the elements of the fix
cycle across social space. Wherever computa-
tions are distributed across social organization,
computational dependencies are also social de-
pendencies. Performance is embedded in real
human relationships. Every action is not only a
piece of the computation, a bit of the task com-
pleted; it is also a social message. Building and
maintaining good social relationships becomes
an important motive for competent performance.
In order to do the computation, the members of
the team must interact. They depend on one an-
other. More precisely, the portion of the compu-
tation for which each is responsible may depend
on the portions for which the others are respons-
ible. In order to plot the next line of position, the
plotter needs the bearing, which means he needs
to communicate with and secure the cooperation
of the pelorus operator.
 907/1540

An important aspect of these interrelated social

and computational structures is that both of
them provide constraints on the behavior of the
participants. One can embed a novice who has
social skills but lacks computational skills in
such a network and get useful behavior out of
that novice and the system. The reason is that
the social structure (and the structure of the
tools) of the task may provide enough con-
straints to determine what turns out to be a well-
organized computational behavior even though
the behavior was not motivated by any under-
standing of the computation. The task world is
constructed in such a way that the socially and
conversationally appropriate thing to do given
the tools at hand is also the computationally cor-
rect thing to do. That is, one can be functioning
well before one knows what one is doing, and
one can discover what one is doing in the course
of doing it.
908/1540
 Page 225

The social structure is not only the framework

on which the communication is based, it is also
the mechanism that is in place prior to the inter-
actions to ensure that they take place as re-
quired. Why should the pelorus operator cooper-
ate? Because adequate performance is the cur-
rency of social interaction. The novice quarter-
master is institutionally located in such a way
that his actions can be taken both as contribu-
tions to the process and as claims to or justifica-
tions of membership in the social world of the
other quartermasters. And, since this is the milit-
ary, a novice who does not perform adequately
can be harshly sanctioned.

How to Say Things with Actions

 910/1540

My initial assumption about work in military

settings was that behaviors are explicitly de-
scribed and that people act more or less as auto-
matons. It should be apparent by now that this is
far from the case. I also naively assumed that
most communication on the job would be part of
the job and nothing more. As I worked with the
data, something that Roy D'Andrade once said
kept coming back to me. A student was making
a point about what people do at work, saying
that in an auto factory people mostly make cars.
Roy said something like: ''How do you know
what they are doing? Maybe what they are mak-
ing is social relationships and the cars are a side
effect."
 911/1540

It is clear that when quartermasters report bear-

ings, assign landmarks, or ask for data, they are
not just constructing position fixes; they are also
constructing social relationships. And the fact
that their respective responsibilities are so well
specified does not eliminate the possibility of
loading social messages into the communication
acts that make up the work. In fact, the well-
formed expectations about what constitutes
competent verbal behavior in this setting may
give the participants an especially subtle means
of communicating social messages. Doing the
absolute bare minimum required when others
know that one has the time and resources to do
more is a clear statement.

Computational Properties of the Navigation

Team
 912/1540

Local functional systems are established in the

individual jobs. Each member of the navigation
team is responsible for the
 Page 226

construction of a number of local functional sys-

tems. These are the processes of bringing media
into coordination as described in chapter 3.

These local functional systems are coordinated

in the interaction of the members of the team. In
their interactions, the team members assemble
the component functional systems into a larger
functional system.

The larger system has cognitive properties very

different from those of any individual. In fact
the cognitive properties of the navigation team
are at least twice removed from the cognitive
properties of the individual members of the
team. The first remove is a result of the trans-
forming effects of the interactions with the tools
of the trade (chapter 3); the second remove is a
consequence of the social organization of dis-
tributed cognition.
 914/1540

In Sea and Anchor Detail the navigation team

implements a distributed problem-solving sys-
tem in which various elements of the computa-
tion are embodied in the operations of functional
systems constructed by the members of the
team. Among the advantages of distributed pro-
cessing discussed by Chandresekaran (1981) are
the following:
 915/1540

The decomposition of processing is a strategy

for controlling the complexity of computation.
By breaking the problem down into pieces, the
team can have several workers operating in par-
allel. This decomposition of the task also per-
mits each member of the team to attend closely
to only a limited set of data. As Chandresekaran
points out, the complexity of computation is of-
ten an exponential function of the size of the in-
put space. If the problem can be divided up,
each person can deal with a tractable problem.
For example, each pelorus operator needs to
deal with the landmarks on only one side of the
ship's track. It is possible to learn the landmarks
for the starboard side without knowing the land-
marks for the port side. There are also filtering
processes applied that prevent the growth of in-
put space. Thus, the recorder does not normally
have to deal with the complexity of the visual
scene outside the ship. His experience of the
bearings is pre-processed by the pelorus
 916/1540

operators and takes the form of strings of

spoken digits. An important advantage of social
distribution of computing is that novices can be
embedded in social arrangements such that
much of the structure required for them to or-
ganize their activity is available in the social re-
lations. Even though the skills have mainly
social
 Page 227

significance to the novices, they can learn a

great many skills that have computational signi-
ficance to the system.

A second property noted by Chandresekaran is

that distributed computing "increases the pro-
spects for graceful degradation" of system per-
formance when components fail. This is appar-
ent in the response of the navigation team to loc-
al failures such as the inability of a pelorus oper-
ator to locate a landmark. Because the members
of the team have overlapping knowledge, it is
possible for them to reconfigure dynamically in
response to a problem. The individuals are a sort
of flexible tissue that moves to ensure the
propagation of task-relevant representational
state. Because their competences overlap and
they have access to one another's activities, they
are able to aid one another and fill in for one an-
other in the event of a local failure.
 918/1540

Adaptation to change may be easier in distrib-

uted than in centralized systems.
Chandresekaran says that "as the external envir-
onment changes, distributed information pro-
cessing makes adaptation to change easier, since
again, as long as the rate of change is not large,
changes to the system can be mostly local." I
will discuss an example that illustrates this in
some detail in chapter 8.

Having the pelorus operators negotiate the

shooting sequence among themselves is an ex-
ample of the uses of modularity. Notice that this
modularity was violated when the starboard
pelorus operator asked the bearing recorder to
settle the shooting and reporting order.
 919/1540

One of the costs of distribution is the filtering

performed by the sensors. The pelorus operators
are expected to pass only the results of their
computations to the bearing recorder. All in-
formation about the process that went into
achieving that result is lost in the report of the
bearing as a single number. This reduces the
bandwidth required for communication (the
phone circuit is adequate for this), and it also re-
duces the processing demands on the central
processor (plotter). However, this kind of filter-
ing makes it more difficult to diagnose the
causes of errors committed by the pelorus oper-
ators, since nothing of the process is normally
communicated.
 920/1540

Representing the bearings symbolically also in-

troduces new possibilities for error. For ex-
ample, the landmarks light 2 and light zulu are
very different from each other in location and
appearance. It is unlikely that one would ever be
mistaken for the other. Their
 Page 228

symbolic representations, "light 2" and "light

Z," however, are very similar and might easily
be confused. This potential was recognized by
the plotter, and he instructed the recorder to put
a slash through the Z.

Another potential cost of distribution is the po-

tential disruption of one processor by another.
The buffers are a way to overcome this sort of
temporal discoordination. The phone circuit has
different properties from the bearing log be-
cause one endures in time while the other does
not.
 922/1540

The problem of the design of the distribution of

labor remains. As we saw with the case of beam
bearings, the mapping from individual perform-
ance to the group configuration is a nontrivial
one. Opportunities exist in the distributed ver-
sion of the task that are simply not present in the
solo-performance case. Finding and exploiting
these opportunities may require reflection on ex-
plicit representations of the work itself, and the
members of the navigation team are ill equipped
to do such reflection.
 923/1540

The theme of this chapter is that organized

groups may have cognitive properties that differ
from those of the individuals who constitute the
group. These differences arise from both the ef-
fects of interactions with technology and the ef-
fects of a social distribution of cognitive labor.
The system formed by the navigation team can
be thought of as a computational machine in
which social organization is computational ar-
chitecture. The members of the team are able to
compensate for local breakdowns by going bey-
ond the normative procedures to make sure that
representational states propagate when and
where they should. The difficulty in mapping a
rule of thumb developed for solo watchstanding
into the group configuration highlights the dif-
ferences between these modes of operation and
provides insights into the limits on the team's
abilities to explicitly plan the coordination of
their actions.
924/1540
 Page 229

5 Communcation

Communication and Task Decompositions

In chapter 4 I argued that the bandwidth of com-

munication available to the members of the nav-
igation team would affect the computational
properties of the team as a cognitive system. I
will report some computer simulation results
that support this claim later in this chapter.
However, casting the phenomenon in terms of a
single quantitative measure, such as bandwidth,
obscures important properties that should be
discussed.
 926/1540

The fact that the navigation team distributes

computational procedures across a social organ-
ization raises the possibility that there may be
better and worse ways to arrange the distribu-
tion. One way in which the distribution of com-
putational procedures can be better or worse
concerns the relation between the kinds of struc-
ture that can be passed between computational
elements and the kinds of structure with which
the passed structures must be coordinated in the
performance of the task.
 927/1540

Consider again the task of the reconciliation of

the chart to the world as it occurs in Sea and An-
chor Detail. One person can see the chart and
another can see the world. They communicate
with one another on a telephone line. Achieving
a reconciliation between chart and world is a
very difficult task, and sometimes verbal com-
munication alone is not sufficient. It would be
easy to say that the telephone circuit provides
only a low bandwidth of communication and
that more information must be transmitted if the
pelorus operator is to locate the correct land-
mark in the world. The problem with that ac-
count is that it commits us to a measure of
quantity of information that may be both prac-
tically and theoretically impossible. On what
grounds could it be claimed that the recorder's
going to the wing and pointing to the landmark
in the presence of the pelorus operator has a
greater bandwidth than a lengthy verbal descrip-
tion? The process of assigning landmarks is
 928/1540

performed under time constraints, but it does not

appear that efforts to communicate the location
of a landmark are ever given up just because
time runs out. These efforts are abandoned be-
cause of a perceived
 Page 230

qualitative rather than quantitative deficiency.

Verbal descriptions typically fail not because
they don't provide enough structure but because
they provide the wrong kind of structure. The
difference between the right and the wrong
kinds of structure is determined by both the
nature of the task and the other structural re-
sources that are available.
 930/1540

The theory of computation by propagation of

representational state poses the question differ-
ently. It asks instead what kinds of structure the
pelorus operator must bring into coordination
with the communicated structure in order to per-
form the task. As described in chapter 3, when
the landmark description is spoken on the tele-
phone, the pelorus operator must coordinate the
spoken description of the landmark with know-
ledge of the landmark's appearance, which must
be coordinated with the visual field. The prob-
lem for the pelorus operator involves searching
the available visual field to find a scene that can
be construed as a match with the target descrip-
tion. Pointing isn't more information than a de-
tailed verbal description: it is a different kind of
information that can be put to work in a differ-
ent way.

Language Behavior as a Determinant of the

Cognitive Properties of Groups
 931/1540

Consider the following example in which the

structure of the lexicon constrains the cognitive
properties of the group: One evening, a Marine
commander on board the Palau, Major Rock,
telephoned the charthouse. Quartermaster Se-
cond Class Smith answered the phone. Major
Rock asked Smith what the phase of the moon
would be that night. Smith asked Chief
Richards, who was sitting nearby. Richards im-
mediately replied "Gibbous waning." Smith re-
layed the answer to Rock. Rock apparently did
not understand the answer, and he and Smith
talked past each other for several conversational
turns. Finally, Smith put his hand over the
mouthpiece and said "Chief, he says it's got to
be one of, 'new', 'first', 'full', and 'last'." Chief
Richards said "It's last." Smith told Rock, and
Rock hung up. Chief Richards had a rich vocab-
ulary for phases of the moon. Rock's vocabulary
was impoverished. Giving Rock the mapping in
terms of the richer vocabulary did no good
 932/1540

because he could not connect it to his simplified

notions. Rock eventually provided both the
question and the possible answers to it, the latter
in the form of the four categories he recognized.
Once the chief knew what Rock was looking
forknew the look of the map Rock
 Page 231

was trying to articulate with the moon phases-he

could settle on 'last' as the nearest match to 'gib-
bous waning'. After Smith hung up the phone,
Chief Richards said: "Rock is a great big guy
with a brain about this big (making a circle with
the tip of his index finger touching the first joint
of his thumb). He must never have taken an am-
phib mission onto a beach at night. He might get
by on a crescent moon, but on a gibbous moon
he'll be dead."
 934/1540

This example raises two points. First, the

amount of information that is conveyed by a
given utterance is not a simple function of the
volume of structure in the utterance. Information
and coding theory define the minimum band-
width required to encode a given set of alternat-
ive messages. From the perspective of informa-
tion theory, natural language is not an efficient
code. Suppose X bits are required to represent
the name of the present phase of the moon. The
amount of information in the message depends
on how many other phases of the moon can be
named, not on how many bits it took to repres-
ent the name of the phase of the moon. Second,
the expressiveness of the code may determine
the cognitive properties of the larger system.
Whether a team of planners can mount a suc-
cessful amphibious landing may depend on the
range of distinctions that can be made in the lan-
guage spoken by the mission planners.
 935/1540

Because so much of the communication within

this system is verbal communication, the proper-
ties of language become important determinants
of the nature of the computation that is accom-
plished. The properties of language change with
the register of the speech and with the medium
in which the utterances are carried. The man-
dated language on the intercom is almost tele-
graphic. This is adequate when the desired com-
munications have been anticipatedwhen the pos-
sible messages have been spelled out and agreed
upon in advance. However, it is difficult to ne-
gotiate a novel understanding of the nature of a
problem or to jointly interpret a complex world
on such a low-bandwidth channel.
 936/1540

Viewing language as one of the structured rep-

resentations produced and coordinated in the
performance of the task highlights the
information-bearing properties of language. In
cognitive science, language is usually thought of
primarily as a human computational capacity
that should be understood in terms of the pro-
cessing that individuals must do to produce or
interpret it. Shifting attention from the cognitive
properties of an individual to those of a system
of socially distributed cognition casts language
in
 Page 232

a new light. The properties of the language itself

interact with the properties of the communica-
tions technology in ways that affect the compu-
tational properties of the larger cognitive
system.
 938/1540

Linguistic determinism is the idea that the struc-

ture of one's native language determines proper-
ties of individual thought. This is not an appro-
priate place to review that literature, but the an-
swer to the question "Does the structure of lan-
guage determine the structure of thought?"
seems to be "Sometimes and sometimes not."
When so-called noncognitive tasks are organ-
ized in such a way that subjects can use the
structure of their language as a mediating re-
source in organizing task performance, then lan-
guage structures thought. When the structure of
language is not useful as a mediating resource in
task performance, then task performance does
not seem to be affected by the structure of lan-
guage. When cognitive activities are distributed
across social space, the language or languages
used by task performers to communicate are al-
most certain to serve as structuring resources,
and the structure of language will affect the cog-
nitive properties of the group even if they do not
 939/1540

affect the cognitive properties of individuals in

the group.

Communication in a Shared World

 940/1540

Communication between persons who are cop-

resent in a shared physical environment differs
in many ways from communication across a re-
stricted bandwidth medium. The Officer of the
Deck tends to trust the navigation plot main-
tained on the bridge better than he trusts the plot
generated in the Combat Information Center.
This is because the OOD can come to the chart
table, look at what the navigation team is doing,
and talk to the quartermasters. In this relatively
rich face-to-face interaction, an understanding
can be negotiated. The work that went into the
recommendation can be displayed and dis-
cussed. Such negotiation of meaning is difficult
when one is dealing with CIC via the phone
talkers. For example, at one point CIC advised
the bridge that the ship could continue on its
present course and speed for 4 hours before
reaching the boundary of the operations area.
The navigator had the conn and doubted this. He
went to the chart table and consulted with
 941/1540

Quartermaster Third Class Charles, who had just

computed a time of 75 minutes to the boundary
of the operations area on present course and
speed. By looking at the charted track and talk-
ing with Charles, the nav-
 Page 233

igator convinced himself that the position shown

there was correct. He told the phone talker to
tell CIC that they were off by 3 hours.

In another example, the weapons officer was un-

happy to have been assigned a general quarters
(GQ) station on the signal bridge rather than on
the navigation bridge. He wanted to have his
duty station changed to the navigation bridge
because he felt it would be easier to communic-
ate a complex set of options to the commanding
officer (CO) in face-to-face interaction than
over the intercom. He was also worried that the
CO would find the arguments of the department
heads who were co-present with him more com-
pelling than those that came to him over the
intercom.
 943/1540

The previous two examples concern intentional

communication in a shared world. A good deal
of our behavior has communicative function
without communicative intent. Segal (1990)
points out the importance of crew-station layout
for the unintended communication among mem-
bers of aircraft flight crews. The following ex-
ample illustrates this and other features of com-
munication in a shared world.

Recorder: Mark it.

SW: Pier 5, 1 1 7.

Recorder: 1 1 7, Pier 5.
 944/1540

[The recorder's echoing the bearing as reported

is (1) an acknowledgement that it has been re-
ceived, (2) a readback with content to permit
checking by the sender, and (3) the communica-
tion of the bearing to the plotter. Thus, it is part
of two conversations at once. The plotter, at this
point, is waiting for the first bearing so he can
begin work. The plotter begins plotting the pier
5 line of position with bearing 117.]

PW: Diving Tower, 2 5 0.

Recorder: 2 5 0, Diving Tower.

(At this point, the chief is still plotting the first

LOP bearing 117. He may be simultaneously
plotting 1 1 7 and subvocally rehearsing the
second bearing, 2 5 0.)

Helm: Steady 3 0 8, Sir. Checking 2 9 2.

CICComm: Steady course 3 0 8.

 945/1540

(There are other numbers being spoken in the

environment. In order to prevent interference
from these, the bearings are frequently sub-
vocally rehearsed.)
 Page 234

(The plotter is setting up the hoey to plot Diving

Tower with a bearing of 2 5 7, rather than the
correct 2 5 0. This may be a datadriven error due
to the interference of 117's being plotted while
250 was heard and possibly rehearsed.)

PW: Stanchion 18, 2 9 7 point 5.

Recorder: For 18, 2 9 7 point 5.

(The plotter aligns the hoey with charted symbol

of Dive Tower with 257 bearing. The LOP is
not near the expected position. The plotter leans
toward the recorder, then looks at the hoey to
read the value he has already aligned on it.)

Plotter: 2 5 7?

Recorder: 2 9 7.
 947/1540

(The plotter leans further toward the recorder,

looks in the bearing record log, and points to the
Dive Tower column heading. ) Plotter: Hm, um,
no.

Recorder: 2 5 0.

(The plotter stands upright and moves the hoey

arm.)

Plotter: OK, I might believe that.

CON: Engine ahead two-thirds.

Leehelm: Engine ahead two-thirds, aye.

(The QMOW has his hand on the chart in the

way of the plotting tool. The plotter whacks it
with the hoey while aligning it for the LOP.)

QMOW: It hurts, Chief.

(The plotter plots the LOP for Diving Tower.)

 948/1540

Plotter: Get your hand out of the way.

QMOW: Yes, sir.

Plotter: And what?

Recorder: 2 9 7 point 5.

(The plotter aligns the hoey and plots the last

LOP.)
 949/1540

In this example the bearings are propagated

bottom-up from the alidades to the chart.
Spoken representations of the bearings 297.5
and 117 are in the environment of the hearing of
250 and seem to interfere with its processing;
250 turns into 257 in this environment. It is not
possible to know which, if either, of these other
signals interfered, but the transformation was
made. The momentary breakdown in commu-
nication is, in part, a consequence of the prop-
erty of the spoken medium. Since speech is eph-
emeral, one
 Page 235

must attend to it as it is being produced. The

lack of a way to buffer this input imposes a need
for temporal coordination between the plotting
activity and the delivery of bearing information.
Since the plotter is still engaged in plotting the
previous line of position, his attention to the
spoken bearing is incomplete. Whatever the
cause of the transformation of the form of the
bearing for Dive Tower, when propagated to the
chart it does not ''work." The line of position
does not fall where expected.
 951/1540

The plotter asks for confirmation of the bearing.

Was it indeed 257? In asking this question, the
plotter refers to the hoey scale, constructing part
of his query by reading the bearing he has plot-
ted. The recorder attempts to match the request
to what he knows to have been the previous
communications. This is a negotiation of the
meaning of the question. The recorder must de-
termine which LOP is being asked about.
 952/1540

The recorder responds "297." He has matched

the plotter's query to one of the reported bear-
ings (297.5) and, in the process, has rounded off
the reported bearing. The plotter leans over to
look in the log and rejects the 297 bearing. He
may know it is not correct because it is even
bigger than 257, and 257 was already too big to
work. The properties of the media are involved
here again. The written record endures, so the
plotter directs his attention there to answer his
own question. Before the plotter is even finished
rejecting the 297 bearing, the recorder has real-
ized that the plotter wants the bearing to the
Dive Tower, not the bearing to stanchion 18.
The plotter has also gestured toward the column
in the bearing record log labeled Dive Tower.
This gesture may be involved in two activities
here. First, it is part of the plotter's procedure for
looking up the bearing. It is a bit of structure
that the plotter creates in his world to guide his
looking in the bearing record log. It is a way to
 953/1540

control the allocation of his own visual atten-

tion. Second, it is simultaneously a clue for the
recorder about which bearing it is that the plot-
ter is looking for. The communicative function
of the gesture is opportunistic. It does not seem
to be intended. The recorder now says "250."
The plotter makes a quick adjustment to the
hoey and sees that a 250 bearing works well.
The plotter closes this negotiation by saying
"OK, I might believe that."

The fix itself is completed when in response to

the plotter's asking "And what?" the recorder re-
sponds with the last of the three bearings, 297.5.
It is interesting that it is not now rounded off, as
it was in response to the question of 257. Then it
was tailored to fit
 Page 236

the query; now that the confusion has been re-

solved, it is reported as it was recorded. The ut-
terance "And what?" is interpretable only in the
context of an understanding of the task being
performed and the place of the plotter in the ex-
ecution of the task. The recorder is able to de-
termine that the plotter intends to plot the last
bearing. He responds in a way that assumes that
"And what?" refers to the one bearing that re-
mains to be plotted. The plotter's continued plot-
ting activityarriving at a satisfactory fixis evid-
ence that the recorder's interpretation and re-
sponse were correct.
 955/1540

The plotter's use of his finger in locating the

bearing in the bearing record log is very inter-
esting. Because the bearing log is a memory for
the observed bearings in this distributed cognit-
ive system, the plotter's action is part of a
memory-retrieval event that is internal to the
system but directly observable. From the per-
spective of the individual, the technology in use
here externalizes certain cognitive processes.
This permits some aspects of those processes to
be observed by other members of the team. Be-
cause of this, the chief's pointing can be both a
part of his private cognitive processing and an
element of communication to the recorder about
the sort of thing the chief is trying to accom-
plish. Some kinds of media support this sort of
externalization of function better than others.
The existence of a gesture that has both private
and public functions suggests that other commu-
nicative features may also have these two roles.
I suspect that prosody might have a similar dual
 956/1540

role in the production of verbal representations,

helping the speaker to shape the allocation of his
own attention while simultaneously providing
the listener with structure that can be used to de-
termine what the speaker is trying to
accomplish.
 957/1540

The above example shows clearly that the norm-

ative description of information flow in the fix
cycle, which maintains that information flows
bottom-up from the alidades to the chart, is
wrong. Far from being a simple one-way traject-
ory for information, the communication is in
fact the bringing together of many kinds of con-
straints in both bottom-up and top-down direc-
tions. The meanings of statements and questions
are not given in the statements themselves but
are negotiated by the participants in the context
of their understandings of the activities under-
way. The participants use guesses about one
another's tasks to resolve ambiguities in commu-
nication. Particular meaningful interpretations
for statements are simultaneously proposed and
presupposed by the courses of
 Page 237

action that follow them. The evidence that each

participant has of successful communication is
the flow of joint activity itself.

The Negotiation of Meaning in Interaction

The following exchange, which is full of break-

downs and repairs, illustrates the negotiation of
meaning in interactions.

Recorder: Stand by to mark. Time 0 ... 0 1 2.

CIC Talker: Combat holds 3100 yards to the

turn. Holds 50 yards left of track.

Recorder: 26 feet under the keel. Mark it.

(After a 13 second pause, the plotter watches the

recorder write down the first bearing, sets the
hoey, and aligns it with the chart.)

Plotter: What's at 2 7 7? (plotting)

 959/1540

Recorder: 2 7 1.

(The recorder's correction may be based on a re-

reading of the log entry or on memory of the ac-
tual bearing reported. Notice the properties of
the medium in which the bearing is represen-
tedthere is a greater potential confusion of 1 and
7 in written form than "one" and "seven" in
spoken form.)

Plotter: What is it?

(The plotter uses the positions of landmarks on

the chart, the projected position of the ship, and
the angle on the hoey in evaluating the bearing.
This one does not fit any of the landmarks he
expects.)

Recorder: Um, carrier tower.

Plotter: Oh, I don't want that thing. It's OK, go

ahead.
 960/1540

(The chief does not want this landmark because

its position is not yet established. It is a back-
plot. Once they have established several of their
positions with known landmarks, they can back-
plot the position of the tower and make it a us-
able landmark in the future.)

Recorder: The aero beacon is 2 9 2. 10th Aven-

ue Terminal is 1 0 5. (then speaking into the
phone circuit) Give me the last bearing you
took.

Plotter: 1 0 5. (The plotter is reading this from

the bearing record log.)

Plotter: Is that 10th Avenue Terminal that is 1 0

5? Or that little pier?
 Page 238

Recorder: 0 5 9 is that little pier. This one is 0 5

9 (pointing to the depiction of the pier on the
chart).

(The referent of "this one" in the recorder's

statement is determined indexically by his ori-
entation to the symbols on the chart.)

Plotter: OK. What's at 1 0 5?

Recorder: 10th Avenue is at 0 5, 1 0 5.

Plotter: Still outside here.

Recorder: Still outside?

Plotter: What's the other one?

(The notion of "the other one" relies on the

knowledge that three lines of position are re-
quired for the fix. The recorder seems to have a
problem with the back-plot in the procedure.)
 962/1540

Recorder: 2 7 1.

Plotter: No, no, that's back-plot.

Recorder: Carrier tower.

Recorder: 2 9 2.

Plotter: Yeah, yeah. What's that?

Recorder: Aero beacon, OK, 1 0 5.

Plotter: What's that?

Recorder: That's 10th Avenue.

Plotter: Oh, it is huh.... is he use ... He must be

using the tip of it. (8 seconds) Well, it's almost
... actually nothing.
 963/1540

In this passage, the recorder and the plotter at-

tempt to communicate a set of landmark-to-
bearing correspondences. The meanings of ut-
terances are established through reference to the
chart itself, the structure of the task, the relation
of the structure of the hoey to the structure of
the chart, and previous elements of the ex-
change. That is, all these structures are brought
into coordination at once in the performance of
the task.
 964/1540

Meanings can only even be imagined to be in

the messages when the environment about
which communication is performed is very
stable and there are very strong constraints on
the expectations. In many endeavors, creating
and maintaining the illusion that meanings
reside in messages requires that a great deal of
effort be put into controlling the environment in
which communication takes place. Meanings
seem to be in the messages only when the struc-
tures with which the message must be brought
into coordination are already reliably in place
and taken for granted. The illusion
 Page 239

of meaning in the message is a hard-won social

and cultural accomplishment.

Confirmation Bias in Individuals and Groups

I have argued above that the cognitive properties

of a group may depend as much on the system
of communication between individuals as on the
cognitive properties of the individuals them-
selves. It is one thing to assert such an effect and
another to demonstrate it. Though I find the ex-
amples from actual interactions compelling,
events in the real world are almost always com-
plicated by unwanted interactions. Fortunately, a
different kind of demonstration is also possible.
In the following pages I will describe a com-
puter simulation that explores the role of com-
munication in the production of the cognitive
properties of a group.
 966/1540

To test the notion that the cognitive properties

of groups may differ from those of the individu-
als who constitute the group, it is necessary to
focus on some particular cognitive property that
is generally agreed to be a property of individual
cognition and then develop some way to show
that whether that property is manifested by a
group depends on the social organization of the
group. For the purposes of this study, I will use
the phenomenon known as confirmation bias.

Confirmation Bias in Formation of

Interpretations
 967/1540

Confirmation bias is a propensity to affirm prior

interpretations and to discount, ignore, or rein-
terpret evidence that runs counter to an already-
formed interpretation. It is a bias to confirm an
already-held hypothesis about the nature of the
world. This is a commonsense notion. We talk
about the difficulty of changing someone's mind
once it is "made up." The importance of "first
impressions" is an obvious corollary of our folk
belief in this principle. There is also compelling
scientific evidence of the generality of confirm-
ation bias across such areas as attribution, per-
sonality traits (Hastie and Kumar 1979), logical
inference tasks (Wason 1968; Wason and
Johnson-Laird 1972), beliefs about important
social issues (Lord et al. 1979), and scientific
reasoning (Fleck 1979; Tweney et al. 1981).

To the extent that this propensity to stick with

prior interpretations and discount disconfirming
evidence often leads us to
968/1540
 Page 240

maintain faulty interpretations of the nature of

the world, it seems maladaptive. After all,
knowing what is going on in the environment is
an important ability for any creature, and, in
general, the more complex the creature, the
more complex is that creature's sense of what is
in the environment. A property of cognitive pro-
cessing that prevents us complex creatures from
finding better interpretations once we have a
good one seems very maladaptive indeed. Why,
then, would such a property survive? Clearly
there must be a tradeoff here between the ability
to move from one interpretation to a better one
and the need to have an interpretationany inter-
pretationin order to coordinate with events in the
environment. A system that maintains a coher-
ent but suboptimal interpretation may be better
able to adapt than a system that tears its inter-
pretations apart as fast as it builds them.
 970/1540

This propensity is widely accepted as a general

feature of individual cognition. If it represents a
sometimes infelicitous tradeoff between keeping
a poor interpretation and having no interpreta-
tion at all, one wonders if it might not be pos-
sible for a group of individuals, each of whom
has this propensity, to make a different sort of
trade-off. That is, might a group be organized in
such a way that it is more likely than any indi-
vidual alone to arrive at the best of several pos-
sible interpretations, or to reject a coherent in-
terpretation when a better one is present? The
plan of the remainder of this chapter is to accept
confirmation bias as a property of individual
cognition and then to ask what properties it
might produce in systems of socially distributed
cognition. What I hope to show is that the con-
sequences of this property of individual cogni-
tion for the cognitive capabilities of groups of
humans depends almost entirely upon how the
group distributes the tasks of cognition among
 971/1540

its members. That is, some ways of organizing

people around thinking tasks will lead to an ex-
acerbation of the maladaptive aspects of this
property of mental systems, whereas other forms
of organization will actually make an adaptive
virtue on the group level of what appears to be
an individual vice.

Interpretation Formation as Constraint

Satisfaction

Many important human activities are conducted

by systems in which multiple actors attempt to
form coherent interpretations of some set of
phenomena. Some of these systems are small,
composed of only a few individuals, while oth-
ers are very large indeed.
 Page 241

The operation of a complex system is often ac-

complished by a team. A shift of operators at a
nuclear power plant, an aircraft flight crew, or
the bridge team on a large ship is a small system
in which multiple individuals strive to maintain
an interpretation of the situation at hand. The
complexity of a system may make it impossible
for a single individual to integrate all the re-
quired information, or the several members of
the group may be present because of other task
demands but may be involved in distributed in-
terpretation formation. Management teams in
business and government are also systems of
distributed interpretation formation, as are juries
in the court system. A community of scientists
may be the best example of a very-large-scale
system in which a group strives to construct a
coherent interpretation of phenomena.
 973/1540

Forming an interpretation is an instance of what

computer scientists call a constraint-satisfaction
problem. Any coherent interpretation consists of
a number of parts; call them hypotheses. Some
of the parts go together with others or support
one another; others exclude or inhibit one anoth-
er. These relationships among the parts of the
interpretation are called constraints. Consider
the following incident taken from Perrow's book
Normal Accidents (1984):
 974/1540

On a beautiful night in October 1978, in the

Chesapeake Bay, two vessels sighted one anoth-
er visually and on radar. On one of them, the
Coast Guard cutter training vessel Cuyahoga,
the captain (a chief warrant officer) saw the
other ship up ahead as a small object on the
radar, and visually he saw two lights, indicating
that it was proceeding in the same direction as
his own ship. He thought it possibly was a fish-
ing vessel. The first mate saw the lights, but saw
three, and estimated (correctly) that it was a
ship proceeding toward them. He had no re-
sponsibility to inform the captain, nor did he
think he needed to. Since the two ships drew to-
gether so rapidly, the captain decided that it
must be a very slow fishing boat that he was
about to overtake. This reinforced his incorrect
interpretation. The lookout knew the captain
was aware of the ship, so did not comment fur-
ther as it got quite close and seemed to be
nearly on a collision course. Since both ships
 975/1540

were traveling full speed, the closing came fast.

The other ship, a large cargo ship did not estab-
lish any bridge-to-bridge communication, be-
cause the passing was routine. But at the last
moment, the captain of the Cuyahoga realized
that in overtaking the supposed fishing boat,
which he assumed was on a near parallel
course, he would cut off that boat's
 Page 242

ability to turn as both of them approached the

Potomac River. So he ordered a turn to the port.
 977/1540

The two ships collided, killing 11 sailors on the

Coast Guard vessel. The captain's interpretation
contained a number of hypotheses (that the other
ship was small, that it was slow, and that it was
traveling in the same direction as his own ship).
These hypotheses were linked to a set of obser-
vations (the ship presented a small image on the
radar; it appeared to the captain to show two
lights; the distance between the ships was clos-
ing rapidly) to form a coherent interpretation in
which the hypotheses were consistent with one
another and with the observations. Several of
the hypotheses of the mate's interpretation were
in direct conflict with some of the hypotheses in
the captain's interpretation. For example, the hy-
pothesis that the ships were meeting head on
and the hypothesis that one was overtaking the
other were mutually exclusive.
 978/1540

A good interpretation is one that is both intern-

ally consistent and in agreement with the avail-
able data. Evidence from the world makes some
of the hypotheses of the interpretation more or
less likely. These hypotheses that are directly
driven by evidence have constraining relations
to other hypotheses for which there is, perhaps,
no direct evidence. For example, in the ship col-
lision described above, there is no direct evid-
ence concerning the speed of the other vessel.
That hypothesis is derived from the hypothesis
that the Coast Guard ship is overtaking the other
ship and the observation that the distance
between the ships is closing rapidly. If those two
things are true, then the other ship must be mov-
ing slowly. The job of forming an interpretation
can thus be seen as attempting to assign likeli-
hoods to the various hypotheses in such a way
that the constraints among the hypotheses and
between the hypotheses and the evidence in the
world are as well satisfied as is possible.
 979/1540

The project at hand is to develop a framework

for describing these situations and the factors
that control the cognitive properties of these so-
cially distributed systems. What is needed is an
abstraction that is pertinent to the phenomena
and that captures the similarities among a num-
ber of classes of distributed interpretation form-
ation in spite of the diversity of details out of
which they are composed. The desired account
should explicitly address the issue of the forma-
tion of interpretations and the ways in which in-
terpretations can be influenced by evidence
from the environment as
 Page 243

well as by evidence communicated by other act-

ors in the setting. It should allow us to look at
what is going on inside individuals and also
what is going on among them. It should allow us
to characterize both the properties of individuals
and the properties of systems composed of sev-
eral individuals.
 981/1540

All these goals can be met by a computer simu-

lation. And while a simulation can permit us to
explore effects of communication in aworld that
is free of unwanted interactions with uncon-
trolled surrounding events, it has limitations that
must be acknowledged in advance. The princip-
al shortcoming of all simulations is that they are,
of necessity, extreme simplifications of the phe-
nomena they are intended to model. In the
present case, many of the important facts about
real communication in human systems are not
represented at all. Questions of indexicality of
reference and of negotiation of meaning are
much too complex to be modeled by simple
simulations and thus are not represented in the
simulation presented here. In spite of these lim-
itations, I believe that the simulation model does
make clear a number of issues that might other-
wise be obscured.

Constraint-Satisfaction Networks
 982/1540

A particular kind of connectionist network

called a constraintsatisfaction network provides
a rough model of individual interpretation form-
ation. Rumelhart et al. (1986) define a con-
straintsatisfaction network as follows:
 983/1540

a network in which each unit represents a hypo-

thesis of some sort (e.g., that a certain semantic
feature, visual feature, or acoustic feature is
present in the input) and in which each connec-
tion represents constraints among the hypo-
theses. Thus, for example, if feature B is expec-
ted to be present whenever feature A is, there
should be a positive connection from the unit
corresponding to the hypothesis that A is
present to the unit representing the hypothesis
that B is present. Similarly, if there is a con-
straint that whenever A is present B is expected
not to be present, there should be a negative
connection from A to B. If the constraints are
weak, the weights should be small. If the con-
straints are strong, the then the weights should
be large. Similarly, the inputs to such a network
can also be thought of as constraints. A positive
input to a particular unit means that there is
evidence from the outside that the relevant
984/1540
 Cognition p244

Figure 5.1 A simple constraint-satisfaction network compos

units. Each unit represents a hypothesis, and the connections
units represent supportive or competitive relations among th
theses represented by the units. In this network, units 1, 2, an
port one another and units 4, 5, and 6 support one another. Th
each of these clusters inhibit the units in the other cluster. The
 986/1540

has two coherent interpretations: one in which units 1-3 are

while units 4-6 are inactive and one in which units 4-6 are ac
units 1-3 are inactive. This figure does not show the pattern o
tion across the units. Evidence in the environment that suppo
ticular hypothesis is implemented as the addition of activatio
unit that represents that hypothesis. This network receives e
that supports the hypothesis represented by unit 4.

feature is present. A negative input means that ther

evidence from the outside that the feature is not pre

With each unit adjusting its activation (likelihood o

true) on the basis of the activations of its neighbors
strengths of the connections to those neighbors, suc
work will eventually settle into a state in which as m
the constraints as is possible will be satisfied.
 987/1540

Imagine a network in which there are two clusters o

(figure 5.1). Among the units within each cluster th
positive connections. Thus, each cluster of units rep
a set of hypotheses that are consistent with one ano
All the connections that go from a unit in one cluste
unit in the other cluster are negative. This means th
hypotheses represented by the units of one cluster a
consistent with the hypotheses represented by the u
the other cluster. When such a network tries to satis
many constraints as it can among the hypotheses, it
end up with all the units in one cluster highly active
the units in
 Page 245

the other cluster inactive. That is, it will arrive at

an interpretation in which one set of hypotheses
is considered true and the other is considered
false. Once it has arrived at such a state, the net-
work will be very insensitive to evidence that
contradicts the interpretation already formed.
Notice that there are two kinds of patterns here:
the pattern of interconnections among the units
and the pattern of activation across the units. An
interpretation of an event is a particular pattern
of activation across the unitsfor example, the
state in which all the units in the left cluster are
active and all the units in the right cluster are in-
active. The stable interpretations of the network
are determined by the pattern of interconnectiv-
ity of the units. In this case the connection
strengths have been carefully arranged so that
there will be just two stable interpretations.
 989/1540

In the collision scenario presented above, the

captain and the mate share the schema for inter-
preting the motion of other ships on the water.
Because of the conventions for lighting ships at
night, seeing two lights supports the hypothesis
that one is viewing the stern of the other ship
whereas seeing three supports the hypothesis
that one is meeting it head on. If it is an overtak-
ing situation, a rapid closing of the distance to
the other ship supports the hypothesis that it is
going slowly. The mate would doubtless en-
dorse these constraints among hypotheses. But
he reached a different interpretation because he
''saw" different evidence than the captain saw. In
the model, sharing the schema is sharing the pat-
tern of connections among the units. Sharing the
interpretation is having the same pattern of ac-
tivation across the units.
 990/1540

For the individual whose constraint-satisfaction

network is represented in figure 5.1, an inter-
pretation is a pattern of activation across the six
units of the network. The space of possible in-
terpretations for this network is thus a six-di-
mensional space. The locations of the two good
interpretations in this space are known to be
{111000} (left cluster active, right cluster not)
and {000111} (right cluster active, left cluster
not). Unfortunately, it is very difficult to think
about events in six dimensions. Fortunately, this
six-dimensional space can be mapped into two-
dimensional space. Since it is possible to com-
pute the Euclidean distance of any pattern of ac-
tivation from the patterns of activation of the
two good interpretations, it is possible to build a
new, two-dimensional space in which the two
dimensions are distance from interpretation 1
and distance from interpretation 2. Thus, the
location of any pattern of activation can be plot-
ted in this interpretation space in terms of its
991/1540
Page 246
993/1540
 994/1540

Cognition p246a
 995/1540

Cognition p246b

Figure 5.2 The trajectories of a single network in inter-

pretation space. Here the interpretation space is plotted in
two dimensions. The x axis shows the distance of the cur-
rent interpretation from interpretation 1: The y axis indic-
ates the distance of the current interpretation from inter-
pretation 2. Thus, for example, the lower right comer is as
close to interpretation 2 as is possible and as far from in-
terpretation 1 as is possible. It therefore represents the loc-
ation of interpretation 2. The diagonal line is everywhere
 996/1540

equidistant between the two interpretations. The track of

the network is shown in time. Its position is plotted at
equal time intervals, so the distances between the plotted
positions is proportional to the speed at which the
network's interpretation is changing. The network shown
in panel a began slightly nearer to interpretation 1 than to
interpretation 2. It began moving slowly at first away
from both interpretations, then it turned toward interpreta-
tion 1 and picked up speed. Finally, it made a slow ap-
proach to interpretation 1. The same network is shown in
panel b, but this time it has evidence from the environ-
ment in favor of interpretation 2. Even though it began
slightly predisposed to interpretation 1, it was swayed by
the evidence from the environment and went to interpreta-
tion 2.
 997/1540

nearness to the two good interpretations. For ex-

ample, the pattern of activation {0.5 0.5 0.5 0.5
0.5 0.5} in which every unit is neither active nor
inactive, is equally distant from the two good in-
terpretations. The plotting of positions in inter-
pretation space is of no computational con-
sequence; it simply makes the motion of the net-
works visible. Note also that not all locations in
the space are possible and that the fact that two
networks are at the same location in the space
does not mean that they have the same pattern of
activation. Figure 5.2a shows the trajectory in
interpretation space of an individual who starts
out about halfway between the two interpreta-
tions. In this individual all the units have ap-
proximately equal levels of activation. This net-
work is not strongly committed to either of the
interpretations. When it begins passing activa-
tion among its units, it moves toward one inter-
pretation and away from the other.
998/1540
 Page 247

Constraint networks can also receive input from

the environment. This corresponds to direct
evidence for one of the hypotheses. Input from
the environment is implemented as the addition
or subtraction of activation to a single unit (fig-
ure 5.1). Thus, if the individual represented in
figure 5.2a is taken back to its starting point and
given evidence that one of the hypotheses con-
sistent with interpretation 2 is true, it will follow
the trajectory shown in figure 5.2b. This simply
demonstrates that a network that has not yet
formed a strong interpretation can be influenced
by evidence from the environment. If, however,
the network had already arrived at interpretation
1, evidence for interpretation 2 would have little
effect on the individual.
 1000/1540

Three variables determine the behavior of the

isolated individual. The first is the pattern of in-
terconnectivity of its units. This is the network's
schema of the phenomena about which the inter-
pretation is formed. The second is the initial pat-
tern of activation across its units. This consists
of the network's preconceptions about the state
of affairs in the world, and t can be seen in the
trajectories of networks in interpretation space
as the point at which a network starts. The third
variable consists of the external inputs to partic-
ular units of the network; it represents the evid-
ence directly in favor of or against particular hy-
potheses that are parts of the interpretations. The
trajectories shown in figure 5.2 demonstrate the
effects of these variables.

COMMUNITIES OF NETWORKS
 1001/1540

The behavior of a single constraint-satisfaction

network mimics, in a rough way, the phenomen-
on of confirmation bias as it is observed in indi-
vidual human actors. Even more complicated
versions of these networks could provide more
accurate models of confirmation bias. The
simple networks presented here are close
enough to serve our present purpose, which is to
find a simulation that allows us to explore the
relationships among properties of individuals
and properties of groups. In order to make such
an exploration, I have created and examined the
behavior of communities of networks. This may
not seem to be the most obvious strategy to pur-
sue. Since the processing in connectionist net-
works is distributed across units in a network,
and the processing in a system of socially dis-
tributed cognition is distributed across a number
of people, there is a strong temptation to adopt a
superficial mapping between the two domains in
 1002/1540

which units in a network are seen as correspond-

ing to
individuals and the connections among units are see
dividuals. In this way, a single network would be ta
lems with this mapping. Let me simply issue the wa
dead end, and suggest instead that the real value of
of cognition will come from a more complicated an
works or assemblies of networks, and in which syst
communities of networks. The latter approach is the

PARAMETERS OF THE MODELS

What happens in a system in which there are two or

form an interpretation? A system composed of two
not present in a single network. Three of these para
state across the individual members of a community
tion among the networks in the community. Table 5
of real communities to which, for the purposes of th

Table 5.1 The principal parameters of the simulatio

cessing they are intended to represent.
IN THE MODEL
 1004/1540

Distributions of properties of individual nets

Pattern of interconnectivity among units in the net

External inputs to particular units in the net

Initial pattern of activation across units in the net

Parameters that characterize communication among nets

Pattern of interconnections among the nets in the community

Pattern of interconnectivity among the units of communicatin

Strengths of connections between nets that communicate

Time course of communication

 Page 249

Distributions of Individual Properties

 1006/1540

The pattern of connectivity among the units

within a network defines the schema for the
event to be interpreted. Thus, the first additional
consideration is the distribution of event
schemata across the members of the community.
Clearly a system in which all the networks have
a consensus about the underlying structure of
the domain of interpretation is different from
one in which different networks have different
patterns of constraint among the hypotheses. As
a simplifying assumption, I have assumed that
all the networks have the same underlying con-
straint structure. This is simply an implementa-
tion of the ethnographer's fantasy that all the in-
dividuals in a culture have the same schemata
for the events to be interpreted (Boster 1990). In
the ship-collision situation it appears that the
Captain and the mate shared the schemata for
interpreting the motion of other ships. Both un-
derstood that seeing two lights was evidence
that one was viewing the stern of the other ship
 1007/1540

and that seeing three lights indicated that one

was meeting it head on. But the mate thought he
saw three lights and the captain thought he saw
two. Thus, consensus on schemata is not the
same thing as consensus about the interpreta-
tions of events. Two individuals could have the
same schema for some phenomenon and still
reach different interpretations of events if their
assessment of the evidence led them to instanti-
ate the schemata in different ways.
 1008/1540

The networks may receive inputs directly from

an environment. The distribution of access to
environmental evidence is an important structur-
al property of a community of networks. If all
networks in the community have the same un-
derlying patterns of constraints among hypo-
theses and all receive input from the same fea-
tures of the environment, then all networks in
the community will arrive at the same interpret-
ation. If different networks have access to dif-
ferent inputs from the environment, then they
may move to very different interpretations of the
world. It turns out that in the ship-collision situ-
ation the captain had some difficulty with his
eyesight.
 1009/1540

At any moment, the pattern of activity across the

units in a particular network represents the cur-
rent state of belief of that network. A coherent
interpretation is a pattern of activation that satis-
fies the constraints of the connections among
units. When a community of networks is cre-
ated, it may be created such that different net-
works have different patterns of activation.
Thus, the third parameter concerns the distribu-
tion of predispositions across
 Page 250

the networks in the community. The initial ac-

tivations in these simulations are always low;
that is, the individuals do not start with strong
beliefs about the truth of any of the hypotheses.

Communication Parameters
 1011/1540

In such a system, at least four additional para-

meters that describe the the communication
between the networks must be considered. For
the sake of simplicity I have modeled the com-
munication between networks as external inputs
applied directly to the units in each network.
(This simplification ignores the fact that com-
munication is always mediated by artifactual
structure. In Hutchins 1991 I modeled this expli-
citly in another community of networks with a
different network architecture. If a particular
node in one network is, say, highly active, then
some fraction of that activity may be applied as
an external input to the corresponding node in
some other network. Thus, in this model, com-
munication between individual networks is rep-
resented by direct communication of the activa-
tion levels of units in one network to units in an-
other network (figure 5.3). This is based on the
assumption that real communication about belief
in a hypothesis from one individual to another
 1012/1540

should have the effect of making the activation

level of the hypothesis in the listener more like
the activation level of the hypothesis in the
speaker. This is the most problematic simplifica-
tion in the system.

A fourth parameter describes the pattern of in-

terconnections among the networks in the com-
munity. This corresponds to the patterns of com-
munication links in a community. Each particu-
lar network in the community may communicate
with some subset of the other networks in the
community.
 1013/1540

Each network that communicates with another

does so by passing activation from some of its
own units to the corresponding units in the other
network. The pattern of interconnectivity among
the units of communicating networks determines
which of the units of each network pass activa-
tion to their corresponding numbers in the other
networks. This corresponds to a determination
of what the networks can talk to one another
about. It could be thought of as a limitation on
vocabulary that permits the networks to ex-
change information about only some of the hy-
potheses that participate in the interpretations.

Recall that a network passes only a fraction of

the activity of its unit to the corresponding unit
in another network. This fraction of
Figure 5.3 Communication among the individual networks. T
of the four communication parameters. In terms of the patt
among the networks (who talks to whom), individual 1 talks
als: individual 2 talks only with individual 1: individual 3 an
each other and with individual 1. In terms of the pattern of int
units of communicating nets (what they talk about), individu
about the most (about hypotheses 2, 4, and 6): individual 3 a
about hypothesis 4; and individual 2 and individual 4 don't
terms of the strengths of connections between nets that comm
 1015/1540

they are), individual 1 and individual 2 find each other's argu

and 6 very persuasive. The last communication parameter, th
nication (when they communicate), is not represented in this
of parameters shown in this figure is intended only to illustr
communicative patterns within a group. Subsequent simulat
the cognitive consequences of a variety of p

the activity of a unit in one network that is applied

to the corresponding unit in the other network may
suasiveness of the source. It determines how impor
agree with its corresponding unit in the other netwo
portance of satisfying the constraints imposed by ot
network.
 1016/1540

The final community-level parameter is the time co

tion. This refers to the temporal pattern of the excha
puts between the networks. This can vary from con
external inputs to no communication at all. In betw
are an infinite number of patterns of connection and
Again, it would be possible to have a different time
nication for every connection between the networks
the persuasiveness of each connection be a function
city, however, this too will be a global parameter, w
either on or off at whatever strength they have been
point in time.
 Page 252

Social Organization and the Cognitive Proper-

ties of Groups

With these simulation pieces, an exploration of

the relationships among the properties of the in-
dividuals and of the group with respect to con-
firmation bias can be made.

THE COMMONSENSE ARCHITECTURE OF

GROUP INTELLIGENCE
 1018/1540

It is often assumed that the best way to improve

the performance of a group is to improve the
communication among the members of the
group, or, conversely, that what is lacking in
groups is communication. In his 1982 novel
Foundation's Edge, Isaac Asimov describes a
world, Gaia, that is a thinly disguised Earth in
the distant future. Whereas James Lovelock's
original concept of Gaia referred only to the no-
tion that Earth's entire biosphere could be taken
to be a single self-regulating organism, Asimov
extended the concept to the cognitive realm. On
Asimov's Gaia, every conscious being is in
continuous high-bandwidth communication with
every other. There is but one mind on Gaia. In
Asimov's book it is a very powerful mind, one
that can do things that are beyond the capabilit-
ies of any individual mind. Is this really an ad-
visable way to organize all that cognitive
horsepower? Our simulations provide us with a
means to answer this question. They indicate
 1019/1540

that more communication is not always in prin-

ciple better than less. Under some conditions,
increasing the richness of communication may
result in undesirable properties at the group
level.
 1020/1540

Consider a simulation experiment in which only

the persuasiveness of the communication among
networks is varied. (Recall that this is imple-
mented by changing the strength of the connec-
tions between units in one network and corres-
ponding units in other networks.) In the initial
community of networks, all the networks have
the same underlying constraint structure, and all
have the same access to environmental evid-
ence, but each has a slightly different initial pat-
tern of activation than any of the others. Further-
more, all the networks communicate with one
another, all the units in each network are con-
nected to all the units in the other networks, and
the communication is continuous. This can be
regarded as a model of mass mental telepathy.
Under these conditions, when the communica-
tion connection strength (persuasiveness) is
zero, the networks do not communicate at all,
and each settles into an interpretation that is de-
termined by its initial predispositions (figure
 1021/1540

5.4a). If the community is started again, this

time with a nonzero persuasiveness, each indi-
vidual network
Page 253
1023/1540
 1024/1540

Cognition p253
 1025/1540

Cognition p253b
1026/1540
 1027/1540

Cognition p253c
 1028/1540

Cognition p253d

Figure 5.4 The trajectories of six individuals in interpreta-

tion space. Each panel in this figure represents the same
six individuals starting in the same locations. The persuas-
iveness of the communication among them is manipu-
lated; all other parameters are held constant. In panel a,
the persuasiveness is 0. The individuals are not commu-
nicating with one another, and each goes to the interpreta-
tion that is nearest its initial position in interpretation
 1029/1540

space. In panel b, with some communication among the

individuals, the individual who started most nearly
equidistant from the two interpretations has been influ-
enced by the three individuals who favor interpretation 2
and has followed them to that interpretation. In panel c, at
a higher level of communication, the unsure individual
has been captured by the two individuals who favor inter-
pretation 1 and has followed them there. In panel d, with
the persuasiveness set even higher, those individuals who
start nearer to interpretation 1 begin to move toward that
interpretation, but as those moving toward interpretation 2
become surer of their own interpretations the persuasive
communication with the others draws them also towards
interpretation 2. Under conditions of high persuasiveness
the system shows a confirmation bias similar to that of the
individual.
Page 254
 1031/1540

moves toward the interpretation that it had

moved to in the absence of communication, but
now it does so more quickly. If the community
is restarted again and again, with the persuasive-
ness increased each time, the velocity of the net-
works in conceptual space increases even more.
They hurry in groups to the available interpreta-
tions (some to one, some to the other), and once
there, they respond only a little to additional
evidence from the environment. Figure 5.4b il-
lustrates such a state in which one of the net-
works has changed the interpretation it arrives at
as a consequence of the influence of the other
networks on it. As the inter-network connection
strength is increased even more, the undecided
individual is drawn back to its original interpret-
ation (figure 5.4c). With the persuasiveness
turned up even more, the networks that started
out toward interpretation 1 are drawn back by
the emerging consensus, and all of them rush to
the same interpretation. Having arrived at that
 1032/1540

interpretation, they remain there, absolutely un-

moved by any amount of evidence from the en-
vironment (figure 5.4d). At high levels of per-
suasiveness, this system thus manifests a much
more extreme form of confirmation bias than
any individual alone. In retrospect, it is easy to
see why. When the level of communication is
high enough, a community of such networks
that receive similar inputs from the world and
that start near one another in the interpretation
space behave as one large network. Wherever
the networks go in interpretation space, they go
hand in hand and stay close together. Because
they are in continual communication, there is no
opportunity for any of them to form an interpret-
ation that differs much from that of the others.
Once in consensus, they stay in consensus even
if they have had to change their minds in order
to reach consensus. When the strengths of the
connections between networks is increased to a
point where it far outweighs the strengths of the
 1033/1540

connections within networks, the networks all

move to a shared interpretation that is incoher-
ent. In this condition, the importance of sharing
an interpretation with others outweighs the im-
portance of reaching a coherent interpretation.

It is clear that a megamind such as that de-

scribed by Asimov would be more prone to con-
firmation bias than any individual mind. It
might be a mind that would rush into interpreta-
tions and that, once it had lodged in an interpret-
ation, would manifest an absolutely incorrigible
confirmation bias. Is it possible for communica-
tion to ever be rich enough in a real human com-
munity to lead to this sort of group pathology?
Perhaps. Even in individual networks, as a co-
herent interpretation forms, units representing
hy-
 Page 255

potheses that have no direct support in the world

receive activation from neighboring units and a
whole coherent schema is filled in. This well-
known effect in individual cognition seems even
more powerful in some group settings. Buck-
hout (1982) asked groups to produce composite
descriptions of a suspect in a crime that all had
witnessed and reported that ''the group descrip-
tions were more complete than the individual re-
ports but gave rise to significantly more errors
of commission: an assortment of incorrect and
stereotyped details." This looks like a case in
which the members of the group settle into an
even more coherent interpretation than they
would do acting alone. That is just what
happened to the networks in the simulation.
 1035/1540

Of course, this extremely tight coupling is very

unusual. The next section considers cases where
there is some interesting distribution of access to
evidence, where the communication connection
strengths are moderate, where the pattern of in-
terconnectivity is partial (that is, where the indi-
viduals don't talk about everything they know,
only about some of the important things), and
where the communication is not continuous.

PRODUCING A DIVERSITY OF
INTERPRETATIONS
 1036/1540

The problem with confirmation bias is that it

prevents an organism from exploring a wider
range of possible interpretations. Although the
first interpretation encountered may well be the
best, a search of the interpretation space may re-
veal another one that better fits the available
evidence. How can this search be accomplished?
I have already shown that in the absence of
communication the interpretations formed by
the individual networksas each exhibits its own
confirmation biasdepend on the three parameters
that characterize the individual networks: the
underlying constraint structure, the access to en-
vironmental evidence, and the initial pattern of
activation. If a community is composed of indi-
viduals that differ from one another in terms of
any of these parameters, then various members
of the community are likely to arrive at different
interpretations. Thus, diversity of interpretations
is fairly easy to produce as long as the
 1037/1540

communication among the members of the com-

munity is not too rich.

Organizational Solutions to the Problem of

Reaching a Decision

Some institutions cannot easily tolerate situ-

ations in which the group does not reach a con-
sensus about which interpretation shall
 Page 256

be taken as a representation of reality. In some

settings it is essential that all members of the
group behave as though some things are true
and others are false, even if some of the mem-
bers have reservations about the solution de-
cided upon. The members of an aircraft crew,
for example, must coordinate their actions with
one another and with a single interpretation of
the state of the environment even if some of
them doubt the validity of the interpretation on
which they are acting. Such institutions may
face the problem of guaranteeing that a shared
interpretation is adopted in some reasonable
amount of time.

HIERARCHY
 1039/1540

A common solution to the problem of reaching a

decision is to grant to a particular individual the
authority to declare the nature of reality. This is
especially easy to see in settings where the rel-
evant reality is socially definedsuch as the law,
where an important state of affairs (guilt or in-
nocence) exists only because some authority (a
judge) says it exists. But this solution is also ad-
opted with respect to physical realities where
time pressures or other factors require a commit-
ment to a particular interpretation. This second
case comes in two versions: one in which the
other members of the community may present
evidence to the authority, and one in which the
authority acts autonomously. Here are two simu-
lation experiments on this theme:

Hierarchy without Communication

 1040/1540

Suppose all members of a group attempt to form

an interpretation, but one network has the au-
thority to decide the nature of reality for all the
members. The cognitive labor of interpreting the
situation may be socially distributed in a way
that permits an exploration of more alternatives
in the interpretation space than would be ex-
plored by a single individual with confirmation
bias; however, if the alternative interpretations
never encounter one another, the wider search
might as well have never happened. The de-
cision reached by the group is simply the de-
cision of an individual. One might imagine this
as a sort of "king" or "dictator" model, but lack
of communication can also bring it about in situ-
ations that are not supposed to have this prop-
erty. The ship collision discussed earlier is an
example of a case in which the correct interpret-
ation of a situation arose within a group but
somehow never reached the individual who had
 1041/1540

the authority to decide which model of reality

the group must organize its behavior around.
 Page 257

Hierarchy with Communication

 1043/1540

This situation is modeled in the simulation by

changing the communication pattern so that one
of the networks (the one in the position of au-
thority) receives input from all the others, but
the others do not receive external inputs from
one another. In the simulation under these con-
ditions, the network that is the authority will fol-
low the weight of the evidence presented to it by
the other networks (figure 5.5). As the other net-
works move in interpretation space, the center
of gravity of the weight of evidence presented
by the other networks also moves. Depending
on the persuasiveness with which the other net-
works communicate with the authority, it may
be pulled to one interpretation or another, or
even change its mind about which is the better
interpretation (figure 5.5c). The authority thus
becomes a special kind of cognitive apparatus;
one that tracks the center of gravity of the entire
community in conceptual space at each point in
time. At very high levels of persuasiveness, this
 1044/1540

authority network may find the evidence for

both interpretations compelling and be drawn to
a state in which it has high activations for the
units representing all the hypotheses in both in-
terpretations (figure 5.5d).

CONSENSUS

Quaker Decision RuleUnanimity or Nothing

 1045/1540

Imagine a world in which each network can at-

tend to only one aspect of the environment at a
time, but all networks communicate with one
another about the interpretations they form on
the basis of what they are attending to. Suppose
further that there is more information in the en-
vironment consistent with one interpretation
(call it the best interpretation) than with anoth-
er. Then, any single individual acting alone will
reach the best interpretation only when it hap-
pens to be attending to some aspect of the envir-
onment that is associated with that interpretation
or when it happens to be predisposed to that in-
terpretation. If there are many networks and the
aspect of the environment each attends to is
chosen at random, then on average more of them
will be attending to evidence supporting the best
interpretation than to evidence supporting any
other interpretation, since by definition the best
interpretation is the one for which there is most
support. If the networks in such a group are in
 1046/1540

high-bandwidth communication with one anoth-

er from the outset, they will behave as the Gaia
system did, rushing as a group to the interpreta-
tion that is closest to the center of gravity of
Page 258
1048/1540
1049/1540
 1050/1540

Cognition p258b
1051/1540
 1052/1540

Cognition p258c
 1053/1540

Cognition p258d

Figure 5.5 The trajectories of an authority. The authority

(plotted as an open circle) is a network that receives input
from the other networks. The other networks do not com-
municate among themselves. The authority follows the
center of gravity of the interpretations of the group. In
panel a, with low persuasiveness by subordinates, it fol-
lows the three who go to interpretation 2. With more per-
suasiveness, as in panel b, it follows the two who move
more quickly toward interpretation 1. With still more
 1054/1540

persuasiveness, it starts toward interpretation 1 but is

eventually drawn to interpretation 2. Panel d shows what
happens with very high values of persuasiveness. The au-
thority is drawn first to interpretation 2. But the two net-
works that have arrived at interpretation 1 make it im-
possible to ignore the elements of that interpretation and
the authority is driven to a state in which there is high ac-
tivation for the units representing all of the hypotheses of
both interpretations.
Page 259
1056/1540
 1057/1540

Cognition p259a
 1058/1540

Cognition p259b

Figure 5.6 Generation of diversity followed by consensus.

In this simulation, the networks start out not communicat-
ing. As time goes on, the persuasiveness with which they
communicate increases rapidly. Panel a shows some indi-
viduals nearly reaching interpretation 1 before being com-
pelled by the others to move to interpretation 2. A di-
versity of opinion is resolved into a consensus. In panel b,
the networks again begin exploring the interpretation
space, this time with a more rapid rate of increase in per-
suasiveness of communication. The networks achieve a
 1059/1540

consensus, but the interpretation on which consensus is

reached does not fit either of the coherent interpretations.

their predispositions regardless of the evidence.

If, however, they are allowed to go their own
ways for a while, attending to both the available
evidence and their predispositions, and then to
communicate with one another, they will first
sample the information in the environment and
then go (as a group) to the interpretation that is
best supported. Figure 5.6a shows the group ex-
ploring the space first (two networks almost
reach interpretation 1) and then, with increasing
persuasiveness of communication, reaching con-
sensus on interpretation 2. Not only is the beha-
vior of each individual different in the group
setting than it would have been in isolation; the
behavior of the group as a whole is different
from that of any individual, because the group,
as a cognitive system, has considered many
kinds of evidence while each individual con-
sidered only one.
 1060/1540

The simulations indicate two shortcomings of

this mode of resolving the diversity of interpret-
ations. First, if some individuals arrive at very
well-formed interpretations that are in conflict
with one another before communication begins,
there may be no resolution at all. They may
simply stay with the interpretations they have
already formed. Sometimes such "hard cases"
can be dislodged by
 Page 260

changing the distribution of access to evidence

in the community by giving stubborn networks
direct access to evidence that contradicts their
present interpretation. However, this may only
drive the network to a state in which it has no
coherent interpretation of the situation (figure
5.6b).

Demographics of Conceptual Space: Voting

 1062/1540

Another set of methods for establishing an inter-

pretation to be acted upon by a group relies on
measuring the demographics of the community
in conceptual space. In the initial state the mem-
bers of the group are sprinkled around in the
conceptual space. The starting location of each
is defined by the preconceptions it has about the
situation. Some may begin closer to one inter-
pretation, some closer to another. As each one
tries to satisfy the constraints of its internal
schemata and the available external evidence, it
moves in conceptual space. This movement is
usually toward one of the coherent interpreta-
tions defined by the underlying schemata. As
was shown above, if the members of the com-
munity are in communication with one another,
they may influence one another's motion in con-
ceptual space. A mechanism for deciding which
interpretation shall be taken as a representation
of reality may be based on the locations of the
members of the community in conceptual space.
 1063/1540

If a majority are at or close to a particular inter-

pretation, that interpretation may be selected as
the group's decision. This is, of course, a voting
scheme.

A majority-rule voting scheme is often taken to

be a way of producing the same result that
would be produced by continued negotiation,
but short-cutting the communication. In these
simulations, voting does not always produce the
same results that would be achieved by further
communication. That this is so can easily be de-
duced from the fact that the result of a voting
procedure for a given state of the community is
always the same, whereas a given state of the
community may lead in the future to many dif-
ferent outcomes at the group level (depending
on the time course and the bandwidth of sub-
sequent communication).

A Fundamental Tradeoff for Organizations

 1064/1540

Many real institutions seem to embody one or

another of these methods for first generating and
then dealing with diversity of interpretation. Ob-
viously, real social institutions come to be or-
ganized as they are for many reasons. For ex-
ample, the political
Page 261
 1066/1540

consequences of various schemes for distribut-

ing the authority to make interpretations real are
important aspects of the actual implementation
of any institution. I do not claim that institutions
are the way they are because they produce par-
ticular kinds of cognitive results. The point is
rather that social organization, however it may
have been produced, does have cognitive con-
sequences that can be described. By producing
the observed structures of organizationslargely
ones in which there are explicit mechanisms for
resolving diversity of interpretationssocial evol-
ution may be telling us that, in some environ-
ments, chronic indecision may be much less ad-
aptive than some level of erroneous commit-
ment. This may be the fundamental tradeoff in
cognitive ecology. The social organization, or
more precisely the distribution of power to
define situations as real, determines the location
of a cognitive system in the tradeoff space.
Where the power to define the reality of
 1067/1540

situations is widely distributed in a "horizontal"

structure, there is more potential for diversity of
interpretation and more potential for indecision.
Where that power is collected in the top of a
"vertical" structure, there is less potential for di-
versity of interpretation, but also more likeli-
hood that some interpretation will find a great
deal of confirmation and that disconfirming
evidence will be disregarded.
 1068/1540

Where there is a need for both exploration of an

interpretation space and consensus of interpreta-
tion, a system typically has two modes of opera-
tion. One mode trades off the ability to reach a
decision in favor of diversity of interpretation.
The participants in the system proceed in relat-
ive isolation and in parallel. Each may be sub-
ject to confirmation bias, but because they pro-
ceed independently, the system as a whole does
not manifest confirmation bias. The second
mode breaks the isolation of the participants and
exposes the interpretations to disconfirming
evidence, the goal being to avoid erroneous per-
severence on an interpretation when a better one
is available. This mode trades off diversity in fa-
vor of the commitment to a single interpretation
that will stand as the new reality of the situation.
Often the two modes are separated in time and
marked by different social structural
arrangements.
 1069/1540

Summary

In this chapter I have tried to take some tentative

steps toward a framework for thinking about
cognitive phenomena at the level of groups. The
simulation models are both a kind of notation
system
 Page 262

that forces one to be explicit about the theoretic-

al constructs that are claimed to participate in
the production of the phenomena of interest and
a dynamical tool for investigating a universe of
possibilities. The simulations show that, even
while holding the cognitive properties of indi-
viduals constant, groups may display quite dif-
ferent cognitive properties, depending on how
communication is organized within the group
and over time. Groups can be better at generat-
ing a diversity of interpretations than any indi-
vidual is; however, having generated a useful di-
versity, they then face the problem of resolving
it. From the perspective presented here, several
well-recognized kinds of social organization ap-
pear to provide solutions to the problems of ex-
ploring a space of interpretations and discover-
ing the best available alternative.
 1071/1540

All the strategies that overcome confirmation bi-

as work by breaking up continuous high-band-
width communication. This is true whether the
strategies are implemented in social organiza-
tion, in the interaction of an individual with an
external artifact, or through the use of an intern-
al mediating structure.
 1072/1540

In the simulations presented in this chapter, the

effects of grouplevel cognitive properties are not
produced solely by structure internal to the indi-
viduals, nor are they produced solely by struc-
ture external to the individuals. Rather, the cog-
nitive properties of groups are produced by in-
teraction between structures internal to individu-
als and structures external to individuals. All hu-
man societies face cognitive tasks that are bey-
ond the capabilities of any individual member.
Even the simplest culture contains more inform-
ation than could be learned by any individual in
a lifetime (Roberts 1964; D'Andrade 1981), so
the tasks of learning, remembering, and trans-
mitting cultural knowledge are inevitably dis-
tributed. The performance of cognitive tasks that
exceed individual abilities is always shaped by a
social organization of distributed cognition. Do-
ing without a social organization of distributed
cognition is not an option. The social organiza-
tion that is actually used may be appropriate to
 1073/1540

the task or not. It may produce desirable proper-

ties or pathologies. It may be well defined and
stable, or it may shift moment by moment; but
there will be one whenever cognitive labor is
distributed, and whatever one there is will play a
role in determining the cognitive properties of
the system that performs the task.
 Page 263

6 Navigation as a Context for Learning

 1075/1540

The primary function of a peacetime military is

what is called "the maintenance of readiness."
The military establishment is a big institution,
full of terrifying weapons systems and other ar-
tifacts. The glue that holds the artifacts together-
that makes the separate ships and planes and
missiles and bombs into something more than a
collection of hardwareis human activity. But
there are high rates of personnel turnover in the
military. The human parts keep passing through
the system. Thus, even though the system is
ready to make war one day, it will not be ready
the next day unless the expertise of the people
departing is continually replaced by the newly
acquired skills of those who have recently
entered. This high turnover of personnel and the
resulting need for the continual reproduction of
expertise makes the military a fertile ground for
research into the nature of learning in cultural
context.
 1076/1540

The Developmental Trajectory of the

Quartermaster
 1077/1540

It takes about a year to learn the basics of the

quartermaster's job. For a young person learning
to be a quartermaster, there are many sources of
information about the work to be done. Some
quartermasters go to specialized schools before
they join a ship. There they are exposed to basic
terminology and concepts, but little more. They
are "trained" in a sense, but they have no experi-
ence. In fact, the two quartermaster chiefs with
whom I worked most closely said they preferred
to get as trainees able-bodied seamen without
any prior training in the rate. They said this
saved them the trouble of having to break the
trainees of bad habits acquired in school. Most
quartermasters learn what to do and how to do it
while on the job. Nonetheless, some of a
trainee's experience aboard ship is a bit like
school, with workbooks and exercises. In order
to advance to higher ranks, the novice must
work through a set of formal assignments that
cover the full spectrum of navigation practice;
 1078/1540

these must be reviewed and approved by a su-

pervisor before the student can progress to the
next rank in the rating.
 Page 264

Sea and Anchor Detail

In the world of navigation, as in many other sys-

tems, novices begin by doing the simplest parts
of a collaborative-work task. Long before they
are ready to stand watch under instruction in
Standard Steaming Watch, novice quartermas-
ters begin to work as fathometer operators and
pelorus operators in Sea and Anchor Detail. As
they become more skilled they move on to more
complex duties, making way for less skilled in-
dividuals behind them. The procedural decom-
position of the task in this work configuration
permits unskilled persons to participate in com-
plex activities. The jobs in Sea and Anchor De-
tail are, in order of increasing complexity, the
following:

monitoring the fathometer

taking bearings
 1080/1540

keeping the deck log

timing and recording bearings

plotting fixes and projecting the dead reckoned

track.

This list of jobs defines a career trajectory for

individuals through the roles of the work group.
Interestingly, it also follows the path of informa-
tion through the system in the team's most basic
computation, position fixing. The simplest jobs
involve gathering sensed data, and the more
complex jobs involve processing those data. The
fact that the quartermasters follow the same tra-
jectory through the system as does sensed in-
formation, albeit on a different time scale, has
an important consequence for the larger system's
ability to detect, diagnose, and correct error. To
see why this is so, however, we need to consider
the distribution of knowledge that results from
this pattern of development of quartermasters.
 1081/1540

System Properties

The Distribution of Knowledge

Analysts often assume that in cooperative tasks

knowledge is partitioned among individuals in
an exhaustive and mutually exclusive manner
such that the sum of the individuals' knowledge
is equal to the total required, with little or no
overlap. Consider the knowledge required to
perform just the input portion of the basic fix
cycle. This requires the knowledge of the pelor-
us operators, the
 Page 265

Cognition p265

Figure 6.1 A nonoverlapping distribution of knowledge

among the members of the navigation team.
 1083/1540

bearing recorder, and the plotter. We could ima-

gine designing an experiment along these lines
by training a different person to perform each of
these roles and then putting the individuals in
interaction with each other. This assumes no
history for the participants except that each is
trained to do his job. This would result in a
nonoverlapping distribution of knowledge, as
shown in figure 6.1. It is certainly possible to or-
ganize a working system along these lines, but
in fact, outside of experimental settings, this is a
very rare knowledge-distribution pattern. More
commonly there is substantial sharing of know-
ledge between individuals, with the task know-
ledge of more expert workers completely sub-
suming the knowledge of those who are less ex-
perienced. At the other end of the knowledge-
distribution spectrum, one can imagine a system
in which everyone knows everything about the
task. This too is a rare pattern, because it is
costly to maintain such a system.
 1084/1540

In many human systems, as individuals become

more skilled they move on to other roles in the
task-performance group, making way for less
skilled individuals behind them and replacing
the more expert individuals before them who ad-
vance or leave the system. This is what we ob-
serve in the case of the development of naviga-
tion skills among quartermasters. A competent
pelorus operator knows how to do his job, but
because of his interaction with the bearing re-
corder he also knows something about what the
recorder needs to do (figure 6.2a). The bearing
recorder knows how to do his job, but he also
knows all about being a pelorus operator, be-
cause he used to be one. Furthermore, he knows
a good deal about the activities of the plotter,
because he shares the chart table with the plotter
and may have done plotting under instruction in
Page 266
1086/1540
 1087/1540

Cognition p266a
 1088/1540

Cognition p266b
1089/1540
 1090/1540

Cognition p266c
 1091/1540

Cognition p266d

Figure 6.2 Overlapping distributions of knowledge among

the members of the navigation team. (a) What the bearing
taker knows. (b) What the bearing recorder knows. (c)
What the plotter knows. (d) The distribution of redundant
coverage in the system as a whole.
 1092/1540

Standard Steaming Watch. What the bearing re-

corder knows is shown in figure 6.2b. Finally, a
competent plotter knows how to plot, but he also
knows everything the bearing recorder and the
pelorus operators must know in order to do their
jobs, because he has done both of those jobs be-
fore advancing to plotting. What the plotter
knows is shown in figure 6.2c, and the distribu-
tion of knowledge that is the sum of these indi-
vidual expertises is shown in figure 6.2d. Thus,
this movement of individuals through the sys-
tem with increasing expertise results in a pattern
of overlapping expertise. Knowledge of the
entry-level tasks is represented most redund-
antly, and knowledge of the expert-level tasks is
represented least redundantly.
 Page 267

The Decompositions of Tasks

 1094/1540

The structure of the distributed task provides

many constraints on the learning environment.
The way a task is partitioned across a set of task
performers has consequences for both the effi-
ciency of task performance and the efficiency of
knowledge acquisition. For example, if the de-
composition into subtasks cuts lines of high-
bandwidth communication (that is, if two pro-
cesses that need to share information often in or-
der to reach completion are distributed across
different task performers), the task performance
may suffer from the effects of a bottleneck in in-
terpersonal communication. What parts of the
process need to communicate with which other
parts and what sorts of structures can be repres-
ented in the available communications media
are important determinants of optimal task parti-
tioning. In chapter 4 the situation in which inex-
perienced pelorus operators attempt to find land-
marks in the world was presented as a way to
explore the computational consequences for the
 1095/1540

larger system of various means of communica-

tion. This chapter considers the implications of
this same phenomenon for learning by individu-
al members of the team.

If the pelorus operator already knows how to

find the landmark in question, then little inform-
ation needs to be passed. The name of the land-
mark may be all that is required. If the pelorus
operator is unsure of the location or appearance
of the landmark, more information may be re-
quired. For example, in the following exchange,
the starboard pelorus operator needed additional
information to resolve an ambiguity concerning
the identity of a landmark. (In this exchange,
SW is the starboard wing pelorus operator and S
is a qualified watchstander working as bearing
recorder. They were communicating via the
sound-powered phone circuit.)

SW: The one on the left or the one on the right?

 1096/1540

S: The one on the left, OK?

SW: Yeah, I got it.

When the confusion or lack of knowledge is

more profound, it is simply impossible to com-
municate enough information, or the right kind
of information, over the phone circuit; someone
has to go to the wing to show the pelorus operat-
or where to find the landmark. A little later dur-
ing the same exit from port, the starboard pelor-
us operator was unable to find the north end of
the 10th Avenue terminal. The plotter C, who is
also the most-qualified and
 Page 268

highest-ranking member of the team, went onto

the wing to point it out to him. On the wing, C
put his arm over SW's shoulders and aimed his
body in the right direction.

C: The north one, all the way up.

SW: OK.

C: If you can't see the light, just shoot the tan-

gent right on the tit of the, the last end of the
pier there.

SW: OK, that pier, where those two ...

C: Yeah, all the way at the end.

SW: All right.

C: There should be a light out there, but if you

can't see the light out there at the end of the pier
just shoot the end of the pier.
 1098/1540

In this example, verbal and gestural interaction

provides the additional identifying information.
Furthermore, the instructions include expecta-
tions about what will be visible at the time of
the next observation and instructions as to what
the pelorus operator should do if he is unable to
see the light on the pier.

The Horizon of Observation

Lines of communication and limits on observa-

tion of the activities of others have con-
sequences for the process of acquiring know-
ledge, because they determine the portion of the
task environment that is available as a learning
context to each task performer. The outer
boundary of the portion of the task that can be
seen or heard by each team member is that
person's horizon of observation.

OPEN INTERACTIONS
 1099/1540

During an early at-sea period, L, the keeper of

the deck log, had served as bearing recorder, but
his performance in that job was less than satis-
factory. That was the job that was next in line
for him, though, and he was eager to acquire the
skills required to perform it. One of the most
important aspects of the bearing recorder's job is
knowing when particular landmarks will be vis-
ible to the pelorus operators on the wings. One
complication of this judgement is the fact that a
large convex mirror is mounted outside the win-
dows of the pilothouse just in front of the port
wing pelorus operator's station. The mirror is
there so that the commanding officer, who sits
inside the pilothouse, can see all of the flight
deck. Unfortunately,
 Page 269

the mirror obstructs the port pelorus operator's

view forward, and the bearing recorder must be
able to judge from his position at the chart table
whether the port wing pelorus operator's view of
a chosen landmark will be blocked by the
mirror.

L was standing at the chart table with C (the

plotter) and S (the bearing recorder). The ship
had just entered the mouth of the harbor, and the
team was running the fix cycle on 2-minute in-
tervals. The previous fix, taken at 36 minutes
after the hour (''time 36"), was complete, and C
had just finished plotting the dead-reckoned
track out through times 38 and 40. S indirectly
solicited C's assistance in deciding which land-
marks should be shot for the next round of bear-
ings. L stood by, watching what S and C were
doing. All the pointing they did in this inter-
change was at the chart.
 1101/1540

1. S: Last set still good? OK Ballast Point, light

Zulu.

2. C: Here's (time) 3 8 (pointing to the DR posi-

tion on the chart).

3. S: So it would be that (pointing to light Zulu),

that (pointing to Bravo Pier) ...

4. C: One, two, three. Same three. Ballast Point,

Bravo. And the next one ...

5. S: (Time) 4 0 should be, Ballast Point ...

6. C: Front Range, Bravo.

7. S: And Balla...

8. L: He may not be able to see Front Range.

9. S: Yeah.
 1102/1540

10. C: Yeah, he can. Once we get up here

(pointing to the ship's projected position for the
next fix).

11. S: Yeah. Up there. OK.

12. C: Down here (pointing to ships current pos-

ition) he can't. It's back of the mirror, but as you
come in it gets enough so that you can see it.

Because the actions of S and C are within L's

horizon of observation, L has a chance to see
how the landmarks are chosen. Furthermore, the
fact that the decision about which landmarks to
shoot is made in an interaction opens the pro-
cess to him in a way that would not be the case
if a single person were making the decision
alone. In utterance 8, L raises the possibility that
the port wing pelorus operator may not be able
to see the landmark. Three days earlier, on an-
other Sea and Anchor Detail, L had made the
same
1103/1540
 Page 270

suggestion about the mirror blocking the port

wing pelorus operator's view and C had agreed
with him. In the present circumstances,
however, L's caveat is inappropriate. S and C
have already anticipated the problem raised by
L, and they jointly counter L's objection, each
building on what the other has said. Clearly, if L
did not share the work space with S and C or if
there was a strict division of labor such that in-
dividuals did not monitor and participate in the
actions of their fellows, this opportunity for L to
have even peripheral involvement in a task that
will someday be his would be lost. Furthermore,
L's horizon of observation is extended because
the decision making about landmarks is conduc-
ted as an interaction between S and C.

OPEN TOOLS
 1105/1540

Simply being in the presence of others who are

working does not always provide a context for
learning from their actions. In the example
above, the fact that the work was done in an in-
teraction between the plotter and the bearing re-
corder opened it to other members of the team.
In a similar way, the design of tools can affect
their suitability for joint use or for demonstra-
tion and may thereby constrain the possibilities
for knowledge acquisition. The interaction of a
task performer with a tool may or may not be
open to others, depending upon the nature of the
tool itself. The design of a tool may change the
horizon of observation of those in the vicinity of
the tool. For example, because the navigation
chart is an explicit graphical depiction of posi-
tion and motion, it is easy to "see" certain as-
pects of solutions. The chart representation
presents the relevant information in a form such
that much of the work can be done on the basis
of perceptual inferences. Because the work done
 1106/1540

with a chart is performed on its surfaceall of the

work is at the device's interface, as it were-
watching someone work with a chart is much
more revealing of what is done to perform the
task than watching someone work with a calcu-
lator or a computer.

The openness of a tool can also affect its use as

an instrument of instruction. When the bearing
recorder chooses a set of landmarks that result
in lines of position with shallow intersections, it
is easy to show him, on the chart, the con-
sequences of his actions and the nature of the
remedy required. Figure 6.3 shows a fix that res-
ulted from landmark assignments made by the
bearing recorder. Bearings off to the side of the
ship rather than ahead or astern are called
"beam" bearings. When the plotter plotted this
fix and saw how it came out, he scolded the
bearing recorder:
Cognition p271
 1108/1540

Figure 6.3 A fix constructed with beam bearings. Because of

chart as a shared workspace, it is easy for the plotter to show
includes beam bearings and why the beam bea

C: What did you take a bunch of beam bearings for

there (points out the front window of the bridge) so
(points to the chart) You shot three beam bearings.
You better tell 'em to shoot from up ahead somepla

Once the fix was plotted, of course, it was easy for

nature of his error. Imagine how much more difficu
adequacy of the landmark assignment if the lines of
equations to be punched into a calculator rather tha

Learning from Error

Most studies of error focus on its reduction or elimi

steps that can be taken to avoid or prevent error. Ye
systems. This is commonly attributed to the fact tha
tired, or confused, or
 Page 272

distracted, or inattentive to our work, or to the

idea that we are inherently fallible in some other
way. These are real contributing factors in many
errors, of course. However, in the case of sys-
tems of cooperative work in the real world there
is a more fundamental reason for the inevitabil-
ity of error: Such systems always rely on on-the-
job learning, and where there is the need for
learning there is room for error.
 1110/1540

Naturally situated systems of cooperative work

must produce whatever it is they produce and
must reproduce themselves at the same time.
They change over time. Sometimes they are re-
organized. Sometimes they change the things
they do, and sometimes the technology they
have to do the job with changes. Even if tasks
and tools could be somehow frozen, the mortal-
ity of the human participants guarantees changes
in personnel over time. Most commonly, relat-
ively expert personnel are gradually lost while
relatively inexpert personnel are added. Even if
the skills required to do the job are learned off
the job, in school, the interactions that are char-
acteristic of cooperative work can generally
only be learned on the job.

Designing for Error

 1111/1540

Norman (1983, 1986, 1987) argues that because

error is inevitable it is important to "design for
error." Designers, he says, can "inadvertently ...
make it easy to err and difficult or impossible to
discover error or to recover from it" (1987,
chapter 5: 24). Norman suggests that designers
should design to minimize the causes of error,
make it possible to "undo'' errors, and make it
easier to discover and correct errors. Each of
these suggestions is aimed at protecting the cur-
rent task performance, yet in the broader per-
spective of production and reproduction in co-
operative work it would be useful if the re-
sponse to error in the current task could also in
some way protect future task performance. That
is, another aspect of designing for error might be
designing systems that can more easily learn
from their errors.
 1112/1540

That would give us three major classes of design

goals with respect to errors: to eliminate, avoid,
or prevent errors wherever possible; to facilitate
the recovery of the system from any errors that
do occur; and to facilitate learning from any er-
rors that do occur so that future errors become
less likely. There is something of an equilibrium
to be maintained here. As career trajectories
take
 Page 273

experienced members out of the work group and

expertise is lost from the system, the likelihood
of error may increase. This increase in likeli-
hood of error must be offset by the decrease in
likelihood of error that comes from learning by
the remaining and new members of the cooper-
ative work group.

Controlling the Effects of Error

ERROR DETECTION

Error detection may require extensive resources.

Observations of the conditions under which er-
rors are detected indicate that the following ele-
ments are necessary:

Access In order to detect an error, the detector

must have access to the behavior that is in error
or some indication of it.
 1114/1540

Knowledge or expectation The detector must

have knowledge of the process being performed
or some expectation about its correct outcome
with respect to which the observed process or
outcome can be judged to be in error.

Attention Whoever detects the error must attend

to the error and monitor it in terms of
expectation.

Perspective Some perspectives are better than

others at bringing interested attention relevant
expectations to bear on the evaluation of
behavior.

Access
 1115/1540

The structure of the task and the extent to which

the behaviors of the participants are available to
the other participants have consequences for er-
ror detection. In the following example, the
team had shot Front Range, Silvergate, and
Light 2 on the previous round. S began to make
a shift that would drop Front Range on the port
side and pick up North Island tower on the star-
board side. After instructing PW to drop Front
Range, he then discovered that SW couldn't yet
see North Island tower. Only a request for clari-
fication from another sailor making a redundant
plot in the combat information center (CIC)
made it clear that S had decided not to shift
landmarks at all and that PW had misunderstood
the situation.

S: (to PW) OK, shift to Silvergate, John.

PW: Drop Front Range.

 1116/1540

S: Drop Front Range. (to SW) Steve, pick up (3

seconds) ah, just stick with number 2.
 Page 274

PW: All right.

CIC: (to S) OK, John, you're gonna shoot Light

2, Silvergate, and the Front Range, right?

S: Yeah, Light 2, Silvergate and Front Range.

CIC: OK.

PW: I thought we dropped Front Range.

S: No, picked that up because he couldn't see the

tower on this side here (starboard).

PW: The Front Range and Silvergate, right?

S: Yeah.
 1118/1540

The point of this example is that the density of

error correction possible depends on the hori-
zons of observation of the team members. Here
PW is on the phone circuit with CIC and S. In
this case, this problem would surely not have
been detected had the communication between S
and CIC not been available to PW.

Knowledge
 1119/1540

The importance of the distribution of knowledge

produced by the overlapping careers of a set of
quartermasters following a career trajectory that
coincides with the flow of sensed information
can now be stated: As a consequence of the
alignment of one's career trajectory with the
path of information through the system, if one
has access to an error, then one also has know-
ledge of the processes that may have generated
it, because one has already, at an earlier career
stage, performed all the operations to which the
data have been subjected before arriving at one's
current position. The overlap of access and
knowledge that results from the alignment of ca-
reer path and data path is not a necessary feature
of these systems, nor is it apparently an inten-
tional one here, but it does give rise to espe-
cially favorable conditions for the detection and
diagnosis of error.

Attention
 1120/1540

The attention required to detect error may be fa-

cilitated or even forced by the nature of coordin-
ated tasks. If errors in the upward propagation of
sensed data are not caught at the lower levels,
they are likely to be noticed at the chart table by
the plotterin part because the plotting procedure
itself is designed to detect error. Any two lines
of position define the location of the ship, but a
position "fix" always consists of three lines of
position. The value
 Page 275

of the third line of position in the fix is that, if

an error is present, it is likely to show up as an
enlarged fix triangle, which will be detected by
the plotter. It is, of course, possible for inde-
pendent errors to conspire to produce a nice
tight fix triangle that is actually in the wrong
place, but such an event is quite unlikely. The
nature of the plotter's task makes errors in the
bearings evident.

Many errors are detected by team members who

are simply monitoring the actions of those
around them. Competition may develop among
peers doing similar tasks. Feedback can be
provided in attempts to show competence, as in
the following example, where PW faults SW's
reporting sequence.

S: Stand by to mark time 1 4.

F: Fifteen fathoms.
 1122/1540

SW: Dive tower 0 3 4.

PW: He didn't say "mark."

S: Mark it. (2 seconds) I've got the dive tower,

Steve, go ahead.

PW: Point Loma 3 3 9.

 1123/1540

Here SW was supposed to wait for the "mark"

signal, but he blurted out the bearing of Dive
Tower when he heard the "stand by to mark"
signal. PW jumped on him. S then gave the
"mark" signal and waited for reports, but the
earlier confusion seemed to have disrupted the
coordination of the reports, and no one said any-
thing for 2 seconds. The work of getting bear-
ings must continue uninterrupted. Stopping to
repair the situation would ruin the fix because
the near simultaneity of the observations would
be violated. S minimized the damage by ac-
knowledging the early report on Dive Tower
and asking PW to report.
 1124/1540

Not only is each member of the team respons-

ible for his own job; each seems also to take re-
sponsibility for all parts of the process to which
he can contribute. Since detection depends on
access, however, the extent to which the activit-
ies of the team members are conducted where
they can be observed (or overheard) by others
may affect the rate of error detection. Error de-
tection also requires attention, which may be a
scarce resource. The consequences of a high
workload may include both an increase in the
rate of error itself (due to the effects of stress)
and a reduction in the rate of error detection
(due to the reduction of resources available for
monitoring the actions of others).
 Page 276

Perspective

The way the plotter thinks about the bearings

and uses them in his task is different from the
way the bearing observers think about the bear-
ings. For the bearing observer, the bearing may
be no more than a string of three digits read
from a scale in a telescopic sight. It is not neces-
sary for the bearing observer to think of the dir-
ectional meaning of the number. In contrast, the
plotter's job is to recover the directional mean-
ing of the reported bearing and combine the
meanings of three bearings to fix the position of
the ship. Different jobs in the team require atten-
tion to different aspects of the computational ob-
jects, so different kinds of errors are likely to be
detected (or missed) by different members of the
team.
 1126/1540

On some ships the job of making a global as-

sessment of the quality of the work of the navig-
ation team is institutionalized in the role of the
evaluator. This evaluator is a qualified naviga-
tion practitioner who is not engaged in doing the
navigation computations themselves, but instead
monitors the process by which the computation
is performed and the quality of the product.
There is a tradeoff here between using the
evaluator's processing power to do the computa-
tions themselves (and possibly achieving lower
error rates but certainly risking lower error-de-
tection rates) and having less processing power
in the task itself (and possibly generating more
errors but certainly detecting more of the errors
that are committed).
 1127/1540

The important structural property of the

evaluator's role is that the evaluator has access
to and knowledge of the performance of the task
but does not participate in the performance. In-
stead, the evaluator attends to the way the task
is done and specifically monitors the perform-
ance for error. The social role of the evaluator is
a way of building into the system attention to an
aspect of the system's behavior that would not
otherwise be reliably present.

RECOVERY FROM ERROR

Not every recovery from error is instructional in

intent or in consequence. Some are simply what
is required to get the job done. There may be no
need to diagnose the cause of the error in order
to know how to recover from it.
 1128/1540

Other error-recovery strategies involve diagnos-

is of the source of the error and perhaps explicit
demonstration of the correct solution, where the
demonstration may also be the performance that
is required. Diagnosis may require modeling the
reasoning of the
 Page 277

person who committed the error. The distribu-

tion of knowledge in which access to an error
and knowledge of its causes are aligned ensures
that most errors that are detected will be detec-
ted by individuals who already have experience
with the operations that led to the error. This
gives each task performer a much better basis
for diagnosing the possible causes of any ob-
served errors than he would have in a system
with discrete knowledge representation. When a
bad bearing is reported, the plotter can examine
it and may develop hypotheses about, for ex-
ample, whether the pelorus operator misread the
gyrocompass scale or whether the bearing re-
corder mistranscribed it into the bearing log.
These hypotheses are based on the plotter's ex-
perience in each of these roles.
 1130/1540

Furthermore, the time, the processing resources,

and the communication channels required for
the composition and delivery of appropriate in-
struction must be available to the person provid-
ing correction at or near the time that the error is
committed. In navigation teams, error feedback
is sometimes reduced to a contentless complaint
or an exhortation to do better. Such limited feed-
back may be of little use to the person who has
committed the error, but it may be all that the er-
ror detector can do under the circumstances of
the task.

LEARNING FROM ERROR

A system can learn from its errors in many

ways.

Learning by Detecting and Correcting

 1131/1540

As a consequence of having engaged in the

activities of detecting and/or diagnosing the
cause of an error, the person doing the detecting
may come to a new insight about the nature of
the operation of the system. This is true whether
the error was committed by that person or by
someone else. It may be particularly important
for novices who detect the errors of others. Get-
ting confirmation of a detection could also be
important. Furthermore, every instance of cor-
rection presents an opportunity to develop error-
detection skills which may save the system from
the consequences of a future error.

Learning from the Correction of One's Own

Mistakes
 1132/1540

This is the most obvious case, the one that

comes first to mind when one asks how re-
sponse to error could improve future perform-
ance. Even feedback that lacks instructional
content can contribute to refinement of under-
standing of the task requirements that may not
be apparent from correct performance alone.
Such
 Page 278

corrections may help the learner induce the prin-

ciples that define correct performance. This can
be especially important with concepts that must
be inferred from cases rather than explicitly
stated. Where there is a solution space to be ex-
plored, errors are a form of exploration of that
space, and the response to an error can guide the
discovery of the concept underlying the
solution.
 1134/1540

Contentful corrections can help a novice learn

recovery strategies that can be applied even
when errors are self-detected. For example, S
and L were plotting together in Standard Steam-
ing Watch. Just after plotting a fix at 0745, L
spanned the distance (7½ miles) that the ship
would cover in the coming 30 minutes and plot-
ted the dead-reckoning fix. However, he incor-
rectly labeled the fix 0800 rather than 0815. S
noticed the error, and he and L constructed the
correction interactively. This is a nice example
of monitoring, error detection, and correction in
a jointly performed task:

S: Isn't that (pointing to the distance along the

track) half an hour?

J: No, it's (the time interval) 15 minutes.

S: Sure, 0745 to 0800 is 15 minutes, but we

don't go that far (distance along track).
 1135/1540

J: Huh?

S: This (time label of DR point) must be 0815.

Otherwise we are going 30 knots.

Learning from the Correction of the Mistakes of

Others
 1136/1540

When an error is detected and corrected in the

context of collaborative work, many participants
may witness and benefit from the response.
Depending again on who has access to what as-
pects of the behavior of others, an error and its
correction may provide a learning context for
many of the participants. For example, a fatho-
meter operator who has not yet worked as a
pelorus operator can learn a good deal about the
task by sharing the phone circuit with the pelor-
us operators and witnessing their mistakes and
the corrections to them. In a system populated
by novices and experts, many errors are likely to
occur; but since there are many sources of error
correction, most errors are likely to be detected.
Witnessing such a correction may be of value to
those who are already competent in the task in
which the error occurred, if they will sub-
sequently be in a position to detect and correct
such errors. They can learn about how to
provide useful feedback by watching the
 1137/1540

corrections of others. This could lead to an im-

provement of sub-
 Page 279

sequent learning for others in the system. Thus,

the value of a response to error for future per-
formances may depend on the horizons of obser-
vation of the various members of the team. This
observation leads to the somewhat paradoxical
conclusion that some nonzero amount of error
may actually be functional on the whole. A low
level of error that is almost certain to be detec-
ted will not, in ordinary circumstances, harm
performance, yet every error-correction event is
a learning context not just for the person who
commits the error but for all who witness it.

Tradeoffs in Error Management

Widening the horizon of observation may lead

to more error detection, but it may also cause
distractions that lead to the commission of more
errors. At present I know of no way to quantify
these tradeoffs, but recognizing their nature is
surely a useful first step.
 1139/1540

The costs of error include the undesirable effects

of undetected errors in the performance of the
current task. Even when errors are intercepted,
there are costs of detecting and recovering from
error. Under some conditions, these costs can be
offset to some extent by benefits derived from
the process of detecting, diagnosing, and cor-
recting errors. Achieving these benefits is not
automatic. There are ways to organize systems
that are more and less likely than others to de-
tect errors, to recover from them adequately, and
to learn from them.

The Social Formation of Competence in

Navigation

Standard Steaming Watch

 1140/1540

As was described in chapter 1, when the ship is

far from land, where the requirements of naviga-
tion are relatively light and the time pressures
are relaxed, navigation is conducted in a config-
uration called Standard Steaming Watch. In this
condition, a novice may stand watch "under in-
struction" with someone who is qualified to
stand watch alone. Depending on his level of ex-
perience, the novice may be asked to perform all
the duties of the quartermaster of the watch.
While he is under instruction, his activities are
closely monitored by the more experienced
watchstander, who is always on hand and who
can help out or take over if the novice is
 Page 280

unable to satisfy the ship's navigation require-

ments. However, even with the help of a more
experienced colleague, standing watch under in-
struction requires a significant amount of know-
ledge, so novices do not do this until they have
several month's experience.
 1142/1540

The task for the novice is to learn to organize

his behavior so that it produces a competent per-
formance. In chapter 4 and the earlier parts of
this chapter, I tried to show how a novice who
lacks specific knowledge can contribute to a
competent performance if the missing know-
ledge is provided by other members of the Sea
and Anchor Detail. This makes learning a
doubly cultural process. It is cultural first and
foremost because what is learned is a set of cul-
turally prescribed behaviors, but it is also cultur-
al to the extent that the participants in the sys-
tem have expectations about who needs what
kinds of help in learning the job. The scaffold-
ing provided to the novice by the other members
of the team is constructed on cultural under-
standings about what is hard and what is easy to
learn. In this sense, what individuals don't do for
one another may be as revealing as what they
do.
 1143/1540

In the following example, a novice quartermas-

ter, Seaman D, was standing watch under in-
struction by C. The tasks were to fill out routine
position-report and compass-report forms. The
position report requires the current latitude and
longitude of the ship. D was unsure how to pro-
ceed, so C asked him to measure the latitude and
longitude of the ship's current position on the
chart and dictate the values to C, who then re-
corded them on the position-report sheet. Here,
the labeled blank spaces on the printed form
provided some of the structure of the task.
Filling the blanks in order, from the top to the
bottom of the form, provided a sequence for the
elements of the task. That sequential structure
was presented explicitly to D by C, who as-
signed the subtasks.
 1144/1540

In this case, the functional unit of analysis is

defined by the requirements of the computation
involved in getting the numbers into the
position-report form. The functional system that
accomplishes this task transcends the boundaries
of the individual participants in the task. Medi-
ating structure is required to organize the se-
quence of actions that will produce the desired
(culturally specified) results. The questions are:
Where does the mediating structure reside? How
does it get into coordination with other bits of
structure to produce the observed actions?

The chief uses the organization of the form as a

resource in organizing his behavior. He employs
a simple strategy of taking items
 Page 281

in order from top to bottom on the form. The

novice coordinates his actions with those of the
chief by responding in order to the tasks presen-
ted by the chief. The whole ensemble, consisting
of the printed form, the chief (with his internal
structuresthe ability to read, etc.), and the novice
(with his own skillshis understanding of English
and his knowledge of how to measure latitude
and longitude on the chart), is the functional
system that accomplishes this element of the
culturally defined activity of navigation.
 1146/1540

The chief's use of the form is both a way to or-

ganize his own behavior and an example to the
novice of a way to use such a resource to organ-
ize behavior. Given the form, the novice might
now be able to reproduce the chief's use of the
form to organize his own behavior without the
chief's being present. The performance of the
task also provides the novice with the experi-
ence of the task and the sequence of actions that
can accomplish it. We might imagine that, with
additional experience, the novice would be able
to remember the words of the chief's queries, re-
member the meanings of the words, and remem-
ber physical actions that went into the satisfac-
tion of those queries. In such a case, we would
have a different functional system accomplish-
ing the same task in the navigation system. Sup-
pose the written form was not present, but the
watchstander wanted to extract the information
required by the form and write it on the form
later. In such a case the functional system would
 1147/1540

produce the specified products by using differ-

ent sorts of mediating structures (several kinds
of memory internal to the watchstander). In the
next chapter I will describe learning itself in
terms of these rearrangements of functional sys-
tems with experience.

Later, while working on the compass report, D

was again unsure what to do. The task is to
make sure the gyrocompass and the magnetic
compass are in agreement. This is done by tak-
ing simultaneous readings from the two com-
passes and then applying corrections to the
magnetic-compass reading and seeing whether
the corrected magnetic heading is the same as
the observed gyrocompass heading. The
magnetic-compass reading is called the checking
head, and the corrections to the magnetic com-
pass include a quantity called deviation. As C
filled in the form, he and D had the following
conversation:
 1148/1540

C: What's our checking head?

D: 0 9 0 and 0 7 4 (reading the gyrocompass and

magnetic compasses at the helm station)
 Page 282

C: What's the table deviation for 0 7 4?

D: One east (reading from the deviation card

posted on the binnacle).
 1150/1540

In asking D these questions, C was not only get-

ting him to practice the subtasks of reading
headings from the compasses and finding the
deviation in a table; he was also guiding D
through the higherlevel task structure. D knew
how to do the component tasks, but he didn't
know how to organize his actions to get the task
done. Some aspect of the organization of action
required is present in the labeled blank spaces
on the compass report form, but D was, by him-
self, unable to make use of that structure. C in-
terpreted that structure for D by asking the ques-
tions that are implied by the spaces. With C
(himself in coordination with the form) provid-
ing the task organization, D became part of a
competent performance. As D becomes more
competent, he will do both the part of this task
that he did in this instance and the organizing
part that was done in this instance by C. The
next time D confronts this task, he may use the
structure of the form to organize his actions.
 1151/1540

When D becomes a fully competent quartermas-

ter, he will not need the form for its organizing
properties; he will be able to say what the form
requires without consulting it, and will use the
form only as a convenient place to compute
corrections.
 1152/1540

The structure required a the novice to organize

his behavior in a competent performance in
Standard Steaming Watch is sometimes
provided by the supervising watchstander. Sim-
ilarly, when the quartermasters work as a team
in Sea and Anchor Detail, each provides the oth-
ers, and the others provide each, with constraints
on the organization of their activities. A good
deal of the structure that a novice will have to
acquire in order to stand watch alone in Stand-
ard Steaming Watch is present in the organiza-
tion of the relations among the members of the
team in Sea and Anchor Detail. The computa-
tional dependencies among the steps of the pro-
cedure for the individual watchstander are
present as interpersonal dependencies among the
members of the team. To the extent that the
novice participant comes to understand the work
of the team and the ways various members of
the team depend on each other, perhaps espe-
cially the ways he depends on others and others
 1153/1540

depend on him, he is learning about the compu-

tation itself and the ways its various parts de-
pend on one another. Long before he knows
how to choose appropriate landmarks to shoot,
the pelorus operator learns
 Page 283

that landmarks must be chosen carefully and as-

signed prior to making observations.
 1155/1540

There are at least two important implications of

the fact that computational dependencies are so-
cial dependencies in this system. First, the
novices' understandings of the social relations of
the workplace are a partial model of the compu-
tational dependencies of the task itself. If it is
true that human minds evolved to process social
relations, then packaging a task in a social or-
ganization may facilitate understanding it. Lev-
inson (1990) argues that this may be related to
the commonplace strategy of explaining mech-
anical and other systems in terms of social rela-
tions among anthropomorphized components.
Second, the communicative acts of the members
of the navigation team are not just about the
computation; they are the computation. When
this is the case, the playing out of computational
processes and the playing out of social processes
are inextricably intertwined. Social moves have
computational as well as social consequences.
 1156/1540

Computational moves have social as well as

computational consequences.

The first of these points is closely related to Lev

Vygotsky's notion of the social origins of higher
mental functions:

Any higher mental function necessarily goes

through an external stage in its development be-
cause it is initially a social function. This is the
center of the whole problem of internal and ex-
ternal behavior... When we speak of a process,
''external" means "social." Any higher mental
function was external because it was social at
some point before becoming an internal, truly
mental function. It was first a social relation
between two people. (Vygotsky 1981: 162,
quoted in Wertsch 1985)
 1157/1540

Vygotsky was, of course, aware that internalized

processes were not simple copies of external
processes: "it goes without saying that internal-
ization transforms the process itself and changes
its structure and functions" (Wertsch 1981:
163). For the sake of clear explication, no doubt,
and perhaps because the primary concern has
been with the development of young children,
many of the examples provided in the literature
of activity theory present cases in which the
structure of the external activity is evident and
the required transformations are fairly simple.
What happens if we consider adults learning
more complicated thinking strategies in more
complex social settings where the primary goal
of the activity is successful task performance
rather than education?
 Page 284

If social processes are to be internalized, then

the kinds of transformations that internalization
must make will be in part determined by the dif-
ferences between the information-processing
properties of individual minds and those of sys-
tems of socially distributed cognition. Let us
consider two such differences that were raised in
chapter 4 in the discussion of the navigation
activity in its individual and socially distributed
forms.
 1159/1540

First, socially distributed cognition can have a

degree of parallelism of activity that is not pos-
sible in individuals. While current research tells
us that much of individual cognition is carried
out by the parallel activity of many parts of the
brain, still, at the scale of more molar activities,
individuals have difficulty simultaneously per-
forming more than one complex task or main-
taining more than one rich hypothesis. These are
things that are easily done in socially distributed
cognitive systems. Ultimately, no matter how
much parallelism there may be within a mind,
there is the potential for more in a system com-
posed of many minds.
 1160/1540

Second, communication among individuals in a

socially distributed system is always conducted
in terms of a set of mediating artifacts (linguistic
or other), and this places severe limits on the
bandwidth of communication among parts of the
socially distributed system. Systems composed
of interacting individuals have a pattern of con-
nectivity that is characterized by dense intercon-
nection within minds and sparser interconnec-
tion between them. A cognitive process that is
distributed across a network of people has to
deal with the limitations on the communication
between people.
 1161/1540

Because society has a different architecture and

different communication properties than the in-
dividual mind, it is possible that there are inter-
psychological functions that can never be intern-
alized by any individual. The distribution of
knowledge described above is a property of the
navigation team, and there are processes that are
enabled by that distribution that can never be in-
ternalized by a single individual. The interpsy-
chological level has properties of its own, some
of which may not be the properties of any of the
individuals who make it up. This, of course, is
no challenge to Vygotsky's position. He didn't
say that every interpsychological process would
be internalized, only that all the higher mental
functions that did appear would get there by be-
ing internalizations of social processes.
 Page 285

That leads one to wonder whether there might

be intrapsychological processes that could not
be transformations of processes that occurred in
social interaction. Finding such a process would
be a challenge to Vygotsky's position, but unless
there are constraints on the possible transforma-
tions there is no way to identify such a process.
 1163/1540

Clearly there are higher mental processes that

could never have been realized in their current
form as interpsychological processes simply be-
cause they exploit the rich communication pos-
sible within a mind in a way that is not possible
between minds. Here is an example we have
already encountered: The task of reconciling a
map to a surrounding territory has as subparts
the parsing of two rich visual scenes (the chart
and the world) and then the establishing of a set
of correspondences between them on the basis
of a complicated set of conventions for the de-
piction of geographic and cultural features on
maps. As performed by an individual, it requires
very high bandwidth communication among the
representations of the two visual scenes. Very
occasionally, this task appears as a socially dis-
tributed task when a pelorus operator has no
idea how to find a particular landmark. In that
case, the restricted bandwidth of communication
between the pelorus operator (who can see the
 1164/1540

world) and the bearing recorder (who can see

the chart) makes the task virtually impossible.
The spatial relations implied by the locations of
symbols on the chart are simply too rich to be
communicated verbally in such a way that the
pelorus operator can discover the correspond-
ences between those verbally expressed rela-
tions and the relations among the objects he can
see in the world.
 1165/1540

Of course, it may be that the real difficulty here

is with the volume of information to be pro-
cessed, and that the actual technique for recon-
ciling map and territory is an internalization of a
social activity in an informationally sparser en-
vironment. Without a much more detailed ac-
count of the acquisition of this process, it will be
impossible to decide this case. For now, one can
do no more than raise the question of whether
internal processes might exist that are not intern-
alizations of external processes. And doing that
seems to throw the spotlight squarely on the
nature of the transformations that occur in the
internalization process.
 Page 287

7 Learning in Context
Throughout the previous chapters, I have tried
to move the boundary of the unit of cognitive
analysis out beyond the skin of the individual.
Doing this enabled me to describe the cognitive
properties of culturally constructed technical
and social systems. These systems are simultan-
eously cognitive systems in their own rights and
contexts for the cognition of the people who
participate in them. I have intentionally not at-
tempted to discuss the properties of individuals
before describing the culturally constituted
worlds in which those properties are manifested.
To do so is a mistake for two reasons.
 1167/1540

First, since cognitive science must content itself

for the foreseeable future with models of unob-
servable processes that are capable of generating
observable behavior, it is important to get the
right functional specification for the human cog-
nitive system. What sorts of things do people
really do? Many behavioral scientists seem to
think that they can answer this question by in-
trospection. I believe, on the contrary, that such
questions can be answered only by the study of
cognition in the wild. The representativeness of
contexts for the elicitation of behavior in labor-
atories is seldom addressed. What cognitive
tasks do people really engage in a normal day?
What is the distribution of such tasks? What
sorts of strategies do people invoke for dealing
with the tasks they do encounter? There is now a
growing literature based on studies of everyday
cognition. This literature is of value to cognitive
theory in the same way that the observations of
early naturalists were important to the
 1168/1540

development of a number of theories in biology.

A close examination of the context for thinking
may change our minds about what counts as a
characteristic human cognitive task.

Second, seeing human cognitive activity as an

integral part of such a larger system may bring
us a different sense of the nature of individual
cognition. Any attempt to explain the cognitive
properties of such a larger system without refer-
ence to the properties of its most active integral
parts would be deficient. Similarly, though, any
attempt to explain the cognitive properties of the
integral parts
 Page 288

without reference to the properties of the larger

system would also be incomplete.
 1170/1540

Human beings are adaptive systems continually

producing and exploiting a rich world of cultural
structure. In the activities of the navigation
team, the reliance on and the production of
structure in the environment are clear. This
heavy interaction of internal and external struc-
ture suggests that the boundary between inside
and outside, or between individual and context,
should be softened. The apparent necessity of
drawing such a boundary is in part a side effect
of the attempt to deal with the individual as an
isolated unit of cognitive analysis without first
locating the individual in a culturally construc-
ted world. With the focus on a person who is
actively engaged in a culturally constructed
world, let us soften the boundary of the indi-
vidual and take the individual to be a very
plastic kind of adaptive system. Instead of con-
ceiving the relation between person and envir-
onment in terms of moving coded information
across a boundary, let us look for processes of
 1171/1540

entrainment, coordination, and resonance among

elements of a system that includes a person and
the person's surroundings. When we speak of
the individual now, we are explicitly drawing
the inside/ outside boundary back into a picture
where it need not be prominent. These boundar-
ies can always be drawn in later, but they should
not be the most important thing.
 1172/1540

In this chapter I will attempt to partially dissolve

the inside/outside boundary and provide a func-
tional description of processes that could ac-
count for learning and thinking in the kind of
cognitive activity that has been described in the
previous chapters. In this and all that follows,
internal representations are identified by their
functional properties only. I make no commit-
ment to proposed mental mechanisms or compu-
tational architectures with which the behaviors
of the representations might be modeled. As far
as I can tell, it is not possible to distinguish
among competing models on the basis of avail-
able evidence, and it is certainly not possible to
do so on the basis of the sorts of evidence that
can be collected in the wild.
 1173/1540

Chapter 3 introduced the idea of a complex

functional system consisting of many media in
simultaneous coordination. The examples given
there included the functional system formed by
the bearing taker, the bearing recorder, and their
technological aids. When the team produces a
record of an observed bearing, the chain of co-
ordination may include the name of the land-
mark, partial
Page 289
 1175/1540

descriptions of the landmark, the visual experi-

ence of the landmark, the hairline of the alidade,
the gyrocompass scale, the knowledge and skills
involved in reading the bearing, the spoken
sounds on the phone circuit, the knowledge and
skills involved in interpreting the spoken bear-
ing, and the digits written in the bearing record
log. In chapter 3 I tried to describe the function-
al properties of some of those internal struc-
tures, but I did not address the question of how
the internal structures could actually develop as
a consequence of particular experiences, or how
a watchstander could manage to get the right
things into coordination to form useful function-
al systems. I couldn't treat those things there be-
cause I had not yet provided a reasonable de-
scription of the other (observable) parts of the
dynamic system in which "internal" structures
form. Dealing with these issues requires an ar-
chitecture of cognition that transcends the
boundaries of the individual.
 1176/1540

The proper unit of analysis for talking about

cognitive change includes the socio-material en-
vironment of thinking. Learning is adaptive re-
organization in a complex system. It is difficult
to resist the temptation to let the unit of analysis
collapse to the Western view of the individual
bounded by the skin, or to let it collapse even
further to the "cognitive" symbol system lying
protected from the world somewhere far below
the skin. But, as we have seen, the relevant com-
plex system includes a web of coordination
among media and processes inside and outside
the individual task performers. The definition of
learning given here works well for learning situ-
ated in the socio-material world, and it works
equally well for private discoveries made in mo-
ments of reflective thought. In this chapter I take
up the question of the formation of internal
structures as a consequence of experience.
 1177/1540

The richness of the universe of possible solu-

tions to real-world problems is often taken as a
reason not to study in the wild. In experimental
design, it is important to ensure that all subjects
in a particular condition are doing the task in the
same way. If subjects are using many different
strategies to solve a problem, it will not be pos-
sible to infer the representational bases of the
performances by comparing the behaviors of
subjects in one condition with those of subjects
in another condition. The difficulty is, as Newell
(1973) says, that we cannot learn about underly-
ing processes by aggregating across methods.
However, the flexibility of the formation of
functional systems in response to real-world
tasks appears to be an important cognitive phe-
nomenon in its own right. This is a
 Page 290

phenomenon that is entirely missed by research

paradigms that, for good reasons, intentionally
limit the methods subjects may use to perform a
task.
 1179/1540

The point of this chapter is to examine the pro-

cess of learning while doing with respect to a
sort of learning task that is frequently en-
countered in the world of navigation. I will at-
tempt to approach this learning problem from
the perspective of a softened boundary of the in-
dividual and to see the learning that happens in-
side an individual as simply adaptation of struc-
ture in one part of a complex system to organiz-
ation in other parts. Individual learning is the
propagation of some kinds of organization from
one part of a complex system to another. Some
of the parts that are reorganized are inside the
skin. It is not possible to understand how that re-
organization takes place without looking at the
other kinds of organization that are present in
the larger system.
 1180/1540

Many tasks in the world of navigation are de-

scribed by written procedures. When a person
uses a written procedure in the performance of a
task, the procedure is a mediating artifact. In or-
dinary usage, a mediating artifact stands
between the person and the task. It mediates the
relationship between the performer and the task.
On closer inspection, however, the situation be-
comes more complex. The stand-between read-
ing of mediation assumes that the task and the
performer can be bounded independently. Rath-
er than focus on the mediating artifact as
something that "stands between," I will view it
as one of many structural elements that are
brought into coordination in the performance of
the task. Any of the structures that are brought
into coordination in the performance of the task
can be seen as a mediating structure. It is diffi-
cult in this context to say what stands between
what, but they certainly all participate in the or-
ganization of behavior. The question of
 1181/1540

individual learning now becomes the question of

how that which is inside a person might change
over time as a consequence of repeated interac-
tions with these elements of cultural structure.

The phenomena of mediated performance are

ubiquitous. For the purposes of exposition, I
have chosen as an example a simple explicit ex-
ternal mediation device: a written procedure.
Many tasks in our culture are mediated by writ-
ten procedures or procedurelike artifacts, but
even considering all of them would not begin to
approach the full range of mediated perform-
ances. Language, cultural knowledge, mental
models, arithmetic procedures, and rules of lo-
gic are all mediating structures too. So are
traffic lights, super-
 Page 291

market layouts, and the contexts we arrange for

one another's behavior. Mediating structure can
be embodied in artifacts, in ideas, in systems of
social interaction, or in all of these at once. I
have chosen the written procedure because it is
a typical artifact in the world of ship navigation
and because it provides a relatively explicit ex-
ample of mediation for which a relatively simple
exposition can be given.
 1183/1540

The task of learning a procedure is interesting

because it can be mediated by so many different
kinds of structures. This is just another way of
saying that there are many possible ways of or-
ganizing functional systems to perform a task.
After discussing the mediation of procedural
performance by the written procedure, I will
consider other forms of mediation. This wide
range of ways to organize the performance of
procedures raises the question: What kind of ar-
chitecture of cognition is required to accom-
modate the flexible constitution of functional
systems of so many kinds?
 1184/1540

Putting the question of the flexible constitution

of functional systems first means approaching
the study of cognition from a different starting
point. It requires a different view of cognition,
and it demands that our models of cognition be
capable of different sorts of computations. This
is a consequence of an attempt to build a theory
of cognition that comes after, rather than before,
a description of the cultural world in which hu-
man cognitive behavior is embedded.

Theoretical Perspective for the Construction of a

Model

I take the fundamentals of an architecture of

cognition and a sense of a unit of analysis from
Gregory Bateson, who said that "the elementary
cybernetic system with messages in circuit is, in
fact, the simplest unit of mind; and the trans-
form of a difference traveling in a circuit is the
elementary idea" (1972: 459).
 1185/1540

The problem of how to bound a unit of analysis

is crisply summed up by Bateson in his well-
known example of a blind man with a stick:

Suppose I am a blind man, and I use a stick. I go

tap, tap, tap. Where do I start? Is my mental sys-
tem bounded at the handle of he stick? Is it
bounded by my skin? Does it start halfway up
the stick? But these are nonsense questions. The
stick is a pathway along which transforms of
difference are being transmitted. The way to
 Page 292

delineate the system is to draw the limiting line

in such a way that you do not cut any of these
pathways in ways which leave things inexplic-
able. (ibid.)
 1187/1540

The proper unit of analysis is, thus, not bounded

by the skin or the skull. It includes the socio-
material environment of the person, and the
boundaries of the system may shift during the
course of activity. Temporal boundaries are im-
portant too. As the analysis of the construction
of the task environment in chapter 3 showed, ar-
bitrary boundaries on the temporal extent of the
unit of analysis also risk cutting pathways in
ways that leave things inexplicable. In the
present context, many things remain inexplic-
able until we consider the history of the person
in the task environment. This seems especially
pertinent to the nature of learning, since learning
must be a consequence of interaction with an
environment through time.
 1188/1540

In a section titled "External Representation and

Formal Reasoning," Rumelhart et al. (1986)
sketch a proposal for a view of symbolic pro-
cessing that fits well with what has been pro-
posed here. They describe doing place-value
multiplication with paper and pencil as follows:

Each cycle of this operation involves first creat-

ing a representation through manipulation of
the environment, then a processing of the (actu-
al physical) representation by means of our
well-tuned perceptual apparatus leading to fur-
ther modification of this representation. By do-
ing this we reduce a very abstract conceptual
problem to a series of operations that are very
concrete and at which we can become very
good.... This is real symbol processing and, we
are beginning to think, the primary symbol pro-
cessing that we are able to do.
 1189/1540

They account for internal symbol processing as

follows:

Not only can we manipulate the physical envir-

onment and then process it, we can also learn to
internalize the representations we create, "ima-
gine" them, and then process these imagined
representationsjust as if they were external.

With experience we learn about the regularities

of the world of external symbolic tokens and we
form mental models of the behaviors of these
symbolic tokens that permit us to perform the
manipulations and to anticipate the possible ma-
nipulations. With even more experience, we can
imagine the symbolic world and
 Page 293

apply our knowledge, gained from interactions

with real physical symbol tokens, to the manipu-
lations of the imagined symbolic worlds.
"Indeed," Rumelhart et al. note, "we think that
the idea that we reason with mental models is a
powerful one precisely because it is about this
process of imagining an external representation
and operating on that."

These ideas also apply to interactions among in-

dividuals. Rumelhart et al. propose that we can
also imagine aspects of interactions and then op-
erate on or with the image of external
representations:
 1191/1540

We can be instructed to behave in a particular

way. Responding to instructions in this way can
be viewed simply as responding to some envir-
onmental event. We can also remember such an
instruction and "tell ourselves" what to do. We
have, in this way, internalized the instruction.
We believe that the process of following instruc-
tions is essentially the same whether we have
told ourselves or have been told what to do.
Thus, even here, we have a kind of internaliza-
tion of an external representational format (i.e.,
language).
 1192/1540

The practice of navigation includes many in-

stances of this kind of instruction. In an example
given at the end of chapter 6, a novice and an
expert quartermaster organized their activity
around a written form. That example was actu-
ally more complex than any considered by
Rumelhart et al., but the processes they propose
should apply here too. In that example, the
novice organized his own actions by coordinat-
ing them with the actions of the expert. The de-
velopment of sequential control of action also
concerns the relationship of public and private
symbol systems and the processes that link
them.

In this chapter I will integrate the functional-

systems perspective with Bateson's unit of ana-
lysis and Rumelhart's notions of imagining ex-
ternal representations into an account of how in-
ternal structures may form as a consequence of
experience.
 1193/1540

Constructing Action Sequences

There are many ways to construct the sequence

of actions that constitutes the fix cycle. Quarter-
masters who get navigation training in school
first encounter the sequence of actions of the fix
cycle as a written list of steps to be performed.
 Page 294

Solution procedures for many navigation tasks

are presented to students in the form of what are
called ''strips." A strip is a list of labeled blank
spaces which the student is supposed to fill in
order. The strip guides the student through a se-
quence of steps that produces a solution to the
problem at hand. One quartermaster chief com-
plained to me that strips don't give the learner
any command of conceptual structure. Students
who learn to fill in the blanks on a strip have no
idea what they have done, and are unable to per-
form the task in the absence of the strip. In what
follows, I will show why strips don't necessarily
provide conceptual understanding. I will also
show how conceptual knowledge can be added
to the sequential knowledge.
 1195/1540

The presupposition of much of military educa-

tion is that memorizing the sequence of ele-
ments in a procedure will lead to successful per-
formance. There are, however, many interpreta-
tions of what it might mean to say that a watch-
stander has "remembered the sequence of ele-
ments in the procedure." This might be taken to
mean that the names of the elements have been
remembered such that they could be recited in
order. Perhaps when the procedure is well
learned the name of each element gives way to
the name of the next, around and around the
cycle, like the words of the chorus of a familiar
song. Another interpretation of the memory for
the procedure consists of a sequence of mental
images of the elements. Perhaps the watchstand-
er can imagine the actions to be taken and can
see the unfolding sequence of elements as if
watching a movie in his "mind's eye." Yet an-
other interpretation for the memory of the ele-
ments of the task consists of a motor memory
 1196/1540

for the sequence of actions involved in doing the

elements of the task. Perhaps a watchstander
who has this sort of memory can simply initiate
the task and observe as his hands remember
what to do. The differences between these inter-
pretations of the phrase "remembered the se-
quence of elements in the procedure" are non-
trivial. In this chapter I will argue that the cor-
rect interpretation might simultaneously include
all these sorts of memory and more.

In order for a sequence to be remembered in any

form, it must have been experienced in some
fashion. This experience may have come in
many forms. Chapters 4, 5, and 6 described
many aspects of the organization of the learning
environment. For the moment, let us consider
the relationships among these different sorts of
representations of task structure.
 Page 295

Written Procedure as Mediating Structure

Consider a person using a written procedure to

organize the performance of a task in a case
where it is essential that the actions of the per-
formance be taken in a particular order and that
all the actions be taken before the performance
is judged complete. Here is a written procedure
for the quartermaster in Standard Steaming
Watch:

1. Choose a fix interval and a first fix time.

2. Choose a set of landmarks and information

sources for the fix.

3. Just prior to the fix time, go to the chart house

and record the fathometer reading.

4. At the fix time, observe the bearings of the

landmarks. (Observe landmarks near the beam
of the ship first.)
 1198/1540

5. Record the observed bearings in the bearing

record book.

6. Plot the observed bearings on the chart.

7. Compare the fix to the previously projected

position for this fix. (determine the effects of
current)

8. Compare the fix to the prior fix, measure dis-

tance and time difference.

9. Compute the ship's speed.

10. Determine the ship's heading.

11. Project the position of the ship two fixes into

the future.

12. Go to step 2 and repeat.

 1199/1540

An actor always incurs some cognitive costs in

coordinating with a mediating structure. To use
the written procedure above, for example, the
watchstander must control his attention and de-
ploy his reading skills. But some types of medi-
ated performance may be less costly to achieve
than others. The reduction of error or increase in
efficiency obtained via the use of the procedure
may compensate for the effort required to use it.
For the unskilled performer, of course, the task
may be impossible without the use of the pro-
cedure, so the economy of mediated perform-
ance in that case is clear.
 1200/1540

In order to use a written procedure as a guide to

action, the task performer must coordinate with
both the written procedure and the environment
in which the actions are to be taken. Achieving
coordination with the written procedure requires
the actor to invoke mental procedures for the
use of the written procedure. These include
reading skills and a strategy of sequential execu-
tion that permits the task performer to ensure
that the steps will be done in the correct order
and that each step will be done once and only
once (figure 7.1). The fixed linear spatial struc-
ture of the written procedure permits the user to
accomplish this by simply keeping track of an
index that indicates the first unexecuted (or the
last executed) item. Written procedures often
provide additional
 Page 296

Figure 7.1 Creating a sequential relationship. A strategy

of sequential execution in interaction with the physical
structure of the written procedure permits the task per-
former to take the elements of the procedure in a particu-
lar sequence. Here and in the rest of the figures in this
chapter, items in lightface are processes or structures in-
side the skin; items in boldface are structures outside the
skin. Seq is the sequential execution strategy. Si and Sj
are written steps in the procedure. The arrow indicates
that the item on the left constrains the development of the
one on the right. The trapezoidish shape to the left indic-
ates a complex coordination process; the rounded rect-
angle to the right indicates a sequential relationship.
 1202/1540

features to aid in the maintenance of this index:

boxes to check off when steps are completed, a
window that moves across the procedure, etc.
The mediating artifact may thus be designed
with particular structural features that can be ex-
ploited by simple interactive strategies to pro-
duce a useful coordination. If the items are writ-
ten in a list, the sequential relations among them
emerge from the interaction of the physical
structure of the list and a particular strategy for
reading it (from top to bottom, for example).
The spatial relations among items on the written
list become sequential relations in the interac-
tion. Notice, however, that the sequential rela-
tions among the steps of the procedure are im-
plicit in the physical structure of the list. They
become explicit only in interaction with some
sequential strategy. The top-to-bottom strategy
produces one set of sequential relations among
steps, but another strategy (say, bottom to top)
would produce a different ordering from the
 1203/1540

very same physical structure. It is important to

note, however, that some spatial structures af-
ford simple strategies whereas others do not. A
written procedure may not be needed if the spa-
tial relations among the things referred to by the
procedure permit the imposition of a simple
strategy. The walk-around inspection of an air-
plane provides an example. In the walk-around,
the pilot examines various parts of the airplane
for flightworthiness. This task is not usually me-
diated by a written procedure, because the spa-
tial arrangement of the parts of the plane them-
selves support a simple strategy that produces
sequence from space. The pilot adopts a "flow"
or trajectory of attention, going around the air-
plane clockwise starting at the boarding ladder
(for example). The items that lie ahead on that
trajectory are the things that remain to be in-
spected; those that lie
 Page 297

behind are those that have already been inspec-

ted. The pilot's own body acts as the delimiter
between the two sets. Every one of these inter-
action strategies can be seen as metamediation-
that is, a mediating artifact that organizes the
use of some other mediating artifact.
 1205/1540

In finding the next step to do in the written pro-

cedure, the actor applies the sequential execu-
tion strategy to the written procedure to determ-
ine which step is the next one, and perhaps to
determine an index of the next step that can be
remembered. There are two related issues con-
cerning this index: where it is located and what
it contains. The index could be encoded in the
memory of the actor, or the actor could take
some action on the world (making a mark on the
written procedure). If the index is simply a mark
on paper, the memory task is only to remember
what the mark means. If the steps are numbered,
the index can be a number. Thus, in the above
example, the quartermaster could remember that
all steps up to step 3 had been accomplished.
The index can also be the lexical or semantic
content of the step's description. The quarter-
master could remember that everything up to
shooting the bearings had been done. Each of
these alternatives requires a different set of
 1206/1540

coordinating actions to implement the sequential

execution strategy. For example, if the content
of the step index is the lexical or semantic con-
tent of the step itself, then finding the next step
(by reading through the steps until you arrive at
one that has not yet been done) and establishing
the step index (identifying a step that has not
been done) are the same action. If the content of
the step index is a mark on paper or a number to
be recorded or remembered, however, then
some action in addition to finding the next step
must be undertaken to establish the step index.
For example, you would have to find the last
mark and move to the step following it, or re-
member the number, increment it by one, and
then find the printed numeral that matches the
new step's index number.
 1207/1540

Although the primary product of the application

of this sequencing strategy is the determination
of the next step to be performed, either the writ-
ten procedure as an object in the environment or
the internal procedure that implements the se-
quential execution strategy may also be changed
as a consequence of the actions involved in find-
ing the next step.

Having generated a step index (in whatever

form), the actor can bring that index into co-
ordination with the written procedure to focus
attention on the current step. Though the goal of
using the
 Page 298

written procedure as a mediating artifact is to

ensure sequential control of the actions taken in
the task domain, it is clear that the task of bring-
ing the written procedure into coordination with
the domain of action may not itself be linearly
sequential. For example, if a user loses track of
the step index, in order to determine the next
step to be taken he may go back to the begin-
ning of the written procedure and proceed
through each step in the procedure, not execut-
ing it but asking of the task world whether the
expected consequences of the step's execution
are present. When a step is reached whose con-
sequences are not present in the task world, one
may assume that it has not yet been executed.
This is a simple illustration of the potential com-
plexity of the metamediation that may be under-
taken in the coordination of a mediating struc-
ture with a task world.
 1209/1540

It is clear from this discussion that the symbols

in figure 7.1 are oversimplified and hide much
of the potential complexity of this relatively
simple task. The remaining figures in this
chapter hide similar complexities. If all these
were included, however, it would be impossible
to assemble the pieces into a coherent whole.

Once the current step has been identified, the

user may coordinate its printed representation
with shallow reading skills in order to produce
an internal representation of what the step says
in words (figure 7.2).
 1210/1540

The shallow reading skills here refer to organ-

ized (perhaps already automated) internal struc-
tures that can create internal representations of
words from their external printed counterparts.
The representation of what the step says may be
in the visual or the auditory modality, or perhaps
in both. Whether this internal representation is
primarily auditory or visual or something else is
not important for the present argument. There is
some evidence that meanings are accessed both
via direct lexical access (from word to meaning)
and via sound codes (from word to sound rep-
resentation to meaning), even in silent reading
(Pollatsek and Rayner 1989).

Cognition p298

Figure 7.2 Finding what the step says. Sj is the written

representation of the ith step of the procedure;
sj is an internal representation of what that step says; Read
 1211/1540

is the shallow reading skills of the person.

The arrow indicates the propagation of representational
state from one medium to another.
The complex trapezoid represents a complex coordination
process.
 Page 299

The internal representation may even be in a

tactile or haptic modality, if the procedure was
written in Braille. The important thing is that the
representation be capable of permitting the actor
to "remember" the lexical content of the step at
a later time. It is obvious that this process may
proceed concurrently with the process of read-
ing what the step means. However, I have separ-
ated shallow and deep readings, primarily be-
cause shallow and deep readings produce differ-
ent sorts of products that can be shown to exist
independently. Thus, a user who does not under-
stand the domain of action may know and be
able to recall what a step "says" without having
any idea of what it "means." The default case for
this example will be a written procedure. The
external mediation is often provided verbally by
another person. In that case, the shallow reading
skills are replaced by listening skills.
 1213/1540

Most models of reading involve the construction

of meaning in the absence of the world de-
scribed. However, reading a procedure is direc-
ted toward understanding a world that is present.
Determining the meaning of what a step says re-
quires the coordination of what the step says
with the task world via the mediation of a deep-
er sort of reading (figure 7.3). This deep reading
relies on two internal structures: one to provide
semantic mappings from linguistic descriptions
provided by the procedure to states in the world
and another to provide readings of the task
world to see what is there. What the words in
the step description are thought to mean may de-
pend on the state of the task world that has been
produced by prior actions. The arrow goes both
ways between the representation of what the
step says and what it means because each de-
pends on the other. The words we think we saw
or heard may depend on what we think makes
sense. Let us assume for the moment that the
 1214/1540

meanings that appear in this medium are im-

agelike and that they are the same kinds of
structures that would result from actual perform-
ance of the task.

It is tempting to think that the words and the

world are coordinated by language in order to
produce the meanings. It is more

Cognition p299

Figure 7.3 Discovering what the step means. L is deep

language skills;
TW is the task world; Mi is the meaning of the ith step.
Page 300
 1216/1540

accurate to say that the meanings, the world, and

the words are put in coordination with one an-
other via the mediating structure of language. It
is difficult to place the meaning of the step
cleanly inside or outside the person, because
some component of the meaning may be estab-
lished by a kind of situated seeing in which the
meaning of the step exists only in that active
process of superimposing internal structure on
the experience of the external world. That is, at
some point in the development of the task
performer's knowledge the step may not have a
meaning in the absence of the world onto which
it can be read. Perhaps the meaning of a step can
reside cleanly inside a person only when the
person has developed an internal image of the
external world that includes those aspects onto
which the mediating structure can be superim-
posed. The structure of language may be
changed by its use, and what is thought to be in
the world may be changed by describing it in a
 1217/1540

particular way. Each of the structures provides

constraints on the others, and all are to some ex-
tent malleable. The system composed of a task
performer, mediating structures, and the task
world settles into a solution that satisfies as
many constraints as is possible. The arrow in
figure 7.3 goes both ways because there are mu-
tual constraints between these structures. This
constraint satisfaction is a computation. We
must keep in mind, however, that side effects of
the computation may include changes in the
constraint structures.
 1218/1540

Finally, having determined what the step means,

the user of the procedure may take actions on
(and in) the world to carry out the step (figure
7.4). The action, like the meaning of the step,
may be difficult to locate cleanly inside or out-
side the actor, because actions taken on the en-
vironment involve phenomena inside and out-
side the actor and because for many mental acts
(those based in mental imagery, e.g.) the task
world itself may be substantially inside the act-
or. In any case, the meaning of the step, the ac-
tion, and the task world are brought into co-
ordination. The arrow between meaning and ac-
tion goes both ways because the meaning of the

Cognition p300

Figure 7.4 Performing the step. Mot is the motor orienta-

tion process;
 1219/1540

Ai is the action that is taken to realize the ith step of the

procedure.
 Page 3

step is used to organize the action, and the monitore

performance of the action may change the understo
meaning of the step. Remember that the meanings w
have discussed so far are equivalent to the sensory
experiences encountered in the actual performance
the task. Having completed this step, the user of the
procedure may find the next step and continue.

In an actual performance of a step of the procedure

all the structures discussed so far may simultaneous
be in coordination with one another as is shown in
figure 7.5.

Consequences of Mediated Task Performance

 1221/1540

While the procedure is being followed, high-level o

ganization of task-related behavior is producedin pa
by the physical structure of the written representatio
of the procedure. The interaction with the procedur
produces for the actor a sequence of experiences of
step descriptions. Each of these experiences may
have several components: what the step says, what
the step means, and the actions in the task world tha
carry out the step. Figures 7.1-7.4 show that many
layers of transforming mediating structure may lie
between a simple mediating artifact (e.g., a written
procedure) and the performance of a task. Now sup
pose a task performer uses this written procedure to
guide many performances of the task. What are the
consequences for structures inside the actor of the r
peated achievement of the coordinations depicted in
figures 7.1-7.4? How might internal structures deve
op as a consequence of interactions with external
structures?
 1222/1540

The discussion above introduced three functionally

distinct internal media: a lexical medium dedicated
representing what the steps of the written procedure
say, a semantic medium dedicated to

Cognition p301

Figure 7.5 Multiple simultaneous coordinations in the perfor

ance of a step of a written procedure.
Shallow reading, deep reading, and motor orientation media
the propagation of state from the printed
representation of the step (Si), to its lexical representation (s
thence to its semantic representation (Mi),
thence to the motor sequences that constitute the performance of the step (A
 Page 302

representing what the steps mean, and a motor

medium dedicated to effecting the actions taken
in the task world. Each medium holds structure
of a particular kind. Now, imagine that each me-
dium has the simple property that it will, as a
consequence of being driven through a sequence
of states, come to have the capacity to reproduce
that sequence of states. That is, when placed in
any state in the sequence, the medium can pro-
duce the next state of the sequence, and then the
next, and so on to the last state in the sequence.

LEARNING THE SEQUENCE OF STEPS

DESCRIPTIONS
 1224/1540

Consider the lexical medium. As the task per-

former reads the steps of the procedure, this me-
dium is driven first into a state that represents
what the first step says, then into a state that rep-
resents what the second step says, and so on
through the entire procedure. This sequence of
states produced by the shallow reading of the
written step descriptions is a mediated sequence.
If the lexical medium has the property described
above, with repeated exposures to the sequence
it will come to be able to reproduce the se-
quence of states that represents what the steps of
the procedure say (figure 7.6). It will be able to
do this without the mediation of the sequential
interaction strategy applied to the written pro-
cedure and the reading skills. The sequential re-
lations among the representations of what the
steps say were originally mediated by the se-
quential relations of the items on the written
procedure. With experience, these sequential re-
lationships become unmediated.
 1225/1540

This newly created internal representation of the

sequence of what the steps say is of the same
class of phenomena as our knowledge of the or-
der of the letters of the alphabet or of the names
of the lower numbers. Originally constructed
through complex mediation processes, it gains
some modularity and autonomy as a con-
sequence of experience. The lexical representa-
tion of each step

Cognition p302

Figure 7.6 Producing an internal sequence of states. Ap-

plying the sequential execution strategy
to the written procedure produces a sequence of internal
states (indicated by the rounded rectangle at right).
 Page 303

of the sequence is explicitly re-represented by a

state that this medium is capable of producing,
but the sequential relations among the states are
implicit in the behavior of the medium.
 1227/1540

Loosely speaking, we could say that the lexical

medium has internalized the written procedure.
We must use the word 'internalize' with caution,
however. The process in question here is spe-
cifically the development of a medium that,
when placed in a state corresponding to the ex-
perience of what step N says, will automatically
undergo a transition to a state corresponding to
the experience of what step N+1says. In a literal
sense, nothing has moved from outside to inside.
Via the mediation of the shallow reading pro-
cess, structure in the written representation of
the steps of the procedure has given rise to cer-
tain internal experiences. Then, as a con-
sequence of repeated experience, a new func-
tional ability has been created out of the entirely
internal states of the lexical medium. Careless
use of 'internalization' here is dangerous because
it glosses over the processes involved and lumps
together many kinds of representations that
 1228/1540

differ from one another in functionally signific-

ant ways.

Once the lexical medium has developed the ca-

pacity to produce, in order, representations of
what the steps say, it may become the con-
trolling structure for subsequent performances.
(See figure 7.7.) This amounts to the task
performer's having learned what the procedure
says, so that instead of reading the next step he
can "remember" what the next step says, use
that to construct the meaning of the next step,
and use that meaning to organize an action. A
performance guided by the memory of the
words in the procedure is still a mediated task
performance, but the mediating structure is now
internal rather than external. The lexical medi-
um that encodes
 1229/1540

Cognition p303

Figure 7.7 The mediation of task performance by the

learned sequence of step descriptions.
The internal representations of what the steps say can be
used to control the sequence of actions.
 Page 304

what the steps of the procedure say provides ex-

plicit representations of the steps of the proced-
ure. It can move through a sequence of states,
each of which corresponds to the experience of
reading what a step of the procedure says. Mov-
ing from external to internal mediation also in-
troduces new possibilities for the relations
between the actor and the environment, because
the environment no longer need contain the me-
diating structure.
 1231/1540

The lexical medium must become an automat-

ized system before it can be used alone to medi-
ate the task performance. This internal medi-
ation system, while having explicit representa-
tional content in its states, relies for its con-
trolling behavior on automatized implicit encod-
ings of relations among its states. The issue of
what is implicit and what is explicit depends on
the question being asked. The internal memory
for the procedure consists of states that represent
explicit descriptions of the actions to be taken.
But the sequential relations among those step
descriptions are implicitly encoded in the beha-
vior of the lexical medium, much as the sequen-
tial relations among the step descriptions in the
external procedure were implicitly encoded in
the spatial relations among elements on the
written-procedure artifact. Consider briefly an-
other common mediating structure: alphabetical
order. It is used in many storage and retrieval
schemes in our culture, so we take care to
 1232/1540

ensure that children learn it. In learning the al-

phabet song, the child is developing an explicit
internal automatized version of the alphabet's
structure. The content of the statesthe words of
the songare explicit, but the sequential relations
among them (which were originally provided by
another mediating system, a teacher) are impli-
cit. A child who knows the song can tell you
what comes after P (perhaps after singing the
names of the first 17 letters), but that same child
will have a difficult time saying why Q follows
P.There is simply no explicit representation of
that in what the child knows.
 1233/1540

When a person first performs a task using writ-

ten instructions, there is an apparent alternation
between coordination with the written procedure
and coordination with the world. One deals first
with the written procedure and then with the
world it describes. However, no alternation of
attention is necessary once one has developed an
internal representation of even the lexical level
of the procedure description. Then the coordina-
tion with the representation of the procedure
(whether lexical, semantic, or motor) and the
world in which the procedure is carried out is no
longer one of al-
 Page 305

ternation of attention focus; it is one of simul-

taneous coordination. Understanding a step in
the description may depend on understanding
the state of the world in which it is to be carried
out. The experience of the meanings of the de-
scriptions of the steps contains experience of the
task world, and the doing of the actions contains
the experience of the meaning of the task steps.
The importance of this is that in this mediated
performance the actor becomes a special sort of
medium that can provide continuous coordina-
tion among several structured media. It fact, the
alternation of focus of attention when one is us-
ing a written procedure is a solution to a prob-
lem of competing demands for visual resources
that is created by the particular physical instanti-
ation of the written procedure. Alternating the
focus of attention is simply a way to time-share
a particular scarce perceptual resource.
 1235/1540

LEARNING THE SEQUENCE OF STEP

MEANINGS

Of course, at the same time that the medium

dedicated to the representation of what the steps
say is driven through a series of states, the medi-
um dedicated to representing the meanings of
the steps is also driven through a series of states.
This is shown in figure 7.8. Once this semantic
medium has been trained, the actor can remem-
ber the meanings of the steps, if necessary
without reference to the memory of what the
steps say. Since the lexical structure is around,
however, and since humans are unrelentingly
opportunistic, it is likely that both the memory
of the meaning of the step and the meaning de-
rived from interpreting the memory of what the
step says will be used in concert to determine
the meaning of the step.
 1236/1540

Figure 7.8 describes one way to establish the se-

quential relations among the meanings of the
steps of the procedure. There is another way that
is mediated by conceptual knowledge about the
task. The

Cognition p305

Figure 7.8 Automatization of the step meaning sequence

in the semantic medium.
Once trained, the semantic medium can produce the suc-
cession of states corresponding
to the meanings of the steps of the procedure.
meanings of the steps represented in this semantic m
gined and observed courses of action. There are oth
ings, however, that concern the conceptual relations
of a procedure. The steps of the fix-cycle procedure
related by a set of computational dependencies. It is
plot a line of position until an observation of a land
made, and it is not possible to make the observation
has been chosen to observe. These kinds of depend
the possible orderings of steps in the procedure, and
quartermaster remember what comes next. Let Mp b
of plotting a line of position, and let Mpbe a represe
concept ''plotting a line of position." Let Mo be the
making an observation of a landmark and let Mo be
of the concept "making an observation of a landmar
conceptual and computational dependency between
that Mo must precede Mp. This relationship explains
the sequential ordering of Mo and Mp(figure 7.9).
 1238/1540

The meanings of the steps would have only implici

another were it not for the potential mediating role
knowledge in the task domain. If conceptual knowl
meanings of the steps, some other medium in the sy
states that explicitly represent a reason why step N
However, such a mediating structure need not be le
sequence of meanings of the steps is learned. Some
why we do some task the way we do long after we h
the task itself.

While I was working on a program to improve rada

ing, I interviewed many navigation instructors. The
complex set of plotting procedures. One of the instr
me that he had spent 3 years at sea doing radar navi

Cognition p306

Figure 7.9 The use of conceptual knowledge to establish seq

among meanings of steps
 1239/1540

in the procedure. The relationships among the meanings of th

ated by knowledge
of the conceptual and computational dependencies among the
ceptual
dependency that holds between Mo and M
 Page 307

tion procedures before he realized the conceptu-

al relationship between relative and geographic
motion. He told me that the insight came to him
as he lay in his bunk one night. He got out of
bed and rushed down to the combat information
center to confirm that his newfound understand-
ing was indeed correct and that it accounted for
the organization of the procedures he had been
executing for years. This story is more dramatic
than most, but the phenomenon is far from rare.
I believe that much of our learning consists of
filling in conceptual details and relationships in
tasks we already know how to do.
 1241/1540

Problem solving and planning in the space of

conceptual dependencies are two of the means
of mediating this task. These kinds of activities
have been the mainstay of an important tradition
of research in cognitive science and artificial in-
telligence. In the case of real-world procedures,
however, there are so many other sources of
structure that it may be relatively rare to find a
person working from the conceptual dependen-
cies alone to determine a sequence of action.

LEARNING THE SEQUENCE OF ACTIONS

 1242/1540

Whether the task is organized by an external

procedure or by an internal representation of it,
the mental apparatus involved in the perform-
ance of the task is driven through a sequence of
states. Because of the nature of the structured in-
teraction of the task performer with the environ-
ment, the sequence of states is repeated more or
less consistently each time the procedure is fol-
lowed. The motor medium begins to encode the
sequential relations among the successive states
(figure 7.10).

Something of the organization of the (N+l)th

state is in the potential of the medium when the
Nth state is present. The capacity of the motor
medium to produce the sequence of states that
 1243/1540

Cognition p307

Figure 7.10 The development of sequential relationships

in the motor medium.
The motor medium can now produce the succession of
states that generate the actions taken in the task world.
There is no longer a need for any mediation via the mean-
ings of steps.
 Page

Cognition p308

Figure 7.11 The assembled tissue of coordination. In expert

formance, the succession of states produced depends on both
zontal and vertical coordination. Every state constrains and is
strained by others. (The conceptual relations among meanin
which also provided constraints, have been omitted from this
ure because of the limitation of a two-dimensional diagram
omitted from this figure because of the limitation of a two-dim
sional diagram.)
 1245/1540

is experienced in and that constitutes the performan

of the task is no different from the capacity of the o
media to produce their respective sequences of state
But unlike the states in the lexical and semantic me
the states of the motor medium are not descriptions
actions. They are not referential; they are not about
anything. They are the states of the motor medium
constitute the performance of the actions. There is n
mediation between the states of the motor medium
the action. When the sequence of these states has be
learned, the medium, once placed in state 1, can do
task automatically without reference to any explicit
representation of the either the steps or their sequen
relations.
 1246/1540

The mediated performances leading up to this cond

tion could be thought of as training trials for the me
um that produces the action. The system has now
reached the condition described by figure 7.11. In t
condition, for a normal task performance, the moto
medium no longer needs the organizing constraints
the meanings of the steps. Once placed in the initial
state, the motor medium simply moves through the
states that constitute the doing of the task. This is th
nature of automatized skill performances-performan
that no longer utilize the organizing constraints of
complexly mediated structure. Of course, if unusua
circumstances arise in the task world, the automatiz
performance may fail, requiring additional recourse
other mediating structure.

This is a simple model in which a person is situated

a sociomaterial world and learning is adaptive reorg
ization of part of the system in coordination with or
ganization in other parts of the system. The system
erates by the propagation of representational
1247/1540
 Page 309

state across media, and the media themselves

acquire functional organization as a con-
sequence of the repeated impositions of repres-
entational state upon them. The lexical medium,
for example, acquires the ability to move se-
quentialy through states that can be read as rep-
resentations of what the step says. The basic
building blocks of the system are the coordinat-
ive processes that move representational state
(horizontally in figure 7.11) and the functional
properties of the representational media that per-
mit them to learn to move through a sequence of
states.
 1249/1540

I began this section with the question of what it

might mean that a watchstander can "remember
the elements of a procedure." The answer is
clearly not that something is retrieved from
memory as a physical structure might be re-
trieved from a storehouse, or even as a pattern
of bits might be retrieved from a magnetic stor-
age medium in a computer. Rather, remember-
ing is a constructive act of establishing coordin-
ation among a set of media that have the func-
tional properties such that the states of some can
constrain the states of others, or that the state of
one at time t can constrain its own state at time t
+ 1. The meaning of the next step in a procedure
can be partially established by sequencing from
the meaning of the current step. It can be con-
structed from the interpretation of the descrip-
tion of the next step in coordination with the
task world. It may be wholly or partially derived
from conceptual dependencies between it and
the other steps (not shown in figure 7.11). It can
 1250/1540

be in part derived from monitoring the real or

imagined motor sequences that realize it in the
world. The multiple representations of each ele-
ment of the procedure are woven into this tight
fabric of relationship and constraint. Remember-
ing is not a retrieval of an identifiable single
structure; rather, it is a process of construction
via simultaneous superimposition of many kinds
of constraints.
 1251/1540

If local structures are sufficient to produce un-

ambiguous states in the task-performance me-
dia, the other media may not be activated. A
very well-learned sequence of actions may run
off without conscious intervention. When snags
are encountered or when ambiguous states arise,
the range of structures that are brought into co-
ordination may expand to include other media.
Perhaps when automatic motor processes reach
an impasse, the semantic medium can provide a
representation of the next step that the motor
planning processes can use to construct a new
motor plan. That is, one can remember what the
step means in order to reconstruct a motor se-
quence to carry it out. When the relations in the
semantic
 Page 310

medium are insufficient to produce a new state,

the lexical medium can be brought into coordin-
ation with the semantic medium to produce a
description of what the next step says, and that
can be used to produce the meaning of the next
step. One can remember what the next step says
in order to reconstruct its meaning. And if all
that fails, one may construct a functional system
that again coordinates with the written proced-
ure to provide representations of the next step.
That is, one can reread the step from the written
procedure.
 1253/1540

While the procedure is being learned, organiza-

tion propagates from left to right and from top to
bottom in figure 7.11, and from outside to inside
and then back to outside. As learning proceeds,
a wave of organization moves across the media
and each medium acquires functional properties
that permit it to produce structured behaviors.
Just as the individual quartermasters follow the
same trajectory in their careers that the data fol-
low through the navigation team, the increment-
al spread of organization of media that is learn-
ing follows the same trajectory that representa-
tions follow in the course of a single perform-
ance of the task.
 1254/1540

The processes by which an individual learns to

perform the task can be seen as the propagation
of a wave of organization moving across a com-
plex set of media. Organization propagates from
external media to internal media and back to ex-
ternal media. The changes that happen inside an
individual as a consequence of learning are ad-
aptations in a part of a larger dynamical system
to organization or structure that is present in
other parts of the system.

WHY WE CAN'T SAY WHAT WE DO

A common observation concerning automatized

skill is that skilled performers may have diffi-
culty saying how they do what they do. This
analysis provides three explanations for this
phenomenon.
 1255/1540

First, the automatized motor medium for the

procedure is a way of producing in the relation
of the person to the environment a sequence of
actions that constitute the doing of the steps de-
scribed by the procedure. Because it encodes a
relationship between the person and the environ-
ment, the execution of the procedure by the
automatized motor medium requires the cooper-
ation of the environment in a way that remem-
bering the procedure does not. For example, the
attempt to do a step can be frustrated by the lack
in the environment of something required by the
step. Yet one may remember a description of a
step even though the conditions re-
 Page 311

quired to carry it out are absent. In the example

above, the actor may be forced by the lack of the
required condition to do some other actions in
preparation for the previously frustrated step. In
giving an account of how to do a task, the per-
former must assume a world (perhaps more cor-
rectly, the report must presuppose a world) in
which the described actions make sense. Except
where the task in question occurs in a very
stable set of environments, the assumed world is
certain to differ from many of the actual worlds
in which the task is attempted, and the descrip-
tion will therefore fail in many of the actual
worlds in which the task is performed.
 1257/1540

Second, the reports skilled performers can give

are sometimes based on the mediating structures
that were used to control their behavior while
they were acquiring automatized skill. The ac-
counts that are given, being descriptions of me-
diating structure, may be just what is needed to
communicate the skill from one person to anoth-
er because the only way to produce the automat-
ized skill in any medium is to have the medium
learn it from experience and the only way for a
novice to experience it is by getting into co-
ordination with mediating structure. However, if
the memory for the mediating structure has at-
rophied as a result of long disuse during auto-
matized performance, then an expert asked how
he does something may simply have no mean-
ingful answer to give. The automated system
does what it has been trained to do, but it has no
explicit representation of what it is doing. The
representation of what it is doing exists only in
 1258/1540

the apparatus that provided the training, that is,

the mediating structure which is now atrophied.

A third situation that results in the expert task

performer's inability to account for his own task
performance arises when the mediating structure
is present as constraints in the environment that
shape the development of the motor medium
directly, without the development of internaliza-
tions of explicit mediating representations. This
seems to be the case for many motor skills.
When asked to describe how the skill is per-
formed, such an expert may describe events in
which the skill was manifested. One view of
such a response might be that the expert is being
uncooperative, but when we understand that the
mediating structure was in the environment of
the skill acquisition we see that describing
events in which the skill was manifested is the
best the expert can do to describe the mediating
structure under which the skill developed.
1259/1540
 Page 312

With this example I have attempted to highlight

the complexity and richness of interaction of
mediation structures of different sorts in the per-
formance of what seemed at the outset to be a
relatively simple mediated task performance. I
don't think this analysis should lead us to change
our minds about the relative simplicity of using
procedures. On the contrary, I hope it heightens
our awareness of the diversity of the kinds of
mediating structure that come into play in every-
day cognitive activities.

OTHER KINDS OF MEDIATING

STRUCTURE
 1261/1540

The example presented at the end of chapter 6

involved mediation by both a written artifact
(the position report form) and the behavior of
another person (the chief petty officer ,who
asked the novice a set of questions in sequence).
Clearly the source of the step descriptions can
be an interaction with another person rather than
with a written procedure. If the mediating struc-
ture is provided by the activities of another per-
son, the novice who has internalized that struc-
ture can then act alone. This echoes Vygotsky's
(1978) general genetic law of development in
which mediating structure appears twice: first in
interpsychological processes and second in in-
trapsychological processes.
 1262/1540

There is much more to "internalization,"

however, than simply imagining an internal con-
versation. It is not that some content is copied
from the outside world into some internal stor-
age medium. Rather, the process of interaction
creates a new process. Notice, for example, that
even the lexical medium (closest to the surface
of any of the media proposed here) has a se-
quence of states implicitly encoded in its beha-
vior. The sequential relations among states were
not a property of the medium from which the
states were learned; rather, they were a property
of a particular pattern of interaction with the ex-
ternal medium. Internalization has long con-
noted some thing moving across some bound-
ary. Both elements of this definition are mis-
leading. What moves is not a thing, and the
boundary across which movement takes place is
a line that, if drawn too firmly, obscures our un-
derstanding of the nature of human cognition.
Within this larger unit of analysis, what used to
 1263/1540

look like internalization now appears as a gradu-

al propagation of organized functional proper-
ties across a set of malleable media.

When individual task performance is considered

in the context of a larger system, individual
learning and mastery of the skills of a job appear
as a shift in the location of the mediating
structures
Page 313
 1265/1540

that constrain the organization of action. In all

cases, the mediating structures must exist some-
where in the functional system. In the case of
team performance, some of the constraints are in
the environment in the form of the behaviors of
the other members of the team. If the team is ex-
perienced, this means that there will be redund-
ant representation of the constraints on se-
quence, since they will exist both in the indi-
vidual actors and in the interactions among
them. Furthermore, conceptual dependencies
may be learned more or less directly from parti-
cipation in the team activity. As was noted in
chapter 6, the computational dependencies
among the steps of the procedure for the indi-
vidual watchstander are present as interpersonal
dependencies among the members of the team.
These dependencies need never be stated to be
learned. They are enacted in social relations, and
they can be learned as patterns of social interac-
tion rather than as words. The processes
 1266/1540

required to get from words to meaningful rep-

resentations of such dependencies are probably
as complex as the processes required to get from
patterns of social interaction to representations
of the same dependencies.

WHY WE TALK TO OURSELVES

 1267/1540

The next chapter describes activities in which

members of the navigation team enter data into
a navigation calculator. While pushing the but-
tons of the calculator, the quartermasters can be
heard to recite the names of the keys they press.
Why do we do this verbal shadowing? Consider
an example from everyday life: I don't seem to
be able to open the combination lock on the
storage shed in my yard without inwardly speak-
ing the numbers of the combination as I do it. I
have had years of experience with the lock. Yet,
in spite of my best efforts to suppress the num-
bers, I still say them inwardly as I spin the lock's
dial. Since I usually do this task entirely alone,
the speech must be there for its self-regulatory
function, not for any communicative function.
How could it be selfregulatory? It is interesting
to note that I do not verbalize the directions in
which the dial should be turned. The dial must
be turned clockwise to the first number, then
counterclockwise past the first number to the
 1268/1540

second number, then clockwise again to the last

number. I seem to have learned the sequence of
directions of turning in the action specification,
because I never say the names of the directions.
I also turn the dial several times in the clockwise
direction as I start without any verbalization. As
the index for the
Page 314
 1270/1540

dial nears the first number, I find myself saying

that number. When I go counterclockwise,
however, I not only say the second number of
the combination; I also tend to say, "back past
33, 21." The direction of turning may be in the
muscles, but the specification of how far to turn
is not. I remember the meaning of the numbers
by the appearance of the dial itself. For ex-
ample, I know that 21 is the mark just to the
right of the large tick labeled 20 without doing
any counting. I just recognize it as being right.
Even though I know the directions of turning in
the action plan, and I seem to know the appear-
ances of the three target numbers in sequence, I
still say the numbers subvocally. The sequence
of number names is a very stable structure for
me. The reason I recite them subvocally, I be-
lieve, is that when I say the numbers aloud I in-
terpret the meanings of the spoken numbers
right onto the medium in which I am construct-
ing the memory of the appearances of the
 1271/1540

numbers. In this way, I add constraints to the

process that is the reconstruction of a memory
that will drive the action. I superimpose multiple
representations of the same action in order to
produce the memory of the action. And the
memory of the action is in the doing of the ac-
tion itself. Of course, the action itself is also
contributing to the construction of the memory,
since I am monitoring the action and seeing
what it means. By subvocally shadowing my ac-
tion, I add a well-organized set of constraints to
an already well-constrained problem. As a res-
ult, the performance is very robust.
 1272/1540

In order to get useful mental work done, of

course, the actor must be capable of bringing
these structures into coordination. As we saw
with the coordination of the procedure with the
task world, bringing mediating structures into
coordination may require still more (metamedi-
ating) structures. The consequences of the lack
of this ability are encoded in our folk wisdom
about "book learning" versus experience. One
may have complete mastery over a major medi-
ating structure for some task without having de-
veloped any of the metamediation required to
put it to work in a real task environment.

Counting
 1273/1540

From the point of view developed here, follow-

ing a written procedure and counting have im-
portant structural similarities. Both involve the
coordination of a sequence of tags with a parti-
tioning of objects or events. In the case of
counting, a set of sequential transitions through
the names for the numbers is coordinated with
the
 Page 315

movement through the collection of individual

objects or events of the partition between those
things that have not yet been counted and those
things that have already been counted. Simil-
arly, following the procedure consists of co-
ordinating the sequential transitions through the
list of steps with the movement through the col-
lection of actions to be taken of the partition
between those things that have not yet been ac-
complished and those things that have already
been accomplished.
 1275/1540

Why do people count on their fingers? Because

it is a strategy that transforms a task by forming
a functional system that includes a representa-
tional media (the fingers) that permits other me-
dia to be coordinated in new ways. Nowadays
counting on one's fingers is viewed as a sort of
cognitive atavism, but "scarcely more than 400
years ago the use of one finger counting tech-
nique was so common among learned
Europeans that arithmetic books had to contain
detailed explanations of it if there were to be
considered complete" (Ifrah 1987: 58). The sys-
tem Ifrah describes is complex and required
some training to master. Still, some of us resort
to finger counting for simple problems. Suppose
today is Tuesday, December 29. What will the
date be on Saturday? I might say to myself
"Let's see: Wednesday, Thursday, Friday,
Saturday" and extend a finger for each day name
spoken. Then I would look at my hand. Next I
would coordinate again, this time bringing in the
 1276/1540

sequence of number names"30 (um, December

has) 31, and 1, 2"directing my attention to the
first raised finger when the first day number is
spoken, and moving it to the next raised finger
with each successive spoken day number until I
arrived at the last finger. The last day number
spoken would be the answer to my question.
Solving the problem in this way brings the day
names and the day numbers into coordination
with the movement of attention along the row of
fingers. The hand serves as a malleable and
handy medium upon which representational
states can be imposed and simple operations can
be performed. The structure of the hand that res-
ults from those operations is a portion of a par-
tial description of the answer to the question.
The remainder of the propagation of representa-
tional state, from structure of the hand to spoken
number, is accomplished either by simple pat-
tern matching or by another simple coordination
operation. The task can be done without using
 1277/1540

the fingers, of course, or even by coordinating

the two sequences (day names and day numbers)
directly. However, trying to simultaneously ma-
nipulate the two sequences internally requires
more memory resources than some of us can
muster.
 Page 316

Many cognitive scientists would think of this as

a silly problem, or as a problem of performance
that only highlights the weaknesses of the hu-
man mind and does not tell us anything about
cognitive architecture. But I believe the real
power of human cognition lies in our ability to
flexibly construct functional systems that ac-
complish our goals by bringing bits of structure
into coordination. That culturally constituted
settings for activity are rich in precisely the
kinds of artifactual and social interactional re-
sources that can be appropriated by such func-
tional systems is a central truth about human
cognition. The processes that create these set-
tings are as much a part of human cognition as
the processes that exploit them, and a proper un-
derstanding of human cognition must acknow-
ledge the continual dynamic interconnectivity of
functional elements inside with functional ele-
ments outside the boundary of the skin.
 1279/1540

What you think cognition is and what you be-

lieve is part of the architecture of cognition de-
pends on what you imagine to be typical or im-
portant cognitive tasks and what you think a
person is. In thinking about the use of a written
procedure, it is clear that there are many ways to
produce the coordination between the physical
structure of the procedure and the processes that
execute the actions described by the steps of the
procedure. Even the simple task of considering
steps in order can be solved in many ways at
different times, or perhaps can be solved by the
confluence of many methods at one time. Given
the ubiquity of such performances in modern
life, I take this to be the sort of cognitive per-
formance for which we should be able to ac-
count. There is also a good deal of this kind of
activity in nonliterate societies. Procedures can
be encoded in structure other than writing. The
arrangement of persons or objects in space (in a
queue, for example) can serve as a mediating
 1280/1540

device for the sequential control of action and

may elicit a set of coordinating procedures like
those observed in interactions with written
procedures.

From this perspective, what we learn and what

we know, and what our culture knows for us in
the form of the structure of artifacts and social
organizations, are these hunks of mediating
structure. Thinking consists of bringing these
structures into coordination so that they can
shape and be shaped by one another. The thinker
in this world is a very special medium that can
provide coordination among many structured
mediasome internal, some external, some em-
bodied in artifacts, some in ideas, and some in
social relationships.
 Page 317

8 Organizational Learning
 1282/1540

In this chapter I will raise some questions about

the processes by which the organization of work
arises. While in the previous chapter I examined
learning by individuals, here I will look closely
at an incident in which learning takes place in
the larger unit of cognitive analysis. Common
sense suggests that work is organized in accord-
ance with plans created by designers who reflect
on the work setting and manipulate representa-
tions of the work process in order to determine
new and efficient organizational structures.
When ''outside" designers are not involved, the
reorganization of work is attributed to conscious
reflection by members of the work group. Ex-
amining the response of the Palau's navigation
team to a change in its informational environ-
ment, I will argue that several important aspects
of a new organization are achieved not by con-
scious reflection about the work but by local ad-
aptations to the emerging conditions of the work
itself. The solution reached is one that we
 1283/1540

recognize in retrospect as being just the sort of

solution we would hope designers could pro-
duce, yet it is a product of adaptation rather than
of design.
 1284/1540

While I was aboard the Palau observing the

navigation team, the ship's propulsion system
failed unexpectedly during an entry into San
Diego Harbor. The opening paragraphs of
chapter 1 describe the event and the bridge
team's response to the difficulties created by the
failure of the propulsion plant. Without steam
pressure, the crew could neither steer the ship
effectively nor bring it to a rapid stop. All
thoughts of continuing to the pier were aban-
doned, and the crew struggled to simply prevent
the ship from going aground until it had lost
enough speed so that the anchor could safely be
dropped. In an impressive exhibition of seaman-
ship, the crew brought the Palau to anchor out
of the main shipping channel. Tugboats were
summoned and the propulsion plant was restar-
ted. The ship later continued to the pier under its
own power.
 1285/1540

Besides taking away the ability to maneuver, the

loss of steam pressure brought a cascade of elec-
trical failures that affected many aspects of the
ship's operation. Among the electrical devices
that
 Page 318

failed was the gyrocompass, which is crucial to

navigation. This incident provided me with an
opportunity to witness and record the response
of a complex organizational system to a very
real crisis.
 1287/1540

The immediate response of the navigation team

to the loss of steam and electrical power was
simply to continue with the fix they were in the
midst of taking. However, one of the pieces of
electrical equipment that lost power was the
main (Mark-19) gyrocompass. There are two
layers of redundant protection for the gyrocom-
pass function: independent emergency electrical
power and a backup gyrocompass. Unfortu-
nately, the emergency power supply for the
gyrocompass failed to come on line, and the
backup gyrocompass had been secured (taken
out of service) earlier because of a maintenance
problem. The main gyrocompass did not fail
completely when the lights went out, but it does
appear to have been mortally wounded. The
gyrocompass operates by spinning a disk at very
high speed and will operate adequately for a
while before it spins down and loses stability.
Sixteen minutes after the loss of power, the
Palau's speed had dropped to less than 4 knots
 1288/1540

and the ship was less than half a mile from its
intended temporary anchorage when word was
passed to the bridge from the forward interior
communications (IC) room that the gyrocom-
pass had ceased operation. This was an espe-
cially critical period for the navigation team.
The chosen anchorage location was out of the
navigation channel and near an area where the
water shoaled rapidly. Dropping the anchor too
soon would leave the ship obstructing traffic in
the channel; dropping it too late might allow the
ship to swing around and ground upon a shoal.
Simply restoring power to a gyrocompass is not
sufficient to bring it to a usable state; several
hours are usually required for the gyro to "spin
up and settle in" so that it will provide reliable
readings.
 1289/1540

Figure 8.1 shows the relations among the vari-

ous terms of the computation. With the gyro-
compass working, the alidade (telescopic sight)
mounted on the pelorus permits the direct meas-
urement of the direction of the bearing of the
landmark with respect to true north ("true bear-
ing" in the figure). When the gyrocompass
failed, all that could be measured by the bearing
takers with the pelorus was the direction of the
landmark with respect to the ship's head ("relat-
ive bearing"). In order to compute the true bear-
ing of the landmark, once the relative bearing
has been determined, it is necessary to determ-
ine the direction of the ship's head with respect
Page 319
 1291/1540

Cognition p319
 1292/1540

Figure 8.1 The relationships among the terms of the bearing-

correction computation. True bearing of landmark from ship
equals compass heading (C), plus deviation (D), plus magnet-
ic variation (V), plus relative bearing (RB).

to true north. The magnetic compass, which does

not require electrical power, measures the direction
of the ship's head with respect to magnetic north (C
in the figure). But the compass reading must first
be corrected for errors, called deviation, that are
specific to the compass and dependent upon the
heading (D in the figure). Cartographers measure
the difference between true north and magnetic
north for all mapped regions of the world. This is
called the variation (V in the figure). The sum of
these terms is the true bearing of the landmark,
which was directly measured by the gyrocompass
when it was working.
 Page 320

There is a mnemonic in the culture of navigation

that summarizes the relations among the terms
that make up the ship's true head. It is "Can
Dead Men Vote Twice?" and it stands for the
expression C + D =M, M + V=T (compass head-
ing plus deviation equals magnetic heading,
magnetic heading plus variation equals true
heading). This specifies a meaningful order for
the addition of the terms in which every sum is a
culturally meaningful object in the world of nav-
igation. Every competent navigation practitioner
can recite this mnemonic, and most can give an
accurate account of what it means. The know-
ledge that is embodied in this formula will be an
important component of the solution discovered
by the navigation team. Notice, however, that
this mnemonic says nothing about relative
bearings.
 1294/1540

The computational structure of the task is well

known. As was described above, computing true
bearings for landmarks from relative bearings
involves adding together the ship's compass
heading, the compass deviation for that heading,
the magnetic variation appropriate for the geo-
graphic location, and the bearing of the land-
mark relative to the ship's head. The procedure
for a single line of position therefore requires
three addition operations. If one used this pro-
cedure for each line of position, the set of three
lines of position that make up a position fix
would require nine addition operations. There is
a substantial saving of computational effort to
be had, however, by modularizing the computa-
tion in a particular way. Since all three lines of
position in any given fix are observed as nearly
simultaneously as is possible, the ship's head for
all of them must be the same. Thus, one can
compute the ship's true heading (sum of com-
pass heading, deviation, and variation) just once
 1295/1540

and then add each of the three relative bearings

to that intermediate sum. This procedure re-
quires only five addition operations for the en-
tire fix (two for the ship's true head and one for
each of the relative bearings), compared to the
nine addition operations required by the non-
modularized procedure. As we shall see when
we consider the details of the actual perform-
ance of the team, even a small saving of compu-
tational effort can be very helpful in this high-
workload environment.

A search of the Palau's operations and training

materials revealed many documents that de-
scribe in detail the nominal division of labor
among the members of the navigation team in
the normal crew configurations, and many that
describe the computational requirements for de-
riving a single line of position from compass
heading, deviation, variation, and relative bear-
ing. There
1296/1540
 Page 321

was, however, no evidence of a procedure that

describes how the computational work involved
in fixing position by visual observation of relat-
ive bearings should be distributed among the
members of the navigation team when the gyro-
compass has failed. The absence of such a pro-
cedure is not surprising. After all, if the ship had
a procedure for this situation, it should have pro-
cedures for hundreds of other situations that are
more likely to occur, and it is simply impractic-
able to train personnel in a large number of pro-
cedures in an organization with a high rate of
turnover.
 1298/1540

What might a procedure for dealing with the

event we are considering be like? Clearly it
should take advantage of the benefits of modu-
larizing the computation. Perhaps it should call
for the computation of ship's true head, followed
by the computation of each of the true bearings
in turn. That much seems straightforward, but
how should one organize the activities of the
separate team members so that they can each do
what is necessary and also get the new job done
in an efficient way? This is a nontrivial problem
because there are so many possibilities for per-
mutations and combinations of distributions of
human effort across the many components of the
computational task. The design should spread
the workload across the members of the team to
avoid overloading any individual. It should
incorporate sequence-control measures of some
kind to avoid discoordinations (in which crew
members undo one another's work), collisions
(in which two or more team members attempt to
 1299/1540

use a single resource at the same time), and con-

flicts (in which members of the team work at
cross-purposes). It should exploit the potential
of temporally parallel activity among the mem-
bers of the team, and where possible, it should
avoid bottlenecks in the computation. This is
quite a complicated design problem, and it looks
even more difficult when we examine the rela-
tionships between the members of the naviga-
tion team and their computational environment.
Given the nature of the task they were perform-
ing, the navigation team did not have the luxury
of engaging in such design activities. They had
to keep doing their jobs, and in the minutes
between the loss of the gyrocompass and the ar-
rival of the ship at anchor the requirements of
the job far exceeded the available resources

The Adaptive Response

 1300/1540

Viewing the navigation team as a cognitive sys-

tem leads us to ask where in the navigation team
the additional computational load
Page 322
 1302/1540

imposed by the loss of the gyrocompass was

taken up and how the new tasks were accom-
plished. To summarize before examining the
performance of the team in detail: The addition-
al computation originally fell to the quartermas-
ter chief, who was acting as plotter. He attemp-
ted to do the added computations to correct the
relative bearings passed to him using mental
arithmetic, but it was more than he could do
within the severe time constraints imposed by
the need for fixes at one-minute intervals. By
trading some accuracy for computational speed
he was able to determine when the ship had ar-
rived at its intended anchorage. After the Palau
came to anchor, the plotter introduced a hand-
held calculator to relieve the burden of mental
arithmetic under stress and recruited the assist-
ance of the recorder in the performance of the
computation. There was no explicit plan for the
division of the labor involved in this added task
between the plotter and the bearing recorder.
 1303/1540

Each had other duties that were related to this

problem. Dropping the anchor did not remove
the requirement to fix the position of the ship. A
ship at anchor may be blown by the wind or
pushed by the tides and swing around its anchor.
As it swings, the ship sweeps over a circle the
diameter of which is the sum of the length of the
ship and the distance from the bow of the ship to
the anchor. Even in shallow water, the Palau
can sweep a circle more than 1500 feet in dia-
meter. Since there was shoal water on one side
and a shipping channel on the other, it was im-
portant to maintain awareness of the location of
the ship.
 1304/1540

Since this correction computation has well-

defined subparts, we may ask how the subparts
of the task were distributed among the parti-
cipants. But here we find that at the outset there
was no consistent pattern. The order in which
the various correction terms were added and
who did the adding varied from one line of posi-
tion (LOP) to the next, and even the number of
correction terms changed over the course of the
66 LOPs that were shot, corrected, and plotted
between the loss of the gyrocompasses and the
arrival of the Palau at its berth. Gradually, an
organized structure emerged out of the initial
chaos. The sequence of computational and so-
cialorganizational configurations through which
the team passed is shown in figure 8.2. After
correcting and plotting about 30 LOPs, a con-
sistent pattern of action appeared in which the
order of application of the correction terms and
the division of labor between the plotter and the
bearing recorder stabilized. While the
 1305/1540

computational structure of this stable configura-

tion seems to have been at least in part intended
by the plotter, the social structure (division of
 1307/1540

1687-323.gif
 1308/1540

Figure 8.2 Computing a line of position. The structure of the

column. Lines of position are numbered across the top of eac
ates an LOP computation performed entirely by the plotter. R
performed entirely by the recorder. RP indicates an LOP com
by the recorder and completed by the plotter. PR indicates a
tured by the plotter and completed by th

labor) seems to have emerged from the interactions

without any explicit planning.

Analysis

The bearing takers out on the wings of the ship wer

the loss of the gyrocompass. For them, it meant onl
ber to shoot the bearings relative to ship's headthe o
of the two azimuth circles in the alidade viewfinder
will therefore focus on the activities of the plotter, Q
Richards, and the bearing recorder, Quartermaster S
 Page 324

We can consider the behavior of the plotter and

the recorder to be a search in a very complex
space for a computational structure and a social
structure that fit each other and that get the job
done. As figure 8.2 shows, these two men ex-
plored 13 different computational structures and
many social configurations on their way to a
stable configuration.

How can we account for this seemingly bizarre

search of computational and social space? I will
claim that there are four main principles that or-
ganize the computation:

computational structure driven by the availabil-

ity of data

the use of a normative description to organize

computation
 1310/1540

the computational advantages of modularizing

the addition task

the fit between computational and social

organization.
 1311/1540

The events that occurred between the failure of

the gyrocompass and the end of the task can be
partitioned into four temporal regions on the
basis of these principles. In the first region, lines
of position 1-15, the plotter did all the computa-
tion himself and the computational structure was
driven primarily by the availability of data. The
end of this region is marked by the introduction
of an electronic calculator. In the second region,
LOPs 16-24, the plotter began to push some of
the computational load onto the recorder. While
providing the recorder with instruction on how
to do the computation, the plotter began to use a
normative description to organize the computa-
tion. In the third region, LOPs 25-33, the modu-
larity of the computation became a shared re-
source for the two men through their joint per-
formance of the modular procedure. In the
fourth and final region, LOPs 34-66, they dis-
covered a division of labor that fit the computa-
tion and they coined a lexical term for the
 1312/1540

modular sum, thus crystallizing the conceptual

discovery in a shared artifact. Now let us look at
the details of the work at the chart table, consid-
ering the lines of position plotted from the time
the gyrocompass failed until the system settled
into its new stable configuration. (Refer to fig-
ure 8.2.)

REGION 1: COMPUTATIONAL
STRUCTURE DRIVEN BY DATA
AVAILABILITY

The plotter computed the first 12 lines of posi-

tion using what would normally be called men-
tal arithmetic. In some cases, this arithmetic was
aided by artifacts in the environment. In the very
first LOP, for example, he used the scale of the
hoey (chart plotting
 Page 325

tool) as a medium for addition, aligning up the

scale index with 29 (the compass course), slid-
ing it 52 gradations upward (the relative bear-
ing), and sliding it an additional 14 gradations to
add the variation. In LOP 2 he used the bearing
log as a memory during the computation, tracing
out the addition columns with his fingers. LOPs
8 and 9 were computed using paper and pencil
in the margins of the chart. The plotter had a
good deal of trouble keeping up with the de-
mands of the task; this is shown by the fact that,
even though three bearings were observed for
each fix, the plotter was able to plot only two
LOPs for the first fix, one for the second, and
two for the third.
 1314/1540

The anchor was dropped at 17:06, just before

the fifth line of position was plotted. Once the
anchor was down, the team went from one-
minute to six-minute fix intervals, but the plotter
was still having trouble keeping up while doing
mental arithmetic.
 1315/1540

The plotter's behavior in this region can be de-

scribed as opportunistic. He used three different
computational orderings and several different
media in computing the first twelve lines of pos-
ition. Though at first glance this behavior looks
unsystematic, there is a simple but powerful reg-
ularity in it. The order in which the plotter took
the terms for addition depends on where the
terms were in his environment and on when and
with how much effort he could get access to
them. For example, in LOP 8 the plotter re-
turned to the chart table verbally rehearsing the
ship's magnetic heading. He began his computa-
tion with that term. In LOP 9, where the plotter
had to consult the recorder in order to establish
the identity of the next relative bearing to add,
he began his computation with relative bearing.
In LOP 10 the plotter was again doing the calcu-
lation on his own, and again he began with
ship's magnetic head. These patterns are hints to
a more general organizing principle that is
 1316/1540

evident throughout this event. In the first two re-

gions of figure 8.2, 12 out of the 15 LOPs for
which the computation is initiated by the plotter
begin with the ship's magnetic head, and 13 out
of 18 computations initiated by the recorder be-
gin with the relative bearing of the landmark.

This regularity appears to be a consequence of

local strategies for individual cognitive eco-
nomy. From the perspective of a person trying
to do the addition, if one of the terms is already
in working memory when it is time to begin the
computation then it is most efficient to start with
that term.
Page 326
 1318/1540

Consider the situation of the bearing recorder.

When he does a computation while interacting
with the bearing takers, he listens to, writes
down, and verbally acknowledges relative bear-
ings. These activities, although not part of the
addition procedure, influence the course of that
procedure because they put the relative bearing
(RB) term into the working memory of the bear-
ing recorder. With RB already in working
memory, in order to do the computation in the
order that supports modularization (C+V+RB),
the recorder must somehow keep RB active in
working memory or must overwrite RB in work-
ing memory and read it again later when it is
needed. If he chooses to maintain RB in work-
ing memory, then it must remain unaltered (and
must not alter the other number representations
present) during the reading of C, the recall of V,
and the addition of C and V. This may require
the recorder to maintain up to 11 digits in work-
ing memory (eight for the addition of V+C, plus
 1319/1540

up to three for RB). If the memory load of that

task is too great, the recorder may choose to let
RB be overwritten in working memory and read
it in again later. Of course, that involves the
wasted effort of overwriting and rereading RB.

In contrast to the costs of this "preferred" order,

taking the terms in the order RB + C + V or RB
+ V + C involves lighter loads on working
memory and no wasted effort. Thus, from the
bearing taker's local perspective it was simply
easier and more efficient to begin each computa-
tion with the relative bearing.
 1320/1540

The plotter was in a different position. In most

cases, he went to the helm station to get the
ship's compass head while the relative bearings
were being reported. This puts the C term into
the plotter's working memory at the beginning
of the fix. Notice in figure 8.2 that, except for
LOPs 5-7, every LOP initiated by the plotter be-
gins with C as the first term. But interaction
with the recorder or with other representational
systems can change the plotter's position in the
computation. In each case where the plotter
began by asking the recorder for a term to add,
that term was the relative bearing and the relat-
ive bearing was taken as the first item in the ad-
dition. On closer inspection, the apparent excep-
tions to the rule in LOPs 5-7 are not exceptions
at all. These computations were not done while
the data were coming in. The observations of the
three relative bearings were made while the
plotter worked to determine the location of the
anchor. Then he set out to compute the LOPs
 1321/1540

with all of the data in the bearing log in front of

him, the relative bearings in the left columns of
the page and the ship's magnetic head in the
 Page 327

rightmost column. This interaction with the

bearing log changed the temporal pattern of
availability of data, which in turn changed the
organization of the most efficient ordering of
terms for the performance of mental arithmetic.

It is unlikely that either man was ever aware of

having made a decision concerning the order in
which to add the terms. Rather, each was simply
trying to do the additions as correctly and as ef-
ficiently as possible. Since the two men experi-
enced different patterns of availability of data,
this principle produced characteristically differ-
ent results for each of them.
 1323/1540

The principle at work so far can be summarized

as follows: Individual actors can locally minim-
ize their workloads by allowing the sequence of
terms in the sum to be driven by the availability
of data in the environment. But since data be-
come available primarily via social interactions,
the computational structure is largely an un-
planned side effect of this interactional struc-
ture. The interactional structure itself is chaotic
because it is shaped by interference from other
tasks and by social interactions with other mem-
bers of the navigation team and with members
of other work teams on the bridge.
 1324/1540

After LOP 12, the recorder initiated a round of

bearings on a twominute interval. The plotter in-
structed him to take the fix on sixminute inter-
vals and complained about not being able to
keep up with the computations using mental
arithmetic. When I asked if he had been able to
keep up with the work, he said: ''No, I was run-
ning it through my head and it wouldn't add. It
wouldn't make numbers, so I was making mak-
ing right right angles in my head to see where
the hell it was at." The recorder said "You take
the variation out of it." "Yes" said the plotter,
"you add the, you add the magnetic head, then
you add the variation." This conversation is the
first evidence of reflection on the structure of
the computation. The plotter explicitly named
the variables: "... you add the magnetic head,
then you add the variation." After this, the plot-
ter remarked that the only way to keep up with
the work would be to use a calculator. Shortly
after this conversation, the plotter went to the
 1325/1540

charthouse and returned with a navigation calcu-

lator. The calculator was capable of computing a
number of specialized navigation functions, but
only addition and subtraction were used in what
followed.

The use of the calculator eliminated the need for

the intermediate sums that the plotter computed
when doing mental
 Page 328

arithmetic. In LOPs 1315 the plotter keyed in

the data. He started each LOP computation by
keying in C +; then he looked for RB in the
bearing book, keyed RB +, then keyed V =.
Here the calculator was not only a computation-
al device; the plotter also used it as a temporary
external memory for the C term while he looked
for the RB term. The immediate consequences
of the introduction of the calculator were that it
eliminated the production of intermediate sums
(this will be important in the development of the
modular solution below) and that it changed the
memory requirements for the plotter by serving
as an external memory. It did not change the
fact that the order in which the terms were ad-
ded was dependent on the pattern of availability
of data in the task environment.
 1327/1540

The dependence of the computational sequence

on the availability of data is the main character-
istic of events in the first region. It will survive
into later regions in the behavior of the recorder,
but the introduction of the calculator marks the
beginning of the end of this sort of data-driven
task organization for the plotter. Up until and in-
cluding the first calculator round, the recorder
has sometimes fed values of RB to the plotter
but has done no arithmetic, mental or otherwise.
That is about to change.

The following conventions are used to record

the verbal and nonverbal actions of the plotter
and the recorder in the transcripts below.

( ) Words enclosed in parentheses are comments

or annotations of the actions observed in the
video record, never verbatim transcriptions.

# Hash marks are used in adjacent lines of tran-

scription to indicate simultaneity of occurrence.
 1328/1540

/?/ This represents an unintelligible portion of an

utterance.

{ } Numbers and actions enclosed in braces de-

note key presses on the calculator.

{3 + } Numbers and actions in boldface en-

closed in braces are key presses on the calculat-
or that are verbally shadowed. This example
would mean that a person pressed the 3 and the
+ key while saying "Three plus." In addition to
numbers, the most frequent key presses are +, -,
=, and clear.

1 20 Spoken numbers have been transcribed

mostly as numerals for convenience. If they are
separated by spaces, each numeral was pro-
nounced separately. If they are not separated by
space, then they are to be read as conventional
numbers.
 Page 329

This example could also have been transcribed

"One twenty."

The following is a key to the notation for the

computation:

C The compass heading of the ship with no

corrections.

D Compass deviation. A function of heading.

V Magnetic variation. Approximately 14° east in

San Diego Harbor.

RB The relative bearing of a landmark. This is

the bearing of the landmark with respect to the
ship's head.

M Ship's magnetic heading (C + D).

T Ship's true head (M + V)

TB True bearing (T + RB).

 1330/1540

( ) Terms enclosed in parenthesis were entered

into the calculator with only + or - operators
among them. The = operator closes the paren-
thesis. Thus, (C + V + RB) means that the three
terms were added together as a group; ((C +V)
+RB) means that the = operator was applied to
(C+V), which was then added to RB.

[] Sums in brackets were spoken as intermediate

sums. Thus, ([(V + D) + C] + RB) denotes the
following actions: key V, key+, key D, key=,
key +, key C, key=, read the displayed value
aloud, key +, key RB, key =.
 1331/1540

How can we know the order in which the terms

of the computation were applied? The computa-
tions involved in each line of position were re-
constructed from the data available in the fol-
lowing way: Usually, the values of all of the
variables were either present in the transcript or
could be determined. In all cases, the variation
was 14°. In LOP 8, for example, I had a record
of the helmsman telling the chief that the ship's
head was 3 3 5 degrees. The relative bearing to
Silvergate was reported by the port pelorus op-
erator as 275°. The problem was to arrange these
in a way that fit with the numbers that were
verbalized. Here is what the plotter said:
 1332/1540

Is 3 3 5, 3 3 5. Oh wow. (Mumbles 3 seconds.

The plotter watched the recorder write down the
bearing to Silvergate. The plotter then jotted the
bearing on the chart and did place value arith-
metic in the margin of the chart.) 1 1 6, 60, 0, 6
from 1 is 5, 2 5 0, 2 5 0 ah, 2 5 0 would give me
2 6 4. 2 6 4, what the hell is it to? Ah, I know
what it is to, it's got to be to Silvergate. Yeah. 2
6 4.
Page 330
 1334/1540

Clearly the plotter added the ship's magnetic

heading, 335°, to the relative bearing of the
landmark, 275°as shown below. The "1 1 6" ap-
peared to refer to the carry digits and the sums
of the leftmost two columns of the addition. It
was impossible to determine which spoken 1 re-
ferred to which carry digit or to the sum of the
central column. Nevertheless, it was certain that
this was the sum being performed. Since this
summed to more than 360, it was necessary to
subtract 360 from the sum. The spoken 60 may
have been the 60 of 360. Then there was the 0
which was the subtraction of the right column,
followed by an entire description of the subtrac-
tion carried out in the center column: "6 from 1
is 5." Up to this point, the addition was done
with paper and pencil on the margin of the chart.
From here on it was conducted mentally. The
outcome, 250, was rehearsed twice; then the
variation was added to it to produce the final
sum.
 1335/1540

The complete reconstruction is shown below.

The numbers in boldface appear explicitly in the
transcript; the numbers in lightface do not ap-
pear in the transcript but can be inferred to have
been present.

31315

275

610

360

50250250

14

264
 1336/1540

Thus, we can infer that the addition of the cor-

rection terms for LOP 8 were taken in the order
(C + RB)+V. Similar reconstructions were done
for all LOPs in the event. In some cases it was
necessary to reconstruct the actual fix itself as
well in order to disambiguate unclear utterances
in the tape recordings. Using this technique, it
was possible to determine the exact order in
which the terms were taken in all cases but
three. The cases where it was not possible to
make a clear determination all involved errors
committed by the members of the team. In those
cases I have attempted to make the most likely
reconstruction.

REGION 2: EMERGENCE OF MEDIATING

STRUCTURE

The most important consequence of the intro-

duction of the calculator was that it created a
new context of interaction between the
1337/1540
 Page 331

plotter and the recorder in which the plotter

gave the recorder instruction in the procedure.
For example in LOP 16 the plotter returned
from the helm station, where he had read the
compass heading and keyed in the value of C:

LOP 16: (C + V + RB)

(Plotter returns from helm.)

Plotter: 2 3 1. What have we got? {231 + }

(Then he slides the calculator in front of the

recorder. )

Here, add these things.

You want ... You want the head. You want the
head# which is 2 3 1

Recorder: #and

add variation.
 1339/1540

Plotter: Plus variation.

Recorder: Oh, 2 3 1 is the head?

Plotter: 2 3 1. Here {clear 2 3 1}

Recorder: I got it. (Recorder puts his hands on

the keys.) {clear, 231}

Plotter: Plus 14.

Recorder: { + 14} OK.

Plotter: OK. (The intermediate sum was not

computed.)

Recorder: { + 0 0 7 =} is 2 5 2 on Silvergate.

Plotter: 2 5 2 Silvergate.

The plotter controlled the order of the arguments

in this LOP. The recorder seemed surprised that
he started with the ship's head.
 1340/1540

In LOPs 17 and 18a, the plotter was busy plot-

ting a previous bearing. The recorder initiated
the computation himself by reading the RB from
the book and beginning with it. In LOP 18a, the
result was in error because the bearing that was
reported was misread by the bearing taker. But
the context of the error provided an opportunity
to restructure the work. The recorder slid the
calculator over in front of the plotter and began
to dictate values starting with what was for him
the most salient term, RB. The plotter, however,
ignored the recorder and began keying in the
data in the sequence C + V. The plotter made an
error and cleared the calculator. The recorder,
having seen the sequence in which the plotter
wanted to add the terms, dictated the terms in
the order (C+V + RB):
 Page 332

LOP 18c (C + V+ RB)

Recorder: 2 3 1, Chief, plus 14, plus #

Plotter: { 2 3 1 + 1 4 + } #OK, what was ah,

Recorder: The bearing was 1 5 7. (3 seconds) #OK

Plotter: {1 5 7 = } #402

Recorder: Minus 3 6 0 # is

Plotter: { - #3 60 = } is 0 4 2. No it ain't. It isn't

no 0 4 2. Its just not working. Look where 0 4 2
goes. (The plotter points to the chart.) If it's 0 4
2, we're sitting over on Shelter Island!

There were three more attempts to compute this

LOP. In LOP 18d, the recorder made a data-
entry error and passed the calculator to the plot-
ter in frustration. In LOP 18e, the plotter made a
dataentry error, cleared the calculator, and
began again.
 1342/1540

We might have thought that the importance of

the introduction of the calculator would be its
power as a computational device. In fact we see
that using the calculator the team was neither
faster nor more accurate than without it! The
important contribution of the calculator was that
it changed the relation of the workers to the
task. When the plotter pushed the calculator
over to the recorder and told him to add the
terms, he engaged in a new task, that of instruct-
ing the recorder in the computation, and he or-
ganized his instructional efforts in terms of the
normative computational structure,
C+D=M+V=T. This was evident in LOP 16,
where the plotter named the variables: "You
want the head, which is 2 3 1....
 1343/1540

Plus variation." Note that the recorder did not

seem to learn from the explicit statements of the
plotter. He returned to taking the RB first in
LOPs 17, 18a, and 18b. However, once the plot-
ter had articulated this structure it became a re-
source he could use to organize his own per-
formance of the task. In LOP 18b, although the
recorder had dictated the RB to him first, he
keyed in C+V. There, the recorder verbally
shadowed the plotter's keystrokes. This joint
performance was the first time the recorder had
taken ship's head as the first term. Once the plot-
ter began behaving this way, the recorder was
able to internalize the strategy that appeared in
interpersonal work, and under certain social
conditions he could use it to organize his own
behavior. Thus, in LOP 18c, where the recorder
took the role of dictating the values to the plotter
(who was keying them in), the recorder said "2 3
1, Chief, plus 14, plus...." But the structure was
 1344/1540

not yet well established for the recorder. In the

next
 Page 333

attempt, LOP 18d, a new RB was observed and,

driven by the data, the recorder began the com-
putation with it.
 1346/1540

The introduction of the calculator and the errors

that were committed with it provided a context
for instruction in which the sequence of terms
could be explicitly discussed. The errors they
were responding to were not sequence errors but
simple key-pressing errors, yet they still served
as contexts for sequence specification. The plot-
ter appeared to learn from his own instructional
statements (intended for the recorder) and
changed his own behavior. Until he tried to in-
struct the recorder on what to do, he took the
terms in the order in which they were presented
by the environment. The recorder appeared to
change his own behavior to fit with what the
plotter did, not what he said. This newly emer-
gent normative structure dominated the plotter's
instructional efforts and came to dominate the
organization of his task performance as well.
 1347/1540

In LOP 21a, the recorder made a key-pressing

error while adding the terms in the order (RB +
C + V). The error drew the plotter's attention,
and he turned to watch the recorder.

LOP 21b: (C +RB +V ) & ((C +V)+RB) =((C

+V)+RB+ V)

Recorder: {clear 2 2 1 # + 14 }

Plotter: #plus 14 is 2 3 5. (C+V The plotter does

it in his head.)

Recorder: 2 3 5?

Plotter: Yeah, its 2 3 5 plus 1 1 8. ((C + V) +

RB)

Recorder: Oh. {clear}

(The recorder doesn't realize that hitting "="

would have produced 235.)
 1348/1540

Plotter: 2 3 5 is #3 3 5, 3 4 5, how about 3 5 3.

Right?

Recorder: {235 # + 1 1 8 + 14 =} How about 00

7? ((C + V) +RB+ V)

Plotter: 0 0 7.

Recorder: Chief, the computer just beat you.

(The plotter glares at the recorder.) Just kidding.
(They all laugh 4 seconds.) The modern
technology.

Plotter: I'll modern technology you.

Here two important things happened. First, the

recorder demonstrated that he could produce the
normative sequence when trying
 Page 334

to show the plotter he could do the addition cor-

rectly. Second, this was the first time the plotter
had organized a properly modular computation.
Unfortunately, it is also clear that the recorder
did not yet understand the meaning of the inter-
mediate sum (C+V), which is the key to the
modularization. He mistook it for C alone and
added in RB and V, thus generating an error.
The plotter seemed intimidated by the calculator
and did not challenge the result. It led to a poor
fix, but he had been getting really poor fixes all
along. The anchor was holding and the ship was
in no danger, but at this point if they had had to
rely on the quality of the fixes they would have
been in trouble.
 1350/1540

In LOP 22, the plotter failed to use the modular

form of the computation. Unless they work to-
gether and make the modularized total available
to each other, there is no advantage in modular-
ization. The modularization is an instance of a
much more general computational phenomenon.
The construction of the compass-deviation table
is part of the computation, but it is a part that
was done by the navigation team days or weeks
before the execution of this task. Similarly, the
measurement of the variation is part of the com-
putation, but it was done years ago by carto-
graphers. In each case, parts of the computation
that are not variable in the instance have been
taken out and crystallized as artifacts (the vari-
ation printed on the chart, the deviation table).
In the same way, the modular sum is a pre-com-
puted invariant of the main computation.
 1351/1540

The plotter performed LOP 23 with the non-

standard sequence (C + RB + V). This,
however, is not a violation of the principles de-
scribed above. The plotter did not get C from
the helm at the beginning of the fix, as he usu-
ally did. Instead, he was busy asking whether
the anchor was being hoisted at this time. The
recorder announced C when the plotter returned
to the table. The plotter looked in the bearing
log for C. He read it aloud, and while still lean-
ing over the book he added in the RB nearest
him in the book, pointing to the place digits in it
with the butt of his pencil as he added the num-
bers. Once again, the availability of data in the
environment drove the organization of the
computation.

LOP 24a: (RB + C + V)

Recorder: 1 1 2 plus 2 2 6 plus 14, 3 5 2 on

ship's head.
 1352/1540

(The recorder means to say "Hamm's light.")

Plotter: Which tower is he shooting for North Is-

land Tower?
 Page 335

(The plotter leaves the table and goes to the port

wing)

Hey, which tower are you shooting for North Is-

land Tower?

(PW points to tower) You are? OK.

PW: Is that the right one?

Plotter: Yep.

(P returns to table)

LOP 24b: (C + V + RB)

Recorder: Which tower #wa-

Plotter: #And ah, what was Hamm's?

Recorder: And Hamm's was {2 2 6+14 + 112 =}

3 5 2. (5 sec)
 1354/1540

Time 5 6 Chief.

In LOP 24a the recorder, working on his own,

took the terms in the order (RB + C + V). A few
moments later, when the plotter asked the re-
corder what the bearing was to Hamm's, instead
of remembering it the recorder recomputed it.
This time, LOP 24b, he did it in the prescribed
order, (C + V + RB). This is evidence that he
knew the sequence preferred by the plotter, but
he seemed to produce it only in interactions with
the plotter.
 1355/1540

This brings us to the end of the second region.

In this region we have seen that a mediating
structure is being remembered by the plotter, but
the recorder's organization of the computation is
still driven largely by the pattern of availability
of data. The clear boundary between this region
and the first one is not marked by the introduc-
tion of the calculator, but by the plotter's order
"Here, add these things." The change in compu-
tational structure follows from a social innova-
tion that was made possible by a technological
change rather than from the technological in-
novation itself.

REGION 3: PARTIAL MODULARIZATION

 1356/1540

In the description of the computational structure

of the task given above, we noted that the true
bearing is the sum of four terms: ship's magnetic
head, C; deviation, D; variation, V; and relative
bearing, RB. By now the team had computed
and plotted 24 lines of position, and the devi-
ation term was not included in any of them. This
seems surprising, since we have ample evidence
that both the plotter and the recorder know well
what deviation is and how to use it. One can
only surmise that they were so busy trying to do
the job that they forgot to include this term.
Luckily, the absence of the
 Page 336

deviation term had no effect on the quality of

the fixes plotted until LOP 22, because until
then the ship was on a heading for which the de-
viation was near zero. Just before LOP 22,
however, the ship's head swung southwest, to a
heading for which there was a 3° deviation. The
fix triangles started opening up and it became
clear to the plotter that something was wrong.
He lay the hoey on the chart from various land-
marks and moved it slightly, seeing what sort of
different bearings would make the triangle smal-
ler. LOPs 25-27 are a reworking of LOPs 22-24,
this time taking deviation into account.

1. Plotter: I keep getting these monstrous god-

damn, these monstrous frigging goddamn tri-
angles. I'm trying to figure out which one is
fucking off.

2. Recorder: You need another round?

 1358/1540

3. Plotter: No, no no, uh uh. 1 2 0 I know what

he's doing. Let me try, let me try, (The plotter
turns and moves to helm station) let me try, with
my new ones, say three. (He reads the deviation
card posted on compass stand.) Say three, add
three to everything.

4. Recorder: Add three?

5. Plotter: Yeah.

6. Recorder: 'Cause he's using magnetic? (The

recorder does not get it yet.)

LOP 25 ([(V + D) + C] + RB)

7. Plotter: On a southwest heading add three. So

its (14+3=)17 plus 2 2, 17 plus 2 2 6 is ah, 2 3
ah

8. Recorder: Plus 2 26 is 3 4 is 2 43 ((V + D)+

C) (The recorder is working on paper with
pencil)
 1359/1540

9. Plotter: Okay, 2 4 3 and 0 1 3 is 2 5 6. 2# 5 6

([(V + D) +C]+ RB)

10. Recorder: #2 5 9 (this is an error)

11. Plotter: 2 5 nuh uh?

12. Recorder: 2 5 9, plus 0 1 3? It's 2 5 9.

13. Plotter: 2 5 9 that's right. OK. And plus 1 1 2

was what?

LOP 26a

14. Recorder: 1 1 2 plus 2 2 6.... (RB + C))

(Here is clear evidence that the recorder doesn't

understand the attempt to modularize.)
 Page 337

LOP 26b ([(V + D) + C] + RB) & (RB + [(V +

D) + C])

15. Plotter: Plus 2 4 3, 2 4 3 plus 1 1 2. ([(V +

D) + C] + RB)

16. Recorder: 1 1 2 plus 2 4 3 is 5 5, 3 5 5. (RB

+ [(V + D) + C])

In the plotter's moment of discovery, line 3,

where he said ''I know what he's doing," he no-
ticed that the geometry of the triangle was such
that a small clockwise rotation of each of the
lines of the previous fix would make the triangle
smaller. A small error that belongs to all the
LOPs suggests deviation. He went to the helm
station and consulted the deviation card to de-
termine the deviation for this heading. Although
he describes the results as "much better," with
deviation included, the two errors introduced by
the recorder still result in a poor fix.
 1361/1540

The plotter had compiled a new deviation table

for the compass while at sea only a few days
prior to this event, and the bearing recorder had
demonstrated his mastery of the use of deviation
in an at-sea exercise 2 months earlier. The prin-
ciples of this computation are well known in the
culture of navigation. I have no doubt that in an
interview the plotter could describe the compu-
tation effortlessly. Their task here is not to dis-
cover these things "in the world" but to discover
them in their own knowledge. Yet it takes the
plotter 55 minutes and 24 lines of position to
discover that he knows the proper order in
which to add the terms to make the corrections.
 1362/1540

The computation of 243° as the ship's true head

and its use in LOP 26b (line 16) is the very first
full modularization of the computation. The
plotter has control of the computations in all
three LOPs, although in LOP 26b he has to fight
the recorder's strong propensity to put the RB
first. The recorder clearly does not yet under-
stand either the benefits of modularization or the
necessity to add the RB last in the modular
form. The structure of LOP 27 was modular too,
but the value of ship's true head, while properly
computed, was not correctly remembered.
 1363/1540

Why the plotter recomputes all the lines of posi-

tion for this fix instead of simply adding 3 to the
earlier results he got is not so clear. It may be an
attempt to eliminate any arithmetical errors that
occurred in the previous round. It was, after all,
a terribly big triangle. Also, all the calculations
are done in this set by hand on paper with pencil
rather than with the calculator. This could be a
way of making sure that it was not the use of the
calculator that was causing the problems.
 Page 338

The plotter seems to have taken the discovery of

deviation and the recomputation of the bearings
as an opportunity to think about the structure of
the computation. The reflection that came in the
wake of the introduction of the calculator led
him to organize the computation in accordance
with the normative form. The reflection that
came with the addition of the deviation term led
him to the modular structure. He never expli-
citly mentioned the advantages of modulariza-
tion; however, if he was not aware of the ad-
vantages when he organized the computation, he
must certainly have been aware of them once
the computation had been performed.
 1365/1540

The recorder computed LOP 28 while the plotter

explained to the keeper of the deck log why the
gyrocompass could not be restarted in time to
help and why they must therefore make the re-
mainder of the trip using magnetic bearings. The
plotter's conversation was interrupted by the re-
corder, who checked on the procedure for using
the deviation table.

LOP 28: ([C + D] + V + RB)

Recorder: Charles? (2 seconds) Head?

Helm: 2 2 6.

Recorder: 2 2 6

Recorder: So it's 2 2 6. You wanna add 3, right?

On a southerly course? (3 seconds) Chief?

Plotter: Say again.

 1366/1540

Recorder: You wanna add 3 to that /?/ southerly

course? (pointing at the entry on the deviation
table.) (2 seconds) It's 2 2 6. The magnetic head
is 2 2 6.

Plotter: Yeah.

Recorder: 2 2 6 plus # 3, OK, so that makes 2 2

9. {2 2 9 + 14}

Plotter: #right.

Recorder: { + 1 1 5 =} (3 seconds) 3 5 8 on
Hamm's light. ([C + D] + V+RB)
 1367/1540

Thus, the recorder took the arguments in the

right order in LOP 28 but did only a partial
modularization. He computed (C+D)= 229 as a
modular sum. Then he added V and added RB
without producing the ship's true head as an in-
termediate sum. In LOPs 29 and 30, the recorder
started with the partially modular sum and ad-
ded the terms in the order [C+D]+RB+V. Even
this partial modularization is an important step
forward for the recorder. It appears to be due to
two factors. First, including deviation in the
 Page 339

computation may have made the C term more

salient. Second, the recorder's location in the
computation has changed. He recorded the relat-
ive bearings as usual, but he had to go to the
helm station himself to get the compass heading
because the plotter was otherwise occupied. At
that point he had the C term in working memory
and it was time to begin the computation. This
change in location meant that what was best for
the computation was also easiest for the record-
er. This is not the best division of labor, but it is
one for which there is a momentary local fit
between social and computational structure. The
pattern of availability of data was not running
counter to the computational structure. Paradox-
ically, then, the extra work that took the plotter
away from the chart table (a burden on the sys-
tem) may have permitted the system to improve.
 1369/1540

After one hour and twenty minutes at anchor,

the Palau weighed anchor and began to move
under its own power toward the pier. LOPs
32-33 are a turning point in the procedure. In
LOP 32 there is a clear conflict of understanding
between the plotter and the recorder. In LOP 33
they perform what will be the stable configura-
tion for the first time.

LOP 32: ([(C + D) + V] + RB)

1. Recorder: You want the aero beacon?

2. Plotter: Yeah, I want the aero beacon now,

yeah. It's just.. 1 8 7, 8 8 , 8 7, 8 8.

3 Recorder: 0 2 0, what's the ship's head?

4. Plotter: Huh? 0 8 7. 8 7, it's # 1 west

5. Recorder: #0 8 7 it's 1 west , 7

6 Plotter: It's 86 (C + D)
 1370/1540

7. Recorder: {8 6}

8. Plotter: And 14 # is 100((C+D)+V)

9. Recorder: #{ + 14 }

10. Recorder: { + 1 0 0 } , hold it

11. Plotter: No, it's 1 0 0 plus whatever. ([(C +

D) + V] + RB)

12. Recorder: 1 0, where are you getting? ...

13. Plotter: 1 0 0 is the heading, the whole thing,

#plus relative.

14. Recorder: #Oh, the whole thing. plus relat-

ive, { + 20 = }, 1 2 0.

15. Plotter: OK

16. Recorder: 1 20 #is for North Island Tower.

 Page 340

LOP 33: ([(C + D) + V] + RB)

17. Plotter: #and Hamm's? (2 seconds) 10 #0

plus whatever for Hamm's.

18. Recorder: #Hamm's

19. Recorder: OK, {100 + 2 2 4 =), 3 2 4 on

#Hamm's

20. Plotter: #3 2 4. That's all three of 'em. I got

'em all.

21. Recorder: OK.

22. Plotter: Looks good. Right on. Perfect. Pin-

point fix.

23. Recorder: All right!

 1372/1540

In LOP 32, the plotter works with the recorder

to recompute the ship's true heading. This joint
work in lines 4-16 provides the opportunity for
the recorder to understand that the "whole thing"
is the modular sum to which the RB can be ad-
ded. The order in which the recorder added the
terms still followed the pattern of data availabil-
ity, but the plotter actively constructed the pat-
tern of data availability so that the sequence pro-
duced by the recorder was the desired one. That
is, the plotter acted as a mediator between the
pattern of data availability in the task environ-
ment and the addition activities of the recorder.
 1373/1540

The most salient features of region 3 were the

emergence of the partial modularization of the
computation and the conflicts between the
plotter's newly solidified conceptual schema and
the recorder's practices. In this region the plotter
began to provide mediating structure that
changed the pattern of data availability experi-
enced by the recorder. In LOP 33 the recorder
showed signs of using this mediating structure
himself. For the recorder, the addition activity
was no longer on the surface being applied op-
portunistically. It now lay behind a conceptual
and social organization that fed it the terms of
the expression in a particular order.

REGION 4: THE NEW STABLE SOLUTION

 1374/1540

In the previous subsection, we saw how the be-

havior of one individual can act as a mediating
device that controls the pattern of availability of
data for the other. In the fourth and final region,
the team discovered a division of labor in which
each member could use a computational se-
quence that followed the availability of data in
the task environment (thus minimizing memory
load and wasted effort) while each simultan-
eously produced for the other patterns of data
availability that supported the modular form of
the computa-
 Page 341

tion. In this region the computational structure

was still driven primarily by the pattern of avail-
ability of data, but the availability of data was
determined by the social organization of the ac-
tions of the members of the team. Thus, the is-
sue here is the fit between the constraints of
cognitive processing (memory limitations, e.g)
and the social organization of work (distribution
of cognitive labor), as mediated by the structure
of the computational task (modularity of
addition).

In LOPs 34-36 the recorder and the plotter tuned

their division of labor. They computed the mod-
ular sum jointly in LOP 34, and the recorder re-
membered the modular sum in LOPs 35 and 36.

LOP 34: ([(C + D) + V] + RB)

Plotter: OK, what's he on? (to helm) What are

ya on right now?
 1376/1540

8, 8 5, 8 5, 0 8 5,0 8 #4 plus 14 0 9 8. ((C + D) +

V)

Recorder: #0 8 5 is 0 8 4 plus 14,{ 8 4 + 14 =}

that's

Plotter: OK

Recorder: 98

Plotter: 9 8 and 2 6

Recorder: 98 { +26=}124. ([(C+D)+V]+RB)

Plotter: #1 2 4

Recorder: #1 2 4 North Island tower.

Plotter: OK

LOP 35:

Recorder: {9 6 + 2 1 2 =} 3 0 8 on Hamm's
light. ([(C + D) + V] + RB)
 1377/1540

(The recorder has misremembered the true head.

It should be 98, not 96)

Plotter: OK

LOP 36:

Recorder: {98 + 3 5 7}

Plotter: Damn near reciprocals.

Recorder: {-3 6 0 =}

Plotter: 3 60 is #0 9 5

Recorder: ah #0 9 5 ([(C + D) + V] + RB)

 1378/1540

This is essentially the pattern of work they

maintained all the way to the pier. By LOP 38
the final pattern was achieved. In this pattern,
the plotter computed the modular sum alone,
finding C and D at the helm station and recalling
V from his long-term memory. Meanwhile, the
recorder recorded the relative bearings. The
plotter
 Page 342

then added the first relative bearing to the mod-

ular sum, usually while the recorder was record-
ing the last of the relative bearings. The plotter
announced the modular sum to the recorder, and
the recorder then added each of the other relat-
ive bearings to the modular sum. The only im-
portant event not included in these first 38 lines
of position was the advent of a linguistic label
for the ship's true head. They called it "total" in
LOP 42 at 18:42. Once they had a name for it,
they could pass it to each other more easily. The
"publication" of the modular sum is essential to
the final solution, since it acts as the bridge
between the portion of the computation done by
the plotter and that done by the recorder.

Discussion
 1380/1540

It appears that four principles control the navig-

ation team's search of the space of computation-
al and social structures. They are (1) the advant-
ages of operating first on the contents of work-
ing memory, which led the computational se-
quence to be entrained by the pattern of availab-
ility of data; (2) the use of normative computa-
tional structure, which permitted the discovery
of (3) the advantages of modularization of com-
putation; and (4) the fit of social to computa-
tional structure. Each region of the adaptation
process was dominated by one of these prin-
ciples. In fact, all of them, except the advantages
of modularization, were present to some extent
in all four regions of the adaptation history.

MEMORY LIMITATIONS AND

AVAILABILITY OF DATA
 1381/1540

In the beginning, the structure of the computa-

tion seemed to be driven exclusively by interac-
tion between limitations of the human cognitive
system (specifically memory limitations) and
the availability of data in the environment
(Newell and Simon 1972; Anderson 1983).
Memory limitations made it advantageous to
add the terms of the correction in the order in
which they became available. The availability of
data depended on the pattern of social interac-
tions. This seemed to characterize the plotter's
behavior until he assumeed a different relation
to the computation at LOP 16. It described the
recorder's behavior at least until LOP 32, and
possibly to the end of the task.
 1382/1540

At LOP 16, the introduction of the calculator

gave rise to a new social arrangement (the re-
corder punched keys while the plotter told him
which keys to press.) This gave the plotter a
new relation to the computational task, which
led, in turn, to the introduction
 Page 343

of the normative computational structure. What

the plotter remembered was acted out in interac-
tion with the recorder. When the recorder took
dictation from the plotter while keying in val-
ues, the plotter was mediating the task for him.
The plotter was changing the recorder's relation
to the task so that what was convenient for the
recorder was also effective for the computation.

THE NORMATIVE COMPUTATIONAL

SEQUENCE, C+D=M, M+V=T, T+RB=TB

There is no doubt that the plotter's computations

were shaped by variants of the normative struc-
ture from LOP 16 on. There was only one ex-
ception to this (LOP 19), and in that case RB
had a value that was particularly easy to handle
(0 0 7). The plotter maintained this structure
even when it ran counter to the pattern of avail-
ability of data, as in LOP 18b.
 1384/1540

The recorder appeared to be capable of produ-

cing the normative sequence when in interaction
with the plotter (LOPs 24b and 27), but when on
his own he seemed to be driven by the availabil-
ity of data. Thus, when computing the true bear-
ings as he recorded the values of relative bear-
ings, he always took RB as the first term. Before
the discovery of the deviation term he used the
sequence (RB + C + V); after the inclusion of
the deviation he used (RB + C + D + V). In one
instance, however, the plotter left the table to do
another task, and the recorder computed the true
bearings alone. After having recorded the relat-
ive bearings and having obtained the ship's mag-
netic head from the helmsman (C term in work-
ing memory), the recorder began with the C
term.
 1385/1540

The computational importance of the normative

sequence is that it makes the modularization
possible. Since addition is a commutative opera-
tion, there is no difference in the sum achieved
by adding the terms in any of the 24 possible se-
quences. But if the addition is to take advantage
of the modularity of the ship's true head, the
terms C, D, and V will have to be added togeth-
er before any of them is added to a relative bear-
ing. The normative structure provides a rationale
for doing this, and it provides culturally mean-
ingful interpretations of the intermediate sums
that are lacking from such non-normative addi-
tions as (RB + V) and (V + D). (See figure 8.1.)

THE MODULAR COMPUTATION

The modular organization of the computation

emerges haltingly from the plotter's attempts to
apply the normative form, but it
 Page 344

seems unlikely that the plotter took up the norm-

ative form for its links to a modularized form of
the computation. It is more likely that the norm-
ative form gave him a better understanding of
what was going on by providing intermediate
sums that have meaningful interpretations in the
world of the ship. For an experienced navigator,
a bearing is not simply a number; it is a body-
centered feeling about a direction in space. Tak-
ing the terms in non-normative sequence results
in intermediate sums that are just numbers. Tak-
ing them in normative sequence results in inter-
mediate sums that are meaningful directions in
the world of the navigator. In this form they be-
come directions that make sense (or don't), and
this gives the navigator another opportunity to
detect error or to sense that the computation is
going well or badly even before it is completed.
 1387/1540

There was a hint of modularity in LOPs 18e and

18f, where the plotter computed C+V and then
asked for RB. Similarly, in LOP 21b he said "...
it's 2 3 5 (C+V) plus 1 1 8 (RB)." In each of
these cases, there was only one LOP involved,
so it was not possible to exploit the advantages
of modularization. The first unambiguous case
of modular computation was in the LOPs
(25-27) that introduced the deviation term.
These were performed in the nonstandard se-
quence ([(V + D) + C] + RB). It is probably sig-
nificant that the plotter chose to perform these
calculations with paper and pencil rather than
with the calculator. The paper-and-pencil com-
putation produced, as a natural side effect, a
written record of the sum [(V + D) + C], which
was then at hand for addition to each of the rel-
ative bearings. The written record of the modu-
lar sum in this instance was functionally similar
to the verbal "publishing" of the labeled modu-
lar sum as "total'' in the later fixes.
 1388/1540

The modularization of the computation echoes

the process of precomputation described in
chapter 3. The modularized form of the compu-
tation captures a short-lived invariant of the en-
vironment in a temporary representation.

FIT OF SOCIAL AND COMPUTATIONAL

STRUCTURE

The modular form of the computation became

stable only when a new division of cognitive
labor was established in LOPs 32 and 33. The
pattern of availability of data produced by the
division of labor in this stable configuration fit
the computational structure of the problem. The
plotter obtained C from the helmsman and D
from the deviation table, added them, and then
added the variation (easily available in
memory). At the same time, the recorder recor-
ded the
 Page 345

relative bearings of the landmarks. The plotter

told the recorder the modular sum, which the re-
corder recorded, and the recorder provided the
plotter with the first relative bearing. The plotter
added this relative bearing to the remembered
modular sum. While the plotter plotted the first
LOP, the recorder then added each of the other
recorded relative bearings to the modular sum.
Thus, the team arrived at a division of cognitive
labor in which the behavior of each of the parti-
cipants provided the necessary elements in the
information environment of the other just when
they were needed. While each man could be-
have as though driven by the availability of data
in the world, as a team they performed the addi-
tions in the sequence that exploited the benefits
of modularization.

Adaptation by Design?
 1390/1540

Since the work of Cyert and March (1963) or-

ganization theory has viewed routines as funda-
mental building blocks. Thus, the processes that
change routines are very important to study. The
description of the operation of the four prin-
ciples that organize the performance of the task
discussed above shows how a variety of solu-
tions may be explored, but it does not in itself
answer the question of how better solutions may
become the routine operations of the system.
 1391/1540

A classical view of organizational change is that

an analyst looks at the behavior of the system,
represents it explicitly, and plans a better solu-
tion. (See, e.g., Chandler 1966.) The better solu-
tion is expressed as an explicit description of the
system's operation that is subsequently imple-
mented in the real system by somehow altering
the behavior of the participants to bring it into
line with the designed solution. We often think
of the organization of a system as a consequence
of this sort of planning or design. We imagine
an "outside" observer who observes the system's
performance, represents it, operates on the rep-
resentation to determine how to change the sys-
tem, and then uses a channel of communication
from outside the system to effect the changes
(figure 8.3).
 1392/1540

In her work on energy policy analysts, Feldman

(1989) adds some complexity to the processes
by which routines become stable elements of
task performance. She describes organizational
routines as "complex sets of interlocking beha-
viors held in place through common agreement
on the relevant roles and expectations." She says
that "any particular set of agreements about
rules
 P

Figure 8.3 The basic design process. A representation that is

the entire system is created from
 1394/1540

observations of the system. This representation is manipulate

der to plan an intervention in the system.
 1395/1540

and roles is a sort of equilibrium satisfying the dem

of many different parties" (p. 136). A similar view
pressed by Nelson and Winter (1982) when they ch
terize routines as memory, truce, and target. This is
more subtle and interactive sense of the nature of th
solutions to the problem of organization. An organi
has many parts, and the operation of the whole eme
from the interactions of those parts. Each part may
taneously provide constraints on the behavior of oth
parts and be constrained by the behavior of other pa
In chapter 2, I referred to this sort of system of mut
adaptive computational parts as a "cognitive ecolog
This describes the sort of solution discovered by the
igation team on the Palau. The parties to the compu
are the plotter and the bearing recorder, and the dem
on them are constructed in the interactions among t
cognitive processing capabilities, the structure of th
computation, the availability of data, and the fit bet
computational and social organization. They settled
a solution that simultaneously satisfied all these con
straints. In the same vein, Feldman (ibid.) writes: "M
 1396/1540

organizations or parts of organizations must coordin

their behavior
 Page 347

in such a way that each can cope adequately

with the pressures and constraints it has to satis-
fy. While there may be many possible solutions
to such a problem, they are not necessarily easy
to find."
 1398/1540

Given that organizations are the kinds of sys-

tems that consist of many interlocking, interact-
ing, and mutually dependent parts, how can
solutions to the organization problem be dis-
covered? Feldman (ibid.) provides one answer
"Even if one of the participants finds a new
solution that will satisfy the constraints of all
parties, the problems of persuading everyone
else that this would be a beneficial change may
still be considerable." Clearly the process de-
scribed in this passage must happen frequently.
Parts of the behavior of the navigation team fit
this description nicely. The plotter's use of the
normative computation scheme and his attempts
to make that scheme explicit for the recorder are
examples. But this answer is a retreat to the
classical view. It posits a designer, albeit "one of
the participants" who "finds a new solution" and
then must "persuade everyone else" that it is a
good solution. And there remain aspects of the
adaptive responses of the members of the
 1399/1540

navigation team, particularly those involving the

changing division of labor, that are simply not
captured by any description that relies on expli-
cit representation of the shape of the solution.

Adaptation and Local Design

In the analysis presented above, there are no in-

stances of anyone's reflecting on the whole pro-
cess. The plotter seems occasionally to represent
the entire computation, but there is no evidence
he ever imagined the structure of the division of
labor. The adaptation process seemed to take
place by way of local interactions, mostly of two
types.
 1400/1540

First, the members of the team put constraints

on each other by presenting each other with par-
tial computational products. When there is no
previously worked out division of labor and as-
signment of responsibilities for various parts of
the computation, team members negotiate the
division of labor by doing what they can, or
what is convenient, and hoping that others can
do whatever else is required. These are changes
that result from the interactions among the beha-
viors of the parts of the system as they adapt to
the information environment and to the behavi-
ors of other parts. There is no need to invoke
any representation of the behavior of any part of
 Page 348

the system to account for these adaptations. The

way the computation was driven by the availab-
ility of data is an example of this kind of unre-
flective adaptation process. Even though they
are not planned, these changes are not necessar-
ily chaotic. If one part of the system behaves in
a systematic way, another part may come to be-
have in a systematic way by adapting to the be-
havior of the first. In the interaction between the
plotter and the recorder we saw that the behavi-
or of one subsystem can be entrained by that of
another.
 1402/1540

A second adaptive process involves local

design. When implicit negotiations of the divi-
sion of labor fail, an actor may become aware of
his inability to keep up with the computation
and attempt to recruit others to take over parts of
it. Thus, the most striking thing the plotter said
during the search for a new configuration was
something he said to the recorder while falling
behind in his attempts to compute bearing cor-
rections with a pocket calculator. He pushed the
calculator at the bearing recorder and said
"Here, add these things." There is no need to at-
tribute a global awareness of the process to the
plotter to account for this. He doesn't have
enough time to do his own work, let alone to re-
flect on the overall division of labor. He is just
acutely aware that he is falling behind and that
he needs help to catch up. This is a case of local
design. As figure 8.4 shows, design processes
may be local to subsystems. This figure depicts
 1403/1540

an overall system that can change in three

modes:

1. without any design activity at all, through the

adaptive interactions among the subsystems

2. through local design activities in which ma-

nipulations are performed on representations of
local subsystems in order to discover more ad-
aptive relationships with the subsystem's envir-
onment (These changes may, in turn, lead to ad-
aptive changes, either designed or not, by the
other subsystems.)

3. through classical global design activities in

which the representation is of the entire system
of interest.
 1404/1540

Modes 1 and 2 are processes that may lead the

system to a a local minimuma nonoptimal solu-
tion from which it is not possible to reach an op-
timal solution. The third mode is supposed to
guard against that possibility. The response of
the system to the change in its environment was
eventually successful; however, it was the con-
sequence of a large number of local interactions
and adjustments, some of which led the system
away from the eventual solution. Many of these
adjustments appear to have been local design
 Cognition p349

Figure 8.4 Local design activity. Subsystems interact with

another's behaviors.
Representations of local subsystem behavior are created an
changes to subsystem operation
These changes may trigger adaptive responses in
 1406/1540

decisions by the participants. Before its discovery b

however, the final configuration appears not to hav
derstood by any of the participants. To the extent th
ful adaptation to a changing environment counts as
this is a case of organizational learning.

Evolution and Design

It seems to me that there is an important difference

change via supervisory reflection and intervention i
view and the process of change via local adjustmen
strongly resembles the difference between design a
1964).

Both evolution and design can be characterized as s

search is conducted by the system in terms of itself
ducted by an "outsider" on representations of the sy
search is the process of adaptation (see Weick's (19
the design search precedes and guides an implemen
intended to be
 Page 350

adaptive. Pure evolution is, in fact, a process

without design (see Dawkins 1986). What we
see in the case of the adaptation by the naviga-
tion team is an organizational change that is pro-
duced in part by an evolutionary process (adapt-
ive search without representation of the search
space) and in part by a process that lies between
evolution and classical global-perspective
design.
 1408/1540

From this perspective, human institutions can be

quite complex because they are composed of
subsystems (persons) that are "aware" in the
sense of having representations of themselves
and their relationships with their surroundings.
Whether we consider a particular change at the
upper system level to be the result of evolution
or the result of design depends on what we be-
lieve about the scope of the awareness of the
subsystems. If we think that some of the subsys-
tems have global awareness, and that they can
represent and anticipate the consequences of
possible changes, then we may view an organiz-
ational change as a result of design. If we be-
lieve that the subsystems do not form and ma-
nipulate representations of system operation,
then we must view organizational change as
evolutionary. What do we say when the indi-
vidual subsystems only engage in local design
activitysay, crying out for help when one is
overworked? In that case, design is clearly
 1409/1540

involved, and the change in the local environ-

ment of the individual that adapts this way is a
designed change. Now, that local designed
change may have undesigned and unanticipated
consequences for other parts of the system. It
may thus provoke local adaptations by other
parts of the system as all the parts seek (either
by design or not) to satisfy the new environment
of constraints produced by the changes in the
behaviors of other parts. Ultimately, this process
may produce a change in the behavior of the
system as a whole. Even when many local
design decisions are involved, such an adapta-
tion at the system level appears to be evolution-
ary in the sense that the systemlevel change that
resulted was never represented. I believe that
most of the phenomena labeled as social or or-
ganizational "evolution" are instances of this
kind of change.
 1410/1540

Is the navigation task setting primarily the

product of evolution or of design? Every parti-
cipant can be both inside and outside in some
sense. The changes in the organization of the
navigation team were brought about by changes
in the thinking of the participants of the sys-
temthat is, by changes in the agreements about
rules and roles that constitute the organizational
routine. To this extent, the structure of the set-
ting is a product of design. But since the ob-
 Page 351

served reorganization was never fully represen-

ted by any of the participants in the system, the
actors' designs alone cannot account for the
solution that was achieved. Thus, the organiza-
tion of the navigation task is also a product of
evolution. Although the participants may have
represented and thus learned the solution after it
came into being, the solution was clearly dis-
covered by the organization before it was dis-
covered by any of the participants.
 1412/1540

The solution to the problem of organizing work

that was discovered by the navigation team was
not saved in the system. The conditions for the
reproduction of this piece of knowledge are
quite rare. The participants who were directly
involved in this event eventually separated from
the Navy without ever encountering this situ-
ation again. One of them went on to a position
aboard a civilian oil tanker, so perhaps the
knowledge constructed in this event will
someday be reproduced in a different organiza-
tional setting.

The fact that the solution was not ultimately

saved does not diminish this event's standing as
an example of the processes of cultural innova-
tion. The processes by which work is accom-
plished, by which people are transformed from
novices into experts, and by which work prac-
tices evolve are all the same processes.
 Page 353

9 Cultural Cognition

The Costs of Failing to See Cognition as a Cul-

tural Process

In this book I have tried to provide a coherent

account of cognition and culture as parts of a
larger system. This view is not widespread in
cognitive science. Yet, there are unnoticed costs
in failing to see cognition as part of a cultural
process.

Marginalization of Culture
 1414/1540

Early in the development of cognitive science,

culture was relegated to a peripheral role. As
Gardner (1985) pointed out, culture, history,
context, and emotion were all set aside as prob-
lems to be addressed after a good understanding
of individual cognition had been achieved. It is
unfortunate that many anthropologists have en-
couraged this view by thinking of culture as
some collection of things. Tylor (1871) defined
culture as "that complex whole which includes
knowledge, belief, art, morals, law, custom, and
any other capabilities and habits acquired by
man as a member of society." Goodenough
(1957) gave cognitive anthropology its founding
ideational definition of culture: "whatever it is
one must know in order to behave appropriately
in any of the roles assumed by any member of a
society." This view has developed in cognitive
anthropology over the years. Attempting to
define a role for anthropology in cognitive sci-
ence, D'Andrade (1981) proposed an intellectual
 1415/1540

distribution of labor in which psychologists

would be responsible for the cognitive processes
and anthropologists would be responsible for
cognitive content. In this view, culture became
simply a pool of ideas that are operated on by
cognitive processes. Tylor's definition stresses
the acquisition of cultural entities and tries to
give a catalog of abilities and artifacts that con-
stitute culture. Goodenough's definition was cru-
cial to the birth of cognitive anthropology, but it
and D'Andrade's formulation completely ignore
the material aspects of culture. I reject both of
these definitions.
 Page 354

Culture is not any collection of things, whether

tangible or abstract. Rather, it is a process. It is a
human cognitive process that takes place both
inside and outside the minds of people. It is the
process in which our everyday cultural practices
are enacted. I am proposing an integrated view
of human cognition in which a major component
of culture is a cognitive process (it is also an en-
ergy process, but I'm not dealing with that) and
cognition is a cultural process.
 1417/1540

Anthropologists are also guilty of accepting this

marginalization of culture, or even enhancing it,
by granting a special place to the powers and
limitations of the mind, as if these can be estab-
lished without reference to culture. Anthropolo-
gical structuralism tries to read the properties of
minds from the structure of public representa-
tions. Sahlins (1976) criticizes it as follows: "It
would seem ... that the main problem of 'reduc-
tionism' besetting modern structuralism has con-
sisted in a mode of discourse which, by giving
mind all the powers of 'law' and 'limitation,' has
rather placed culture in a position of submission
and dependence. The whole vocabulary of 'un-
derlying' laws of the mind accords all force of
constraint to the mental side, to which the cul-
tural can only respond, as if the first were the
active element and the latter only passive."
 1418/1540

Marginalizing culture by reducing it to some

collection of ideational contents hides the many
ways in which cognition is part of the cultural
process. Culture is a process, and the "things"
that appear on list-like definitions of culture are
residua of the process. Culture is an adaptive
process that accumulates partial solutions to fre-
quently encountered problems. It is unfortunate
that cognitive science left culture, context, and
history to be addressed after the understanding
of the individual had matured. The understand-
ing of the individual that has developed without
consideration of cultural process is fundament-
ally flawed. The early researchers in cognitive
science placed a bet that the modularity of hu-
man cognition would be such that culture, con-
text, and history could be safely ignored at the
outset, and then integrated in later. The bet did
not pay off. These things are fundamental as-
pects of human cognition and cannot be com-
fortably integrated into a perspective that
 1419/1540

privileges abstract properties of isolated indi-

vidual minds. Some of what has been done in
cognitive science must now be undone so that
these things can be brought into the cognitive
picture.
 Page 355

Mistaking the Properties of the System for Those

of the Individual

Another cost of failing to see the cultural nature

of cognition is that it leads us to make too much
of the inside/outside boundary or to assume the
primacy of that boundary over other delimita-
tions of cognitive systems.

CONSTRUCTION OF PRIMITIVE
THOUGHT
 1421/1540

A surprising side effect of the heavily drawn in-

side/outside boundary is that it reinforces the
idea that individuals in primitive cultures have
primitive minds. The firm drawing of the inside/
outside boundary creates the impression that in-
dividual minds operate in isolation and encour-
ages us to mistake the properties of complex so-
ciocultural systems for the properties of indi-
vidual minds. If one believes that technology is
the consequence of cognitive capabilities, and if
one further believes that the only place to look
for the sources of cognitive capabilities is inside
individual minds, then observed differences in
level of technology between a ''technologically
advanced" and a "technologically primitive" cul-
ture will inevitably be seen as evidence of ad-
vanced and primitive minds. Differences in
mental capacity seem necessary to account for
differences in level of technology. I tried to
show in chapters 2-6 that moving the boundaries
of the unit of cognitive analysis out beyond the
 1422/1540

skin reveals other sources of cognitive accom-

plishment. These other sources are not mysteri-
ous, they simply arise from explicable effects
that are not entirely internal to the individual.

Overattribution
 1423/1540

Overlooking the cultural nature of cognition has

another costone that may be the most interesting
and far-reaching for the field of cognitive sci-
ence itself. When one commits to the notion that
all intelligence is inside the inside/outside
boundary, one is forced to cram inside
everything that is required to produce the ob-
served behaviors. Much of cognitive science is
an attribution problem. We wish to make asser-
tions about the nature of cognitive processes that
we cannot, in general, observe directly. So we
make inferences on the basis of indirect evid-
ence instead, and attribute to intelligent systems
a set of structures and processes that could have
produced the observed evidence. That is a ven-
erable research strategy, and I have no objection
to it in principle. However, failing to recognize
 Page 356

the cultural nature of cognitive processes can

lead to a misidentification of the boundaries of
the system that produced the evidence of intelli-
gence. If we fail to bound the system properly,
then we may attribute the right properties to the
wrong system or (worse) invent the wrong prop-
erties and attribute them to the wrong system. In
this attribution game, there has been a tendency
to put much more inside than should be there.

How Cognitive Science Put Symbols in the

Head

If there are fundamental deficiencies in the dom-

inant conceptions of cognition in cognitive sci-
ence, how did that come about?
 1425/1540

It is sometimes difficult to say things that are

quite simple. The words we must say are simple,
but sometimes it takes a lot of work to construct
the conceptual framework in which those simple
words have the right meanings. There are many
possible readings for the sentences I want to
write. In the previous chapters I tried to con-
struct some of the conceptual background that
will allow me now to say some simple things.
However, one hurdle remains. Some of what I
have done here departs from the mainstream of
cognitive science. And some of the unexamined
assumptions of the field make my words unruly.
What I want to say cannot be said simply in that
framework.
 1426/1540

In order to construct a new framework, I will

have to deconstruct the old one. In what follows
I will give a brief "Official" History of Cognit-
ive Science. This is a history as seen by the pro-
ponents of the currently dominant paradigm. I
will then reread the history of cognitive science
from a sociocultural perspective. In doing this I
will identify a number of problems in contem-
porary cognitive science and attempt to give
new meanings to some of the familiar events in
its history.

The "Official" History of Cognitive Science

I begin the official history of cognitive science

with a quote from Herbert Simon and Craig Ka-
plan (1989): "The computer was made in the im-
age of the human."
 1427/1540

The ideas on which cognitive science is based

are so deeply ingrained in our culture that we
can scarcely see how things could be otherwise.
The roots of representationalism go back at least
to Descartes.
 Page 357

Dreyfus (1992) summarizes the history of Good

Old Fashioned Artificial Intelligence (GOFAI)
as follows:

GOFAI is based on the Cartesian idea that all

understanding consists in forming and using ap-
propriate symbolic representations. For Des-
cartes, these representations were complex de-
scriptions built up out of primitive ideas or ele-
ments. Kant added the important idea that all
concepts are rules for relating such elements,
and Frege showed that the rules could be form-
alized so that they could be manipulated without
intuition or interpretation.

The entities that are imagined to be inside the

mind are modeled on a particular class of entit-
ies that are outside the mind: symbolic
representations.
 1429/1540

Symbolic logic has a special place in the history

of cognitive science. The idea that a computer
might be in some way like a person goes back to
the formalization of logic and mathematics. In
the early years of cognitive science, develop-
ments in information theory, neuroscience, psy-
chology, and computer science came to have a
synergistic interrelationship. In information the-
ory the notion of a binary digit (bit) as the fun-
damental unit fit with speculations by McCul-
loch and Pitts that neurons could be character-
ized as on/off devices. Thus, the brain might be
seen as a digital machine (this turned out to be
wrong, but at the time that did not interfere with
the developing synergy). Both of these ideas fit
well with Turing's work showing that any func-
tion that could be explicitly specified could be
computed by a class of machine called a univer-
sal machine and with his demonstration that the
imaginary Turing Machine that operated on a
 1430/1540

binary code was an example of a universal

machine.

The symbol-processing model of cognition has

something else going for it as well: "A universal
machine can be programmed to compute any
formally specified function. This extreme plasti-
city in behavior is one of the reasons why com-
puters have from the very beginning been
viewed as artifacts that might be capable of ex-
hibiting intelligence." (Pylyshyn 1989: 54) This
was an essential component of the history of the
field. Referring to the human cognitive architec-
ture, Newell et al. (1989: 103) say that "the
central function of the architecture is to support
a system capable of universal computation." By
choosing a formalism that is capable of any spe-
cifiable computation, the early theorists were
surely casting a wide enough net to capture hu-
man cognition, whatever it might
 Page 358

turn out to be. It seemed that the only viable

challenge to this view would be a demonstration
that human cognition might not be formally spe-
cifiable. There are many varieties of systems
capable of universal computation. Newell and
his colleagues and most others in the classical
camp have taken what is called a "physical sym-
bol system" as the primary architecture of hu-
man cognition. "A physical symbol system is an
instance of a universal machine. Thus the sym-
bol system hypothesis implies that intelligence
will be realized by a universal computer."
(Newell and Simon 1990) Newell and Simon
(ibid.) define a physical symbol system this
way:
 1432/1540

A physical symbol system consists of a set of en-

tities, called symbols, which are physical pat-
terns that can occur as components of another
type of entity called an expression (or symbol
structure). Thus a symbol structure is composed
of a number of instances (or tokens) of symbols
related in some physical way (such as one token
being next to another). At any instant of time the
system will contain a collection of these symbol
structures. Beside these structures, the system
also contains a collection of processes that op-
erate on expressions to produce other expres-
sions: processes of creation, modification, re-
production, and destruction. A physical symbol
system is a machine that produces through time
an evolving collection of symbol structures.
Such a system exists in a world of objects wider
than just these symbolic expressions themselves.
 1433/1540

According to Pylyshyn (1989), the notion of

mechanism that underlies the classical concept
of cognition is "concerned only with abstractly
defined operations such as storing, retrieving,
and altering tokens of symbolic codes."

Simon and Kaplan (1989) cite the Logic Theor-

ist of Newell and Simon (1956) as an example
of abstract intelligence and note the role of psy-
chological research in its design:

The earliest artificial intelligence programs (for

example, the Logic Theorist (Newell and Simon
1956)) are perhaps best viewed as models of ab-
stract intelligence; but nonetheless their design
was informed by psychological research on
memory and problem solvingnote, for example,
the use of associative structures in listpro-
cessing programming languages and sub-
sequently the frequent use of means-ends ana-
lysis for inference.
 1434/1540

By embodying the growing knowledge of hu-

man informationprocessing psychology in com-
puter programs, the early researchers
 Page 359

were able to express their theories about cogni-

tion as working models that, in many cases,
were capable of actually reproducing many im-
portant aspects of the behavior of human
subjects.
 1436/1540

Artificial intelligence (AI) and information-pro-

cessing psychology thus have a synergistic rela-
tionship to each other. Information-processing
psychology investigates humans as information
processors via the computational metaphor of
mind, while AI investigates machine imple-
mentations of intelligent processes. The opera-
tion of machines that are purportedly built in the
image of humans is believed to shed light on
natural human intelligence. Since the properties
of abstract systems of intelligence are not de-
pendent on the implementational details of the
machines on which they run, intelligence in gen-
eral (in addition to specifically human intelli-
gence) can be investigated with this technology.
The hope is that these traditions will continue to
synergistically feed each other. In the most op-
timistic versions of the story, AI and
information-processing psychology are the prin-
cipal motors of scientific progress in cognitive
science.
 1437/1540

An Alternative History of Cognitive Science

 1438/1540

Let us back up and examine the history of com-

puters a bit more. The digital computer is a
physical device that can support a mechanized
version of a formal system. And it is this capa-
city that makes it a potential model of intelli-
gence. Understanding computers requires an un-
derstanding of formal systems. We know that
formal systems go back several thousand years
in the history of our species. I do not know
when the formal aspects of formal systems were
first understood. I suspect that real understand-
ing of the formal aspects of formal systems did
not come until the revolutionary work on math-
ematics and logic at the beginning of this cen-
tury that was critical to the foundation of cognit-
ive science. Formal systems themselves are
much older than our explicit understanding of
them. The first arithmetic systems are at least
3000 years old, so we may take that as a minim-
um age of formal systems in the human experi-
ence. The idea of a formal system is that there is
 1439/1540

some world of phenomena, and some way to en-

code the phenomena as symbols. The symbols
are manipulated by reference to their form only.
We do not interpret the meanings of the symbols
while they are being manipulated. The manipu-
lation of the symbols results in some other sym-
bolic expression. Finally, we may interpret a
newly
 Page 360

created string of symbols as meaning something

about the world of phenomena.
 1441/1540

Being able to find sets of syntactic manipula-

tions of symbols that preserve this relationship
so that we can reinterpret symbolic expressions
into the world is of paramount importance. As
Pylyshyn (1989) says: "One might ask how it is
possible for symbolic expressions and rules to
keep maintaining their semantic interpretation,
to keep the semantics of the expressions coher-
ent. It is one of the important discoveries of
formal logic that one can specify rules that oper-
ate on symbolic expressions in such a way that
the sequence of expressions always corresponds
to a proof." If we built the right formal system,
we could now describe states of affairs in the
world that would have been impossible or im-
practical to observe directly. Such a state of af-
fairs might be something in the future, which we
cannot observe directly, but which can be pre-
dicted. I consider the mastery of formal systems
to be the key to modern civilization. This is a
very, very powerful idea.
 1442/1540

The system of ship navigation that I have

presented in this book is based on formal manip-
ulations of numbers and of the symbols and
lines drawn on a chart. It is a system that ex-
ploits the powerful idea of formal operations in
many ways. But not all the representations that
are processed to produce the computational
properties of this system are inside the heads of
the quartermasters. Many of them are in the cul-
turally constituted material environment that the
quartermasters share with and produce for each
other.
 1443/1540

Now, here is what I think happened. It was dis-

covered that it is possible to build machines that
can manipulate symbols. The computer is noth-
ing more than an automated symbol manipulat-
or. And through symbol manipulation one can
not only do things we think of as intelligent, like
solving logical proofs or playing chess; we
know for a fact that through symbol manipula-
tion of a certain type it is possible to compute
any function that can be explicitly specified. So,
in principle, the computer could be an intelligent
system. The mechanical computers conceived
by Charles Babbage to solve the problem of un-
reliability in human compilers of mathematical
and navigational tables were seen by his ad-
mirers to have replaced the brain: "The won-
drous pulp and fibre of the brain had been sub-
stituted by brass and iron; [Babbage] had taught
wheelwork to think" (H. W. Buxton, cited in
Swade 1993). Of course, a century later it would
 1444/1540

be vacuum tubes that created the "electronic

brain."
 Page 361

But something got lost in this move. The origin

myths of cognitive science place the seminal in-
sights of Alan Turing in his observations of his
own actions. Dennett (1991) describes the con-
text of Turing's discoveries:

He was thinking self-consciously and introspect-

ively about just how he, a mathematician, went
about solving mathematical problems or per-
forming computations, and he took the import-
ant step of trying to break down the sequence of
his mental acts into their primitive components.
"What do I do," he must have asked himself,
"when I perform a computation? Well, first I ask
myself which rule applies, and then I apply the
rule, and then write down the result, and then I
look at the result, and then I ask myself what to
do next, and...."
 1446/1540

Originally, the model cognitive system was a

person actually doing the manipulation of the
symbols with his or her hands and eyes. The
mathematician or logician was visually and
manually interacting with a material world. A
person is interacting with the symbols and that
interaction does something computational. This
is a case of manual manipulation of symbols.

Notice that when the symbols are in the environ-

ment of the human, and the human is manipulat-
ing the symbols, the cognitive properties of the
human are not the same as the properties of the
system that is made up of the human in interac-
tion with these symbols. The properties of the
human in interaction with the symbols produce
some kind of computation. But that does not
mean that that computation is happening inside
the person's head.
 1447/1540

John Searle's "Chinese room" thought experi-

ment provides a good example of this effect.
Imagine a room inside of which sits the philo-
sopher Searle. Chinese people come up to the
room and push strings of Chinese characters
through a slot in the door. Searle slips back oth-
er strings of characters, which the Chinese take
to be clever responses to their questions. Now,
Searle does not understand Chinese. He doesn't
know the meaning of any Chinese character. To
him, the characters of written Chinese are just a
bunch of elaborate squiggles. However, Searle
has with him in the room baskets of Chinese
characters, and he has a rulebook which says
that if he gets certain sequences of characters he
should create certain other sequences of charac-
ters and slide them out the slot.

Searle intends his thought experiment as a

demonstration that syntax is not sufficient to
produce semantics. According to Searle,
1448/1540
 Page 362

the room appears to behave as though it under-

stands Chinese; yet neither he nor anything in
the room can be said to understand Chinese.
There are many arguments for and against
Searle's claims, and I will not review them here.
Instead, I want to interpret the Chinese room in
a completely different way: The Chinese room
is a sociocultural cognitive system. The really
nice thing about it is that it shows us very
clearly that the cognitive properties of the per-
son in the room are not the same as the cognit-
ive properties of the room as a whole. There is
John Searle with a basket of Chinese characters
and a rulebook. Together he and the characters
and rulebook in interaction seem to speak
Chinese. But Searle himself speaks not a word
of Chinese.
 1450/1540

Let us be clear, then, on the distinction between

the cognitive properties of the sociocultural sys-
tem and the cognitive properties of a person
who is manipulating the elements of that
system.
 1451/1540

The heart of Turing's great discovery was that

the embodied actions of the mathematician and
the world in which the mathematician acted
could be idealized and abstracted in such a way
that the mathematician could be eliminated.
What remained was the essence of the applica-
tion of rules to strings of symbols. For the pur-
poses of producing the computation, the way the
mathematician actually interacted with the ma-
terial world is no more than an implementational
detail. Pylyshyn (1989) claims that while Turing
was developing the notion of the mechanically
"effective procedure" he was looking "at what a
mathematician does in the course of solving
mathematical problems and distilling this pro-
cess to its essentials." The question of what con-
stitutes the essentials here is critical. For Turing,
the essentials evidently involve the patterns of
manipulations of the symbols, but they ex-
pressly do not involve the psychological pro-
cesses which the mathematician uses in order to
 1452/1540

accomplish the manipulations. The essentials of

the abstract manipulation of symbols are pre-
cisely not what the person does. What Turing
modeled was the computational properties of a
sociocultural system.

When the manipulation of symbols is auto-

mated, neither the cognitive processes nor the
activity of the person who manipulated the sym-
bols is modeled. The symbols themselves are
dematerialized and placed inside the machine, or
fed to it in a form that permits the straightfor-
ward generation of internal representations.
What is important about this is that all the prob-
lems the mathematician faced when interacting
with a world of material symbol to-
 Page 363

kens are avoided. That is good news, if those

things are considered unimportant, because they
are a nuisance to model anyway. The rulebook
(or the mathematician's scribbled notations of
rules) is replaced by abstract rules, also inside
the computer. The mathematician who was a
person interacting with a material world is
neither modeled by this system nor replaced in it
by something else. The person is simply absent
from the system that performs automatic symbol
manipulation. What is modeled is the abstract
computation achieved by the manipulation of
the symbols.
 1454/1540

All that is fine if your goal is to extend the

boundaries of human computational accom-
plishments. But it is not necessarily a model of
the processes engaged in by a person doing that
task. These programs produce the properties, not
of the person, but of the sociocultural system.
This is a nontrivial accomplishment. But the
culture of cognitive science has forgotten these
aspects of its past. Its creation myths do not in-
clude this sort of analysis. The physicalsymbol-
system architecture is not a model of individual
cognition. It is a model of the operation of a so-
ciocultural system from which the human actor
has been removed.
 1455/1540

Having failed to notice that the central metaphor

of the physicalsymbol-system hypothesis cap-
tured the properties of a sociocultural system
rather than those of an individual mind, AI and
information-processing psychology proposed
some radical conceptual surgery for the modeled
human. The brain was removed and replaced
with a computer. The surgery was a success.
However, there was an apparently unintended
side effect: the hands, the eyes, the ears, the
nose, the mouth, and the emotions all fell away
when the brain was replaced by a computer.

The computer was not made in the image of the

person. The computer was made in the image of
the formal manipulations of abstract symbols.
And the last 30 years of cognitive science can be
seen as attempts to remake the person in the im-
age of the computer.
 1456/1540

It is no accident that the language of the

physical-symbol-system hypothesis captures so
much of what is happening in domains like ship
navigation. The physical-symbol-system hypo-
thesis is based on the operation of systems of
this type. Conversely, there is nothing meta-
phorical about talking about the bearing record
book as a memory, or about viewing the erasure
of lines drawn in pencil on a chart as forgetting.
Sometimes my colleagues ask me whether I feel
safe metaphorically extending the language of
what's happening
 Page 364

inside people's heads to these worlds. My re-

sponse is "It's not a metaphorical extension at
all." The computer was made in the image of the
sociocultural system, and the human was re-
made in the image of the computer, so the lan-
guage we use for mental events is the language
that we should have used for these kinds of so-
ciocultural systems to begin with. These are not
examples of metaphorical extension from the
base domain of mental events to the target do-
main of cultural activity. Rather, the original
source domain for the language of thought was a
particular highly elaborated and culturally spe-
cific world of human activity: that of formal
symbol systems.
 1458/1540

At first, the falling away of the apparatus that

connects the person to the world went un-
noticed. This may have been because there was
a lot of justifiable excitement about what could
be done with this technology. All that remained
of the person, however, was the boundary
between the inside and the outside. And this
boundary was not the same as the boundary of
the Chinese room. The boundary that remained
was assumed to be the boundary of the per-
sonthe skin or the skull. In fact, it was the
boundary of the formal system. The boundary
between inside and outside became the bound-
ary between abstract symbols and the world of
phenomena described by symbols. The walls of
the Chinese room were mistaken for the skin of
the person. And the walls of the room surroun-
ded the symbols, so the symbols were assumed
to be inside the head.
 1459/1540

This separation between the boundaries of the

formal system and the skin shows up in the lan-
guage of cognitive science. "Symbol systems
are an interior milieu, protected from the extern-
al world, in which information processing in the
service of the organism can proceed." (Newell et
al. 1989: 107 [my emphasis]). Or:

Act*, as is typical of many theories of cognition,

focuses on the central architecture. Perception
and motor behavior are assumed to take place
in additional processing systems off stage. Input
arrives in working (memory), which thus acts as
a buffer between the unpredictable stream of en-
vironmental events and the cognitive system.
(ibid.: 117)
 1460/1540

The "off stage" metaphor of Newell et al. ex-

presses the isolation of the cognitive system
from even the sensory and motor experiences of
an organism. In fact, many cognitive scientists
take the word 'cognitive' as an antonym to 'per-
ceptual' or 'motor'. Here is a
Page 365
 1462/1540

typical example of this usage: ''This is especially

true for tasks that are primarily cognitive, in
which perceptual and motor operations play
only a small role in the total sequence." A strong
claim about the modularity of the human cognit-
ive system is implicit in this use of language. It
places a large divide between cognition and the
world of experience. But the existence of per-
ceptual and motor processes that are distinct and
separate from so-called cognitive processes is
not an empirical fact: it is simply a hypothesis
that was made necessary by having constructed
cognition out of a mechanized formal symbol
processing system. Proponents of the physical-
symbol-system hypothesis point to the existence
of various sensory and motor memories that can
act as buffers between cognition and the world
of experience as evidence of this modularity. In
fact, there may be many other uses for such buf-
fers. We are unlikely to discover these other
uses, however, as long as we keep cognition
 1463/1540

isolated from the world. For example, such buf-

fers may be essential in maintaining training sig-
nals after the disappearance of stimuli while
learning is taking place.

The model of human intelligence as abstract

symbol manipulation and the substitution of a
mechanized formal symbol-manipulation system
for the brain result in the widespread notion in
contemporary cognitive science that symbols are
inside the head. The alternative history I offer is
not really an account of how symbols got inside
the head; it is a historical account of how cognit-
ive science put symbols inside the head. And
while I believe that people do process symbols
(even ones that have internal representations), I
believe that it was a mistake to put symbols in-
side in this particular way. The mistake was to
take a virtual machine enacted in the interac-
tions of real persons with a material world and
make that the architecture of cognition.
 1464/1540

This mistake has consequences. Why did all the

sensorimotor apparatus fall off the person when
the computer replaced the brain? It fell off be-
cause the computer was never a model of the
person to begin with. Remember that the sym-
bols were outside, and the apparatus that fell off
is exactly the apparatus that supported interac-
tion with those symbols. When the symbols
were put inside, there was no need for eyes,
ears, or hands. Those are for manipulating ob-
jects, and the symbols have ceased to be materi-
al and have become entirely abstract and
ideational. The notion of abstractness was ne-
cessary to bleach the material aspect out of the
symbols so that they could become freed from
any particular
 Page 366

material instantiation. Calling logic and math-

ematics "abstract" more than misses the point of
their concrete nature as human activities; it ob-
scures it in a way that allows them to be impor-
ted into a cognitive inner sanctum. The physic-
ality of material symbols in the environment has
been replaced by the physicality (causality) of
the computer; thus, while the physical is ac-
knowledged in the physical-symbol-system hy-
pothesis, it is rendered irrelevant by the claim
that the physical aspect is an implementational
detail. This idea may also help to explain the in-
difference that cognitive science generally
shows to attempts to study implementation in
real human systems.
 1466/1540

Observe how a proponent of the classical view

treats the manipulation of a computational arti-
fact. Here Pylyshyn (1989: 56) constructs an ex-
ample of manipulations of symbols that are
codes for numbers:
 1467/1540

If you can arrange for the computer to trans-

form them systematically in the appropriate
way, the transformations can correspond to use-
ful mathematical operations such as addition or
multiplication. Consider an abacus. Patterns of
beads represent numbers. People learn rules for
transforming these patterns of beads in such a
way that the semantic interpretation of before-
andafter pairs corresponds to a useful mathem-
atical function. But there is nothing intrinsically
mathematical about the rules themselves; they
are just rules for moving beads around. What
makes the rules useful for doing mathematics is
that we are assured of a certain continuing cor-
respondence between the formal or syntactic
patterns of beads and mathematical objects
(such as numbers).
 1468/1540

There are no hands or eyes in this description.

There are only the formal properties of the pat-
terns of beads. Pylyshyn is using the example of
the abacus to show how the manipulation of
symbols produces computations. He provides a
very nice illustration of the power of this cultur-
al artifact. He is not interested in what the per-
son does, or in what it means for a person to
learn, to "know," or to apply a rule. Rather, he is
interested in the properties of the system enacted
by the person manipulating the physical beads.
That is fine as a description of the computation-
al properties of a sociocultural system, but to
take this as being about cognitive processes in-
side the skin is to recapitulate the error of mis-
taking the properties of the sociocultural system
for the properties of a person. It is easy to do. It
is something we do in our folk psychology all
the time. But
 Page 367

when one is really careful about talking about

cognition, one must carefully distinguish
between the tasks that the person faces in the
manipulation of symbolic tokens and the tasks
that are accomplished by the manipulation of the
symbolic tokens.
 1470/1540

A failure to do this has led to a biased view of

the tasks that are properly considered cognitive.
Problem solving by heuristic search is taken as a
representative cognitive activity. This is tailor-
made for the symbol-shuffling apparatus. The
definition of cognition has been unhooked from
interaction with the world. Research on games
and puzzles has produced some interesting in-
sights, but the results may be of limited general-
ity. The tasks typically chosen for laboratory
studies are novel ones that are regarded by sub-
jects as challenging or difficult. D'Andrade
(1989) has likened the typical laboratory cognit-
ive tasks to feats of athletic prowess. If we want
to know about walking, studying people jump-
ing as high as they can may not be the best ap-
proach. Such tasks are unrepresentative in an-
other sense as well. The evolution of the materi-
al means of thought is an important component
of culturally elaborated tasks. It permits a task
that would otherwise be difficult to be re-coded
 1471/1540

and re-represented in a form in which it is easy

to see the answer. This sort of development of
material means is intentionally prohibited in
puzzle tasks because to allow this sort of evolu-
tion would destroy the puzzling aspects of the
puzzle. Puzzles are tasks that are preserved in
the culture because they are challenging. If the
performance mattered, we would learn to re-rep-
resent them in a way that removed the chal-
lenge. That would also remove their value as
puzzles, of course. The point is that the tasks
that are "typical" in laboratory studies of
thought are drawn from a special category of
cultural materials that have been isolated from
the cognitive processes of the larger cultural
system. This makes these tasks especially unrep-
resentative of human cognition.
 1472/1540

Howard Gardner (1985) is very kind to cognit-

ive science when he says that emotion, context,
culture and history were deemphasized in early
cognitive science because, although everyone
believed they were important, everyone also
knew that they complicated things enormously.
According to Gardner, getting the program star-
ted required a simple model of cognition. The
field therefore deferred consideration of affect,
culture, context, and history until such time as
there was a good model of how an individual
worked in isolation. It was hoped that these
things could be added in later. That is a charit-
able reading of the history, I think. I can see
why
 Page 368

there were compelling reasons to see it as it was

seen, and not to notice that something is wrong
when AI was producing "deaf, dumb, and blind,
paraplegic agents" (Bobrow 1991) as models of
human cognition.
 1474/1540

Newell et al. (1989) seemed genuinely puzzled

by the fact that no one had succeeded in integ-
rating emotion into the system of cognition they
had built. Yet this failure is completely predict-
able from the assumptions that underlie the con-
struction of the symbolmanipulation model of
cognition. The person was simply omitted from
what was taken as the model of the cognitive
system. The model of cognition came from ex-
actly that part of the system that was material
rather than human. Within this underlying the-
ory of cognition there can be no integration of
emotion, because the part of the cultural system
that is the basis of the physical symbol system
excludes emotion. The integration of cognition
with action will remain difficult because the
central hypothesis separates cognition and ac-
tion by definition. History and context and cul-
ture will always be seen as add-ons to the sys-
tem, rather than as integral parts of the cognitive
 1475/1540

process, because they are by definition outside

the boundaries of the cognitive system.

Adherents of the physical-symbol-system hypo-

thesis are obviously aware of the presence of a
world in which action takes place, and they have
attempted to take it into account. Consider the
following passage from Newell and Simon's
seminal 1972 book Human Problem Solving:

For our theory, specification of the external

memories available to the problem solver is ab-
solutely essential. These memories must be spe-
cified in the same terms as those we have used
for the internal memories; symbol capacities,
accessing characteristics, and read and write
times. The problem solving program adopted by
the information-processing system will depend
on the nature of its "built in" internal STM and
LTM [short-term memory and long-term
memory].
 1476/1540

From a functional viewpoint, the STM should be

defined, not as an internal memory, but as the
combination of (1) the internal STM and (2) the
part of the visual display that is in the subject's
foveal view....

In short, although we have few independent data

suited to defining precisely how external
memory can augment STM, the two components
do appear to form a single functional unit as far
as the
 Page 369

detailed specification of a problem solving

information-processing system is concerned.
 1478/1540

This is a good start on the problem, but I think it

is fair to say that in the twenty years since the
publication of Human Problem Solving the use
of material structure in the problem-solving en-
vironment has not been a central topic in the
physical-symbol-system research agenda. Some
recent work within this tradition takes the "ex-
ternal world" into account (Larkin 1989; Vera
and Simon 1993) but treats the world only as an
extra memory on which the same sorts of opera-
tions are applied as are applied to internal
memories. Structure in the world can be much
more than an augment to memory. The use of
cultural structures often involves, not just the
same process with more memory, but altogether
different processes. The overattribution of in-
ternal structure results from overlooking the co-
ordination of what is inside with what is outside.
The problem remains that the nature of the inter-
action with the world proposed in these systems
is determined by the assumptions of the
 1479/1540

symbolic architecture that require the bridging

of some gap between the inner, cognitive world
and an outer world of perception and action.
 1480/1540

These criticisms by themselves are not sufficient

grounds for rejecting the notion that humans are
symbol-processing systems. Newell and Simon
(1990) wisely acknowledge that the
physicalsymbol-system hypothesis is a hypo-
thesis and that the role of symbolic processes in
cognition is an empirical question. It has proved
possible to interpret much of human problem-
solving behavior as if the very architecture of
human cognition is symbol processing. It's a hy-
pothesis. A lot of data can be read as failing to
reject it. Yet, the hypothesis got there under sus-
picious circumstances. There are no plausible
biological or developmental stories telling how
the architecture of cognition became symbolic.
We must distinguish between the proposition
that the architecture of cognition is symbolic
and the proposition that humans are processors
of symbolic structures. The latter is indisput-
able, the former is not. I would like to be able to
show how we got to be symbol manipulators in
 1481/1540

relation to how we work as participants in so-

ciocultural systems, rather than assume it as an
act of faith. The origins of symbolic processes
have not been explored this way, though, be-
cause they were obfuscated by the creation myth
that maintains that the computer was made in
the image of humans.
 Page 370

Increasingly, the physical-symbol-system hypo-

thesis is a perspective into which things don't fit.
It was a bet or a guess, grounded in a nearly reli-
gious belief in the Platonic status of mathemat-
ics and formal systems as eternal verities rather
than as historical products of human activity.
This is an old dispute that lies at the heart of the
developing split in cognitive science between
those who feel there is more to be learned from
the physical-symbol-system research and those
who feel it has been exhausted. (See the special
January-March 1993 issue of Cognitive
Science.) By advocating this alternative view, I
am espousing what might be called a "secular"
view of cognitionone that is grounded in a secu-
lar perspective on formal systems, in contrast
with the quasi-religions "cosmic truth" view put
forth by the symbolists.
 1483/1540

Why cognition became disembodied is clear

from the history of the symbolic movement. An
important component of the solution is to re-em-
body cognition, including the cognition of sym-
bol processing.
 1484/1540

I believe that humans actually process internal

representations of symbols. But I don't believe
that symbol manipulation is the architecture of
cognition. Historically, we simply assumed that
symbol processing was inside because we took
the computer as our model of mentality. Hu-
mans (and, I suspect, most other animals) are
good at detecting regularities in their environ-
ment and at constructing internal processes that
can coordinate with those regularities. Humans,
more than any other species, spend their time
producing symbolic structure for one another.
We are very good at coordinating with the regu-
larities in the patterns of symbolic structure that
we present to one another. As was described in
chapter 7, the internal structures that form as a
consequence of interaction with symbolic mater-
ials can be treated as symbolic representations.
Ontogenetically speaking, it seems that symbols
are in the world first, and only later in the head.
 1485/1540

Studying Cognition in the Wild

Many of the foundational problems in cognitive

science are consequences of our ignorance of
the nature of cognition in the wild. Most of what
we know about cognition was learned in laborat-
ory experiments. Certainly, there are many
things that can be learned only in closely con-
trolled experiments. But little is known about
the relationships of cognition in the captivity of
the laboratory to
 Page 371

cognition in other kinds of culturally constituted

settings. The first part of the job is, therefore, a
descriptive enterprise. I call this description of
the cognitive task world a "cognitive ethno-
graphy." One might have assumed that cognitive
anthropology would have made this sort of work
its centerpiece. It has not. Studying cognition in
the wild is difficult, and the outcomes are
uncertain.
 1487/1540

Cognitive systems like the one documented in

this book exist in all facets of our lives. Unfortu-
nately, few truly ethnographic studies of cogni-
tion in the wild have been performed. (Beach
1988, Frake 1985, Gladwin 1970, Goodwin
1993, Goodwin and Goodwin 1992 and 1995,
Latour 1986, Lave 1988, Lave et al. 1984, Ochs
et al. (in press), Scribner 1984, Suchman 1987,
and Theureau 1990 are lonely exceptions to this
trend.) We trust our lives to systems of this sort
every day, yet this class of phenomena has
somehow fallen into the crack between the es-
tablished disciplines of anthropology and psy-
chology and appears to be excluded by founda-
tional assumptions in cognitive science. This
book is an attempt to show what a natural his-
tory of a cognitive system could be like.
 1488/1540

Among the benefits of cognitive ethnography

for cognitive science is the refinement of a func-
tional specification for the human cognitive sys-
tem. What is a mind for? How confident are we
that our intuitions about the cognitive nature of
tasks we do on a daily basis are correct? It is a
common piece of common sense that we know
what those tasks are because we are human and
because we engage in them daily. But I believe
this is not true. In spite of the fact that we en-
gage in cognitive activities every day, our folk
and professional models of cognitive perform-
ance do not match what appears when cognition
in the wild is examined carefully. I have tried to
show here that the study of cognition in the wild
may reveal a different sort of task world that
permits a different conception of what people do
with their minds.
 1489/1540

Cognitive science was born in a reaction against

behaviorism. Behaviorism had made the claim
that internal mental structure was either irrelev-
ant or nonexistent-that the study of behavior
could be conducted entirely in an objective char-
acterization of behavior itself. Cognitive
science's reaction was not simply to argue that
the internal mental world was important too; it
took as its domain of study the internal mental
environment largely separated from the external
world. Interaction with the world was reduced to
read and write operations conducted at either
end of extensive processing
 Page 372

Figure 9.1 A moment of human practice.

activity. This fit the computer metaphor very

well, but it made the organization of the environ-
ment in which thinking took place seem largely
irrelevant. Both behaviorism and cognitivism
must be wrong.
 1491/1540

Cognition in the Intersection of Cultural

Processes

The cube depicted in figure 9.1 represents any

moment in navigation practice (or, in fact, any
moment in any human practice). The arrows
passing through the cube represent three devel-
opmental sequences of which every moment of
practice is simultaneously a part. I have adopted
some simple conventions to capture several as-
pects of the situation in this single diagram. The
thickness of the arrow represents the density of
interaction among the elements in that dimen-
sion. The length of the shaft of the arrow emer-
ging from the cube represents the rate at which
states in that dimension are changing. The length
of the tail of the arrow going into the cube rep-
resents the duration of the relevant history of the
activity in the given dimension.
 1492/1540

It is essential to keep in mind that these things

are all happening at the same time in the same
activity. Having reinstated a whole human being
in a culturally constituted activity, I see the
following.

The conduct of the activity proceeds by the oper-

ation of functional systems that bring representa-
tional media into coordination with one another.
The representational media may be inside as
Page 373
 1494/1540

well as outside the individuals involved. These

functional systems propagate representational
state across the media. In describing the ongoing
conduct of navigation tasks, it is possible to
identify a number of cognitive systems, some
subsuming others. One may focus on the pro-
cesses internal to a single individual, on an indi-
vidual in coordination with a set of tools
(chapter 3), or on a group of individuals in inter-
action with one another and with a set of tools
(chapter 4). Each system produces identifiable
cognitive properties, and in each case the prop-
erties of the system are explained by reference
to processes that transform states inside the sys-
tem (chapter 5). The structured representational
media in the system interact in the conduct of
the activity. Each medium is put to use in an op-
erational environment constituted by other me-
dia. As indicated in figure 9.1, the conduct of
the activity itself has a relatively short history.
An entry into a harbor, for example, involves a
 1495/1540

few hours of preparation and takes about an

hour to complete. Changes in this dimension
happen quickly, and the elements of the task
performance are in relatively intense interaction
with one another. The conduct of the activity
creates elements of representational structure
that survive beyond the end of the task. These
elementsthe logbooks, pencil marks on charts,
the quartermasters' memories of the eventsare
the operational residua of the process.
 1496/1540

In this adaptive system, the media may be

changed by the very processes that constitute the
conduct of the activity. The operations of the
navigation team produce a structured experience
for the participants that contains opportunities
for individual learning (chapter 6). As a con-
sequence of their participation in the task per-
formance, the quartermasters may acquire in-
ternal organization that permits them to coordin-
ate with the structure of their surroundings. In
this way, learning can be seen as the propaga-
tion of organization through an adaptive system
(chapter 7). The development of the practition-
ers themselves takes years. Through a career, a
quartermaster gradually acquires the skills that
are exercised in the performance of the job.
Changes to the organization of the internal me-
dia that the quartermasters bring to the job take
place more slowly than the changes to the states
that the media support. That is, it takes longer to
learn how to plot a fix, for example, than it does
 1497/1540

to plot a fix. But since most learning in this set-

ting happens in the doing, the changes to intern-
al media that permit them to be coordinated with
external media happen in the same processes
that
 Page 374

bring the media into coordination with one an-

other. The changes to the quartermasters' skills
and the knowledge produced by this process are
the mental residua of the process.
 1499/1540

The setting of navigation work evolves over

time as partial solutions to frequently en-
countered problems are crystallized and saved in
the material and conceptual tools of the trade
and in the the social organization of the work.
The development of the practice takes place
over centuries (chapter 2). The very same pro-
cesses that constitute the conduct of the activity
and that produce changes in the individual prac-
titioners of navigation also produce changes in
the social, material, and conceptual aspects of
the setting. The example given in chapter 8 il-
lustrates the creation in interaction of a new
concept and a shared lexical label for it (the
"total" in the modular form of the true-bearing
computation). The microgenesis of the cultural
elements that make up the navigation setting is
visible in the details of the ongoing practice.
 1500/1540

All this happens simultaneously in cognition in

the wild. It is in this sense that cognition is a
fundamentally cultural process.
 Page 375

References
Alexander, C. 1964. Notes on the Synthesis of
82866990447203f57400372e62d765b5.gif
Form. Harvard University Press.

Anderson, J. 1983. Architecture of Cognition.

82866990447203f57400372e62d765b5.gif
Harvard University Press.

Asimov, I. 1982. Foundation's Edge. Ballantine.

82866990447203f57400372e62d765b5.gif

Aveni, A. F. 1981. Tropical archeoastronomy.

82866990447203f57400372e62d765b5.gif
Science 213 (4504): 161-170.

Bateson, G. 1972. Steps to an Ecology of Mind.

82866990447203f57400372e62d765b5.gif
Ballantine.
 1502/1540

Beach, K. 1988. The role of external mnemonic

symbols in acquiring an occupation. In M. M.
Gruneberg, P. E. Morris, and R. N. Sykes (eds.),
82866990447203f57400372e62d765b5.gif
Practical Aspects of Memory: Current Research
and Issues, volume 1. Wiley.

Bearden, B., and Wedertz, B. 1978. The

Bluejacket's Manual (twentieth edition). U.S.
82866990447203f57400372e62d765b5.gif
Naval Institute.

Benedict, R. 1946. The Chrysanthemum and the

82866990447203f57400372e62d765b5.gif
Sword. Meridian.

Bobrow, D. 1991. Dimensions of interaction.

82866990447203f57400372e62d765b5.gif
AAAI Magazine, fall: 64-80.

Boster, J. S. 1985. Requiem for the omniscient

informant: There's life in the old girl yet. In J.
82866990447203f57400372e62d765b5.gif
Dougherty (ed.), Directions in Cognitive An-
thropology. University of Illinois Press.
 1503/1540

Boster, J. S. 1990. The information economy

model applied to biological similarity judge-
ment. In L. Resnick, J. Levine, and S. Teasley
82866990447203f57400372e62d765b5.gif
(eds.), Perspectives on Socially Shared Cogni-
tion. APA Press.

Bowditch, N. 1977. American Practical Navig-

ator (An Epitome of Navigation). U.S. Defense
82866990447203f57400372e62d765b5.gif
Mapping Agency Hydrographic Center.

Buckhout, R. 1982. Eyewitness testimony. In U.

Neisser (ed.), Memory Observed: Remembering
82866990447203f57400372e62d765b5.gif
in Natural Contexts. Freeman

Chandler, A. 1966. Strategy and Structure. Har-

82866990447203f57400372e62d765b5.gif
vard University Press.

Chandrasekaran, B. 1981. Natural and social

system metaphors for distributed problem solv-
82866990447203f57400372e62d765b5.gif
ing: Introduction to the issue. IEEE Transaction
on Systems, Man, and Cybernetics 11: 1-5.
 1504/1540

Cole, M., and Griffin, P. 1980. Cultural amplifi-

ers reconsidered. In D. R. Olson (ed.), The So-
82866990447203f57400372e62d765b5.gif
cial Foundations of Language and Thought.
Norton.

Cotter, C. 1983a. A brief historical survey of

British navigation manuals. Journal of Naviga-
82866990447203f57400372e62d765b5.gif
tion 36 (2): 237-248.

Cotter, C. 1983b. A brief history of sailing dir-

82866990447203f57400372e62d765b5.gif
ections. Journal of Navigation 36 (2): 249- 261.

Cutler, A., and McShane, R. 1975. Trachten-

berg Speed Mathematics Self-Taught.
82866990447203f57400372e62d765b5.gif
Doubleday.

Cyert, R. M., and March, J. G. 1963. A Behavi-

82866990447203f57400372e62d765b5.gif
oral Theory of the Firm. Prentice-Hall.

D'Andrade, R. G. 1981. The cultural part of cog-

82866990447203f57400372e62d765b5.gif
nition. Cognitive Science 5: 179-195.
 1505/1540

D'Andrade, R. G. 1989. Cultural cognition. In

M. Posner (ed.), Foundations of Cognitive
82866990447203f57400372e62d765b5.gif
Science. MIT Press.

Dawkins, R. 1986. The Blind Watchmaker.

82866990447203f57400372e62d765b5.gif
Norton.

Dennett, D. C. 1991. Consciousness Explained.

82866990447203f57400372e62d765b5.gif
Little, Brown.

Dreyfus, H. L. 1992. What Computers Still

Can't Do: A Critique of Artificial Reason
82866990447203f57400372e62d765b5.gif
(second edition). MIT Press.

Feldman, M. S. 1989. Order without Design.

82866990447203f57400372e62d765b5.gif
University of California Press.

Finney, B. R. 1979. Hokulea: The Way to Tahiti.

82866990447203f57400372e62d765b5.gif
Dodd, Mead.

Finney, B. R. 1991. Myth, experiment, and the

reinvention of Polynesian voyaging. American
82866990447203f57400372e62d765b5.gif
Anthropologist 93(2): 383-404.
 1506/1540

Fleck, L. 1935. The Genesis and Development

of a Scientific Fact. University of Chicago
82866990447203f57400372e62d765b5.gif
Press, 1979.
 Page 376

Frake, C. 1985. Cognitive maps of time and tide

82866990447203f57400372e62d765b5.gif
among medieval seafarers. Man 20: 254- 270.

Fuson, R. H. 1987. The Log of Christopher

82866990447203f57400372e62d765b5.gif
Columbus. International Marine.

Gardner, H. 1985. The Mind's New Science.

82866990447203f57400372e62d765b5.gif
Basic Books.

Geertz, C. 1983. Local Knowledge: Further Es-

82866990447203f57400372e62d765b5.gif
says in Interpretive Anthropology. Basic Books.

Gelman, R., and Gallistel, C. R. 1978. The

Child's Understanding of Number. Harvard
82866990447203f57400372e62d765b5.gif
University Press.

Gentner, D., and Grudin, J. 1985. The evolution

of mental metaphors in psychology: A ninety-
82866990447203f57400372e62d765b5.gif
year retrospective. American Psychologist 40:
181-192.
 1508/1540

Gladwin, T. 1970. East is a Big Bird. Harvard

82866990447203f57400372e62d765b5.gif
University Press.

Goodenough, W. 1953. Native Astronomy in the

82866990447203f57400372e62d765b5.gif
Central Carolines. University of Pennsylvania.

Goodenough, W. 1957. Cultural anthropology

and linguistics. In Report of the Seventh Annual
Round Table Meeting in Linguistics and Lan-
82866990447203f57400372e62d765b5.gif
guage Study (Language and Linguistics mono-
graph 9), Georgetown University.

Goodwin, C. 1993. Perception, Technology and

Interaction on a Scientific Research Vessel. Re-
82866990447203f57400372e62d765b5.gif
search report, University of South Carolina.

Goodwin, C. 1994. Professional vision. Americ-

82866990447203f57400372e62d765b5.gif
an Anthropologist 96 (2): 606-633.
 1509/1540

Goodwin, C., and Goodwin, M. H. 1995. For-

mulating planes: Seeing as situated activity. In
D. Middleton and Y. Engestrom (eds.), Cogni-
82866990447203f57400372e62d765b5.gif
tion and Communication at Work. Cambridge
University Press.

Goody, J. 1977. The Domestication of the Sav-

82866990447203f57400372e62d765b5.gif
age Mind. Cambridge University Press.

Grudin, J. 1988. Why CSCW applications fail:

Problems in the design and evaluation of organ-
izational interfaces. In Proceedings of the
82866990447203f57400372e62d765b5.gif
CSCW'88 Conference on Computer- Supported
Cooperative Work, Portland, Oregon.

Hastie, R., and Kumar, P. 1979. Person

memory: Personality traits as organizing prin-
82866990447203f57400372e62d765b5.gif
ciples in memory for behavior. Journal of Per-
sonality and Social Psychology 37: 25-38.
 1510/1540

Hewson, J. B. 1983. A history of the Practice of

Navigation (second edition). Brown, Son and
82866990447203f57400372e62d765b5.gif
Ferguson.

Hutchins, E. 1983. Understanding Micronesian

navigation. In D. Gentner and A. L. Stevens
82866990447203f57400372e62d765b5.gif
(eds.), Mental Models. Erlbaum.

Hutchins, E. 1991. The social organization of

distributed cognition. In L. Resnick, J. Levine,
82866990447203f57400372e62d765b5.gif
and Stephanie Teasley (eds.), Perspectives on
Socially Shared Cognition. APA Press.

Hutchins, E., and Hinton, G. E. 1984. Why the

82866990447203f57400372e62d765b5.gif
islands move. Perception 13: 629-632.

Hutchins, E., Hollan, J., and Norman, D. A.

1986. Direct manipulation interfaces. In D. A.
Norman and S. Draper (eds.), User Centered
82866990447203f57400372e62d765b5.gif
System Design: New Perspectives in Human-
Computer Interaction. Erlbaum.
 1511/1540

Ifrah, G. 1987. From One to Zero: A Universal

82866990447203f57400372e62d765b5.gif
History of Numbers. Penguin.

Kann, P. 1978. Leningrad in Three Days.

82866990447203f57400372e62d765b5.gif
Progress.

Kirsh, D. 1990. When is information explicitly

represented? In P. Hanson (ed.), Information,
82866990447203f57400372e62d765b5.gif
Thought, and Content. UBC Press.

Kyselka, W. 1987. An Ocean in Mind.

82866990447203f57400372e62d765b5.gif
University of Hawaii Press.

Langacker, R. 1987. Foundations of Cognitive

82866990447203f57400372e62d765b5.gif
Grammar, volume 1. Stanford University Press.

Lantz, D., and Stefflre, V. 1964. Language and

cognition revisited. Journal of Abnormal and
82866990447203f57400372e62d765b5.gif
Social Psychology 69 (5): 472-481.
 1512/1540

Larkin, J. 1989. Display-based problem solving.

In D. Klahr and K. Kotovsky (eds.), Complex
82866990447203f57400372e62d765b5.gif
Information Processing: The Impact of Herbert
A. Simon. Erlbaum.

Latour, B. 1986. Visualization and cognition:

Thinking with eyes and hands. Knowledge and
82866990447203f57400372e62d765b5.gif
Society 6: 1-40.

Latour, B. 1987. Science in Action. Harvard

82866990447203f57400372e62d765b5.gif
University Press.

Lave, J. 1988. Cognition in Practice. Cambridge

82866990447203f57400372e62d765b5.gif
University Press.

Lave, J., Murtaugh, M., and de la Rocha, O.

1984. The dialectic of arithmetic in grocery
shopping. In B. Rogoff and J. Lave (eds.),
82866990447203f57400372e62d765b5.gif
Everyday Cognition: Its Development in Social
Context. Harvard University Press.
 Page 377

Law, J. 1987. Technology and heterogeneous

engineering: the case of the Portuguese expan-
sion. In W. E. Bijker, T. P. Hughes, and T.
82866990447203f57400372e62d765b5.gif
Pinch (eds.), The Social Construction of Tech-
nological Systems. MIT Press.

Levinson, S. 1990. Interactional Biases in Hu-

man Thinking. Working paper 3, Project Group
82866990447203f57400372e62d765b5.gif
in Cognitive Anthropology, Max Planck Gesell-
schaft, Berlin.

Lewis, D. 1972. We the Navigators. University

82866990447203f57400372e62d765b5.gif
of Hawaii Press.

Lewis, D. 1976. A return voyage between Pu-

luwat and Saipan using Micronesian navigation-
82866990447203f57400372e62d765b5.gif
al techniques. In B. R. Finney (ed.), Pacific
Navigation and Voyaging. Polynesian Society.

Lewis, D. 1978. The Voyaging Stars: Secrets of

82866990447203f57400372e62d765b5.gif
Pacific Island Navigators. Norton.
 1514/1540

Lord, C., Lepper, M., and Ross, L. 1979. Biased

assimilation and attitude polarization: The ef-
fects of prior theories on subsequently con-
82866990447203f57400372e62d765b5.gif
sidered evidence. Journal of Personality and So-
cial Psychology 37: 2098-2110.

Maloney, E. 1985. Dutton's Navigation and

Piloting (fourteenth edition). Naval Institute
82866990447203f57400372e62d765b5.gif
Press.

Marr, D. 1982. Vision: A Computational In-

vestication into the Human Representation and
82866990447203f57400372e62d765b5.gif
Processing of Visual Information. Freeman.

Miyake, N. 1986. Constructive interaction and

the iterative process of understanding. Cognitive
82866990447203f57400372e62d765b5.gif
Science 10 (2): 151-177.

National Maritime Museum. 1976. The Plani-

spheric Astrolabe. Her Majesty's Stationery
82866990447203f57400372e62d765b5.gif
Office.
 1515/1540

Neisser, U. 1976. Cognition and Reality.

82866990447203f57400372e62d765b5.gif
Freeman.

Nelson, R., and Winter, S. 1982. An Evolution-

ary Theory of Economic Change. Harvard
82866990447203f57400372e62d765b5.gif
University Press.

Newell, A. 1973. You can't play 20 questions

with nature and win. In W. G. Chase (ed.), Visu-
82866990447203f57400372e62d765b5.gif
al Information Processing. Academic Press.

Newell, A., and Simon, H. A. 1972. Human

82866990447203f57400372e62d765b5.gif
Problem Solving. Prentice-Hall.

Newell, A., and Simon, H. A. 1990. Computer

science as empirical inquiry: Symbols and
search. In J. L. Garfield (ed.), Foundations of
82866990447203f57400372e62d765b5.gif
Cognitive Science: The Essential Readings. Par-
agon House.
 1516/1540

Newell, A., Rosenbloom, P. S., and Laird, J. E.

1989. Symbolic architectures for cognition. In
82866990447203f57400372e62d765b5.gif
M. Posner (ed.), Foundations of Cognitive
Science. MIT Press.

Nickerson, R., and Adams, M. J. 1979. Long

term memory for a common object. Cognitive
82866990447203f57400372e62d765b5.gif
Psychology 11: 287-307.

Norman, D. A. 1981. Categorization of action

82866990447203f57400372e62d765b5.gif
slips. Psychological Review 88: 1-15.

Norman, D.A. 1983. Design rules based on ana-

lyses of human error. Communications of the
82866990447203f57400372e62d765b5.gif
Association of Computing Machinery 4:
254-258.

Norman, D.A. 1986. Cognitive engineering. In

D. A. Norman and S. Draper (eds.), User
82866990447203f57400372e62d765b5.gif
Centered System Design: New Perspectives in
Human-Computer Interaction. Erlbaum.
 1517/1540

Norman, D. A. 1987. The Psychology of Every-

82866990447203f57400372e62d765b5.gif
day Things. Basic Books.

Norman, D. A. 1991. Approaches to the study of

intelligence. Artificial Intelligence 47 (1-3):
82866990447203f57400372e62d765b5.gif
327-346.

Norman, D. A. 1993. Things That Make Us

82866990447203f57400372e62d765b5.gif
Smart. Addison-Wesley.

Ochs, E., Jacoby, S., and Gonzalez, P. 1994. In-

terpretive journeys: How physicists talk and
82866990447203f57400372e62d765b5.gif
travel through graphic space. Configurations 2
(1): 151-171.

Perrow, C. 1984. Normal Accidents. Basic

82866990447203f57400372e62d765b5.gif
Books.

Pollatsek, A., and Rayner, K. 1989. Reading. In

M. Posner (ed.), Foundations of Cognitive
82866990447203f57400372e62d765b5.gif
Science. MIT Press.
 1518/1540

Pylyshyn, Z. W. 1989. Computing in cognitive

science. In M. Posner (ed.), Foundations of
82866990447203f57400372e62d765b5.gif
Cognitive Science. MIT Press.

Reisberg, D. 1987. External representations and

the advantages of externalizing one's thoughts.
82866990447203f57400372e62d765b5.gif
In program of ninth annual conference, Cognit-
ive Science Society.

Riesenberg, S. 1972. The organization of navig-

ational knowledge on Puluwat. Journal of the
82866990447203f57400372e62d765b5.gif
Polynesian Society 1 (81): 19-55.

Roberts, J. 1964. The self-management of cul-

tures. In W. Goodenough (ed.), Explorations in
82866990447203f57400372e62d765b5.gif
Cultural Anthropology: Essays in Honor of Ge-
orge Peter Murdock. McGraw-Hill.
 Page 378

Romney, A. K., Weller, S. C., and Batchelder,

W. H. 1986. Culture as consensus: A theory of
82866990447203f57400372e62d765b5.gif
culture and informant accuracy. American An-
thropologist 88 (2): 313-338.

Rumelhart, D. E., Smolensky, P., McClelland, J.

L., and Hinton, G. E. 1986. Schemata and se-
quential thought processes in PDP models. In J.
L. McClelland, D. E. Rumelhart, and the PDP
82866990447203f57400372e62d765b5.gif
Research Group (eds.), Parallel Distributed
Processing: Explorations in the Microstructure
of Cognition, volume 2. MIT Press.

Sahlins, M. 1976. Culture and Practical

82866990447203f57400372e62d765b5.gif
Reason. University of Chicago Press.

Sarfert, E. 1911. Zur kenntnis der

schiffahrtskunde der Karoliner. Korresponden-
zblatt der Deutschen Gesellschaft für Anthropo-
82866990447203f57400372e62d765b5.gif
logie, Ethnologie und Urgeschichte 42:
131-136.
 1520/1540

Schück, A. 1882. Die astronomischen, geo-

graphischen und nautischen kennitnisse der Der-
wohoner der Karolinen und Marshall Inseln im
82866990447203f57400372e62d765b5.gif
Westlichen Grossen Ozean. Aus Allen Weltthei-
len 13 : 51-57, 242-243.

Schwartz, T. 1978. The size and shape of a cul-

ture. In F. Barth (ed.), Scale and Social Organ-
82866990447203f57400372e62d765b5.gif
ization. Universitetsforlaget (Oslo).

Scribner, S. 1984. Studying working intelli-

gence. In B. Rogoff and J. Lave (eds.), Everyday
82866990447203f57400372e62d765b5.gif
Cognition: Its Development in Social Context.
Harvard University Press.

Searle, J. R. 1990. Is the brain's mind a com-

puter program? Scientific American 262 (1):
82866990447203f57400372e62d765b5.gif
26-31.

Segal, L. 1990. Effects of cockpit design on

crew communication. In Editor, Contemporary
82866990447203f57400372e62d765b5.gif
Ergonomics. Taylor and Francis.
 1521/1540

Simon, H. A. 1981. The Sciences of the Artifi-

82866990447203f57400372e62d765b5.gif
cial (second edition). MIT Press.

Simon, H. A., and Kaplan, C. A. 1989. Founda-

tions of cognitive science. In M. Posner (ed.),
82866990447203f57400372e62d765b5.gif
Foundations of Cognitive Science. MIT Press.

Suchman, L. 1987. Plans and Situated Actions:

The Problem of Human-Machine Communica-
82866990447203f57400372e62d765b5.gif
tion. Cambridge University Press.

Swade, D. D. 1993. Redeeming Charles

Babbage's mechanical computer. Scientific
82866990447203f57400372e62d765b5.gif
American 268 (2), 86-91.

Taylor, E. G. R. 1971. The Haven Finding Art.

82866990447203f57400372e62d765b5.gif
American Elsevier.

Taylor, F. J. 1984. Residue arithmetic: a tutorial

82866990447203f57400372e62d765b5.gif
with examples. IEEE Computer.
 1522/1540

Theureau, J. 1990. Introduction a l'etude du

cours d'action: Un programme de recherche en
ergonomie st anthropologie cognitive. Labor-
82866990447203f57400372e62d765b5.gif
atoire Communication et Travail, Université
Paris-Nord.

Tweney, R., Doherty, M., and Mynatt C. 1981.

On Scientific Thinking. Columbia University
82866990447203f57400372e62d765b5.gif
Press.

Tylor, E. B. 1871. Primitive Culture. Murray.

82866990447203f57400372e62d765b5.gif

Vera, J., and Simon, H. 1993. Situated action: A

symbolic interpretation. Cognitive Science 17
82866990447203f57400372e62d765b5.gif
(1): 7-48.

Vygotsky, L. S. 1978. Mind in Society: The

Development of Higher Psychological Pro-
82866990447203f57400372e62d765b5.gif
cesses. Harvard University Press.
 1523/1540

Wason, P. 1968. Reasoning about a rule.

Quarterly Journal of Experimental Psychology
82866990447203f57400372e62d765b5.gif
20: 273-281.

Wason, P., and Johnson-Laird, P. 1972. Psycho-

logy of Reasoning: Structure and Content.
82866990447203f57400372e62d765b5.gif
Batsford.

Waters, D. 1976. Science and the Techniques of

Navigation in the Renaissance. Maritime Mono-
82866990447203f57400372e62d765b5.gif
graphs and Reports, National Maritime Mu-
seum, London.

Weick, K. E. 1979. Social Psychology of

82866990447203f57400372e62d765b5.gif
Organizing. Addison-Wesley.

Wertsch, James, V. 1985. Vygotsky and the So-

cial Formation of Mind. Harvard University
82866990447203f57400372e62d765b5.gif
Press.
 1524/1540

Wickens, C., and Flach, J. 1988. Information

processing. In E. Wiener and D. Nagel (eds.),
82866990447203f57400372e62d765b5.gif
Human Factors in Aviation. Academic Press.

Zhang, J. 1992. Distributed Representation: The

Interaction between Internal and External In-
formation. Technical report 9201, Department
82866990447203f57400372e62d765b5.gif
of Cognitive Science, University of California,
San Diego.
Index

Adaptation, 354, 373

82866990447203f57400372e62d765b5.gif

in organizations, 219, 223-224, 227,

82866990447203f57400372e62d765b5.gif 317, 321-3

Alidade, 30, 45, 119-124, 136-137, 142, 157-158,1

82866990447203f57400372e62d765b5.gif
186,201,234,318, 323

(see also Navigation tools)

82866990447203f57400372e62d765b5.gif

Analog-digital conversion, 65, 93, 95, 103, 106,123

82866990447203f57400372e62d765b5.gif

Arithmetic, mental, 171-172, 222, 322, 324-327

82866990447203f57400372e62d765b5.gif

Artificial intelligence, 171, 307, 357-359, 363

82866990447203f57400372e62d765b5.gif

Astrolabe, 96-99, 102, 106-107, 113

82866990447203f57400372e62d765b5.gif
 1526/1540

as analog computer, 98
82866990447203f57400372e62d765b5.gif

Attention
82866990447203f57400372e62d765b5.gif

allocation of, 235-236, 295-297, 304-305,333

82866990447203f57400372e62d765b5.gif

and distraction, 201,235

82866990447203f57400372e62d765b5.gif

in error detection, 273-276

82866990447203f57400372e62d765b5.gif

to goals, 202-203
82866990447203f57400372e62d765b5.gif

Attribution, 132, 169,173,355-356,369

82866990447203f57400372e62d765b5.gif

Bateson, G., 291-293

82866990447203f57400372e62d765b5.gif

Beam bearings, 156, 204-219, 270-271, 295

82866990447203f57400372e62d765b5.gif

Bottom-up processes, 190, 234, 236

82866990447203f57400372e62d765b5.gif

C
 1527/1540

Chandrasekaran, B., 226-228

82866990447203f57400372e62d765b5.gif

Charts
82866990447203f57400372e62d765b5.gif

as analog computers, 61, 73

82866990447203f57400372e62d765b5.gif

computational properties of, 55, 61-65, 73,107,

82866990447203f57400372e62d765b5.gif
146, 171,173

construction of, 36, 64, 107-112, 165,

82866990447203f57400372e62d765b5.gif 168,173

projections of, 63-64, 125, 143, 146,

82866990447203f57400372e62d765b5.gif 173

storage of, 19, 36, 160

82866990447203f57400372e62d765b5.gif

use of, 13, 29-30, 36-37, 44-48, 65, 119, 125-12

82866990447203f57400372e62d765b5.gif
143-145, 159,161-163, 166,181,270

and world, 12-14, 62, 110, 136, 229-

82866990447203f57400372e62d765b5.gif 230,285

''Chinese room" thought experiment, 361-362,364

82866990447203f57400372e62d765b5.gif

Chip log, 103-107, 112, 152

82866990447203f57400372e62d765b5.gif
 1528/1540

Cognition
82866990447203f57400372e62d765b5.gif

disembodied, 132, 362, 365, 367-368,

82866990447203f57400372e62d765b5.gif 370

group, 242-261
82866990447203f57400372e62d765b5.gif

unit of analysis for, 49, 118, 123,

82866990447203f57400372e62d765b5.gif
128-129,142,155,157-159,175,280,287-293,321

Cognitive amplifiers, 153-155,170

82866990447203f57400372e62d765b5.gif

Cognitive architecture, 357-358,364-365, 369-370

82866990447203f57400372e62d765b5.gif

Cognitive ecology, 107, 112-114, 152, 168, 346

82866990447203f57400372e62d765b5.gif

tradeoffs in, 261

82866990447203f57400372e62d765b5.gif

Cognitive economy, 92, 295, 325

82866990447203f57400372e62d765b5.gif

Cognitive ethnography, 371

82866990447203f57400372e62d765b5.gif

Cognitive labor, division of, 134, 175-178, 180, 182

185,201,219,224,228-230, 239-240,256-262,267-27
82866990447203f57400372e62d765b5.gif
344-345,347
 1529/1540

Compass (see also Gyrocompass)

82866990447203f57400372e62d765b5.gif

magnetic, 39, 58-59, 70, 93, 108, 319

82866990447203f57400372e62d765b5.gif

sidereal, 69-70, 73-74, 92

82866990447203f57400372e62d765b5.gif

Computational constraints, 52-55

82866990447203f57400372e62d765b5.gif

combinations of, 52-58,118-119

82866990447203f57400372e62d765b5.gif

in Micronesian navigation, 92-93

82866990447203f57400372e62d765b5.gif

physical embodiment of, 95-102, 107-110,

82866990447203f57400372e62d765b5.gif
123,130-131,150-151,154, 165

social embodiment of, 202-204,

82866990447203f57400372e62d765b5.gif
207-209,282-283,312,316,327,332,347

Computer simulation
82866990447203f57400372e62d765b5.gif

of group cognition, 242-261

82866990447203f57400372e62d765b5.gif

limits of, 234,261

82866990447203f57400372e62d765b5.gif
 1530/1540

Conceptual dependencies, 305-307

82866990447203f57400372e62d765b5.gif

Confirmation bias, 239-240, 247, 252-256, 260-262

82866990447203f57400372e62d765b5.gif

Constellations, 67-69, 93, 99

82866990447203f57400372e62d765b5.gif

Constraint satisfaction
82866990447203f57400372e62d765b5.gif

as computational process, 55-56, 125,

82866990447203f57400372e62d765b5.gif 144,300,3

and interpretation formation, 240

82866990447203f57400372e62d765b5.gif

and networks, 243-248

82866990447203f57400372e62d765b5.gif

Context, 115, 330, 367, 372

82866990447203f57400372e62d765b5.gif

of arithmetic practice, 155

82866990447203f57400372e62d765b5.gif

Coordination
82866990447203f57400372e62d765b5.gif

of action, 175, 183, 186-189, 199-220,

82866990447203f57400372e62d765b5.gif 228,235
 1531/1540

of media, 6, 98, 117-118, 120-125,

131-145,153,155-159,167,170,172,186-189, 19
82866990447203f57400372e62d765b5.gif
280-281,289-290, 295-316,369,372-374
 Page 380

Coordination
82866990447203f57400372e62d765b5.gif

by superimposition, 68, 93, 96, 100-102,

82866990447203f57400372e62d765b5.gif
110, 120, 123,126-127, 135,167,309

Counting, 138-139,314-316
82866990447203f57400372e62d765b5.gif

Cultural processes, 111, 130, 155, 165,

82866990447203f57400372e62d765b5.gif
168-169,280,353-354,372-374

Culture, definitions of, 353-354

82866990447203f57400372e62d765b5.gif

Daemons, 191
82866990447203f57400372e62d765b5.gif

Data, collection of, 21-24

82866990447203f57400372e62d765b5.gif

Dead reckoning, 56
82866990447203f57400372e62d765b5.gif

Depth-contour matching, 56-57, 92-93

82866990447203f57400372e62d765b5.gif
 1533/1540

Description, Marr's levels of, 50, 51, 119,

82866990447203f57400372e62d765b5.gif
129-131,153,170-173

Design
82866990447203f57400372e62d765b5.gif

for error, 272-274

82866990447203f57400372e62d765b5.gif

of organizations, 204-219, 317, 345-351

82866990447203f57400372e62d765b5.gif

of procedures, 321
82866990447203f57400372e62d765b5.gif

of tools, 107, 270, 290

82866990447203f57400372e62d765b5.gif

Distance-rate-time constraint, 57-58, 93,

82866990447203f57400372e62d765b5.gif
147-155, 165.

See also Computational constraints

82866990447203f57400372e62d765b5.gif

DRAI, 34
82866990447203f57400372e62d765b5.gif

Error
82866990447203f57400372e62d765b5.gif
 1534/1540

designing for, 105,272-273

82866990447203f57400372e62d765b5.gif

detection of, 29, 35, 128, 138-139, 179-182,

82866990447203f57400372e62d765b5.gif
196, 221,227, 264, 273-276, 344

learning from, 271-272,277-279

82866990447203f57400372e62d765b5.gif

recovery from, 276-277

82866990447203f57400372e62d765b5.gif

as source of change in system, 331,

82866990447203f57400372e62d765b5.gif 333

Evolution, 349
82866990447203f57400372e62d765b5.gif

cultural, 116, 152, 367, 374

82866990447203f57400372e62d765b5.gif

Expertise, reproduction of, 263, 272, 351, 373

82866990447203f57400372e62d765b5.gif

Expert system, 155

82866990447203f57400372e62d765b5.gif

Field site, access to, 6-7, 11-12, 21-26

82866990447203f57400372e62d765b5.gif
 1535/1540

Fix cycle, 26-29, 42-48, 117-118, 133-159,

165,168,178,191,195-196,199-201,222,
82866990447203f57400372e62d765b5.gif 224,
236, 293-295

Formal systems, 132, 292, 357-360,

82866990447203f57400372e62d765b5.gif
363-366,370

Frake, C., 99-102

82866990447203f57400372e62d765b5.gif

Functional system, 142,153-158, 163,165, 170,

172,189, 194, 219, 225-226,
82866990447203f57400372e62d765b5.gif
280-281,288-291,293,310, 313,315-316,
372-373

Geographic position, 59
82866990447203f57400372e62d765b5.gif

Great Circle, 63-64

82866990447203f57400372e62d765b5.gif

Groups, cognitive properties of, 123, 128-129,

82866990447203f57400372e62d765b5.gif
170, 175-178, 180, 185-191, 252-262,284
 1536/1540

in adaptation, 321-323, 332,342, 345

82866990447203f57400372e62d765b5.gif

with beam bearings, 209, 217-219

82866990447203f57400372e62d765b5.gif

with confirmation bias, 239-262

82866990447203f57400372e62d765b5.gif

linguistic determinants of, 229-232

82866990447203f57400372e62d765b5.gif

mistaken for those of individual, 51,

82866990447203f57400372e62d765b5.gif
170-173,366

in sea and anchor detail, 196, 199-203,

82866990447203f57400372e62d765b5.gif
223-228

Guyot hopping, 29, 57, 92-93.

82866990447203f57400372e62d765b5.gif

See also Depth-contour matching

82866990447203f57400372e62d765b5.gif

Gyrocompass, 2, 30, 32, 34, 39, 45, 92,

119-126, 138,142,158,197,277,281,
82866990447203f57400372e62d765b5.gif
289,318-324,338

H
 1537/1540

Hoey, 46, 124-126, 130, 142-147, 166-167,

82866990447203f57400372e62d765b5.gif
188-189,194,219,223,234-238, 324,336

Horizon of observation, 268, 274-275, 279

82866990447203f57400372e62d765b5.gif

Information, access to, 180, 197, 223, 227,

82866990447203f57400372e62d765b5.gif
249,252,255,260,273-278,325-327

Information buffers, 194-195, 228, 235, 364-365

82866990447203f57400372e62d765b5.gif

Information processing, psychology of,

82866990447203f57400372e62d765b5.gif
358-359,363

Interactions, open, 268-270

82866990447203f57400372e62d765b5.gif

Internalization, 140, 171-172,282-285,

82866990447203f57400372e62d765b5.gif
289,301-313,332,373

Interpretations, formation of, 240-242, 260-262

82866990447203f57400372e62d765b5.gif

K
 1538/1540

Knowledge, distribution of, 176-178, 218,

82866990447203f57400372e62d765b5.gif
223,249,255,262-266,277

Laboratory research, 287-290, 367, 370

82866990447203f57400372e62d765b5.gif

Learning, 289-294, 301-310, 373

82866990447203f57400372e62d765b5.gif

as adaptive reorganization, 288-290,

82866990447203f57400372e62d765b5.gif
308,310

context of, 267-271, 279-285, 312-313,

82866990447203f57400372e62d765b5.gif 336

from error, 271-272,277-279

82866990447203f57400372e62d765b5.gif

on-the-job, 263,267-272
82866990447203f57400372e62d765b5.gif

of procedures, 294-310
82866990447203f57400372e62d765b5.gif

of sequences, 292-313
82866990447203f57400372e62d765b5.gif

Linguistic determinism, 230-232

82866990447203f57400372e62d765b5.gif
 1539/1540

Logarithms, 106-107,171
82866990447203f57400372e62d765b5.gif

Meaning, negotiation of
82866990447203f57400372e62d765b5.gif

in interaction with the world, 299-301,

82866990447203f57400372e62d765b5.gif 344

in social interaction, 214, 217-219,

82866990447203f57400372e62d765b5.gif 232-239

Measurement, units of, 58-60, 94, 108-110

82866990447203f57400372e62d765b5.gif

Mediation, 280-284,290-316,330-335
82866990447203f57400372e62d765b5.gif

Memory
82866990447203f57400372e62d765b5.gif

in artifacts, 96-98, 105, 125, 134, 142,

82866990447203f57400372e62d765b5.gif
156,221,236,325-328

for bearings, 123,196

82866990447203f57400372e62d765b5.gif
@Created by PDF to ePub