Natural Language Processing (NLP)
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 1
UNIT – IV
Semantic Parsing II:
Predicate-Argument Structure,
Meaning Representation Systems
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 2
Semantic Parsing II:
Definition:
Semantic Parser–II also called Deep Semantic Parsing is the process of
converting a sentence into a formal representation of its meaning, such as a
logical form or semantic graph.
It goes beyond just labelling roles like Subject, Object and tries to represent the
complete meaning of the sentence in a machine-understandable form.
To make the computer understand the actual meaning of natural language,
so it can reason, answer questions, or perform logical operations.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 3
Example
Sentence:
“John eats an apple.”
Semantic Representation:
eat(John, apple)
Or
∃e [eat(e) ∧ agent(e, John) ∧ theme(e, apple)]
This captures the meaning that “John” is the one performing the action “eat,” and
“apple” is the thing being eaten.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 4
Difference Between Semantic Parser–I and II
Feature Semantic Parser–I (Shallow) Semantic Parser–II (Deep)
Identifies who did what to whom (semantic Builds a complete meaning
Focus
roles) representation
Also called Shallow Semantic Parsing / SRL Deep Semantic Parsing
Input Parsed syntactic structure Parsed syntax + semantics
Output Predicate–argument structure (roles) Logical form / semantic graph
Representation Roles like Agent, Theme, Instrument Logical relations like eat(John, apple)
Example [ARG0: John] [V: eats] [ARG1: an apple] eat(John, apple)
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 5
Predicate-Argument Structure:
Predicate–Argument Structure also called Shallow Semantic Parsing or Semantic
Role Labelling – SRL is the process of finding how the parts of a sentence
arguments are semantically related to the main verb predicate.
It answers the question: “Who did what to whom, when, where, and how?”
Key Points:
In linguistics, a predicate usually means the main verb in a sentence.
The predicate takes arguments — these are the participants or things involved in the
action.
SRL (Semantic Role Labelling) tells us how each argument is related to the
predicate.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 6
Main Parts
1.Predicate:
1.The main verb or action in a sentence.
2.Example: eat, load, hit, run
2.Arguments:
1.The people or things taking part in the action.
2.Each argument has a semantic role (like doer, object, place, time).
3.Adjuncts:
1.Extra information such as time, place, or manner.
2.Example: yesterday, at the park, with a stick
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 7
Example of Different Forms (Same Meaning):
All these sentences mean the same thing:
1.“Yesterday, Kristina hit Scott with a baseball.”
2.“Scott was hit by Kristina yesterday with a baseball.”
3.“With a baseball, Kristina hit Scott yesterday.”
4.“Kristina hit Scott with a baseball yesterday.”
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 8
Predicate: hit
Arguments:
Kristina → Agent (doer)
Scott → Patient (receiver)
baseball → Instrument
yesterday → Time
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 9
FrameNet:
A semantic database for English that provides frame-specific annotation of
predicates.
Uses tagged sentences extracted from the British National Corpus (BNC).
Helps in understanding the meaning of words in context.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 10
Components of FrameNet
Component Description Example
Conceptual structures
Commerce_buy frame
Frames representing situations/events
(buying something)
and their participants
Words or phrases that evoke a
Lexical Units (LUs) frame (verbs, nouns, buy, purchase, sell
adjectives, adverbs)
Frame Elements Roles or participants in the Buyer, Goods, Seller,
(FEs) frame (agent, object, etc.) Money
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 11
Example: Commerce_buy Frame
Sentence: “John bought a car from Mary for $500.”
Element Role in Frame Example
Buyer Person doing the buying John
Goods Item being bought car
Seller Person selling Mary
Money Amount paid $500
Predicate / LU Word evoking frame bought
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 12
Meaning Representation Systems (MRS)
Meaning Representation Systems are essential in computational linguistics
and natural language processing (NLP) for convert natural language into a
structured, formal representation of meaning.
Aim to capture the semantic content of sentences in a way that is precise and
computationally usable.
Help bridge the gap between the ambiguity of natural language and the need
for clear representation in computational tasks
Purpose
•Represent “who did what to whom, when, where, and how”.
•Enable reasoning, question answering, and AI applications.
•Make semantic understanding computationally tractable.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 13
Components of Meaning Representation Systems
Component Description
Functions or relations that describe actions, events, states, or
Predicates properties. They express the core meaning of a sentence and connect
arguments.
Entities that participate in the action or relation described by the
Arguments
predicate. Can be nouns, noun phrases, or other constructs.
Logical Operators such as AND, OR, NOT, IMPLIES used to build complex
Connectives representations from simpler ones.
Words like some, all, none, every to specify quantities or conditions
Quantifiers
within the representation.
Placeholders representing entities, actions, or properties, often used
Variables
with quantifiers to express general statements.
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 14
Example:
Sentence: “Every student reads a book.”
Predicate: read
Arguments: student (reader), book (object)
Logical Connective: implicit in sentence structure
Quantifier: every (for student), a (for book)
Variable representation: ∀x (student(x) → ∃y (book(y) ∧ read(x, y)))
PREPARED BY GAJJELA.KUSHUBU@GMAIL.COM 15
