UNIVERSITY OF CALIFORNIA Santa Barbara Grammatical Relations in Tobelo A Thesis submitted in partial satisfaction of the requirements for the degree of Master of Arts in Linguistics by Gary Holton Committee in charge: Professor Susanna Cumming, Chairperson Professor Marianne Mithun Professor Sandy Thompson June 1997 ii The thesis of Gary Holton is approved ______________________________ ______________________________ ______________________________ Committee Chairperson June 1997 iii 1 January 1997 Copyright by Gary Holton 1997 iv Acknowledgments This study would not have been possible without the warm hospitality and support of many Tobelo people in Halmahera. These include Yohanis Labi, Matias Oga, Paltiel Oga and family, Bapak Guru Kukihi and family, Domingus Diba, Frans Diba, Paulina Tindagi, Tobias Tjileni, and Jason Moloku. This list unfortunately omits scores of persons in Tobelo, Gura Belakang, Meti and LabiLabi who generously shared their thoughts and language with me during my stay in their villages. Initial bibliographic work at the University of Hawaii Hamilton Library was supported by the University of Hawaii Center for Southeast Asian Studies. Field work investigating the Tobelo language was conducted in Halmahera for four months in 1995 under the auspices of the University of Hawaii — Universitas Pattimura Sago Research Project, with support from the Henry Luce Foundation (grant #P95280F348B164; Principal Investigator: Byon Griffin). Outside Halmahera, many people have offered advice since I began investigating the Tobelo language. Foremost among these are my committee members in Santa Barbara — Susanna Cumming, Marianne Mithun, and Sandy Thompson — all of whom have taught me the value of letting Tobelo speak for itself rather than accepting artificially constructed paradigms. I also owe a great debt of gratitude to Jim Collins, whose unceasing confidence and enthusiasm encouraged me to return to Maluku and to eventually write this thesis. And to Talmy Givón, my first linguistics teacher, I thank you for the challenge. Thanks also to Colette Grinevald for giving me the tools and inspiration with which to meet that challenge. A special note of thanks goes to my wife, Wendy Camber, who constantly reminds me that Tobelo is more than just a collection of semantico-syntactic macro-roles, lexical NPs and non-volitional arguments. v Abstract Grammatical Relations in Tobelo by Gary Holton The verbal cross-reference paradigm in the Papuan language Tobelo is here shown to pattern as an active-stative system of grammatical relations, based on an aspectual distinction between dynamic and time-stable predicates. Thus, grammatical relations in Tobelo behave quite differently than those in other split intransitive systems which base verbal cross referencing on semantic properties of the participants. Although Durie (1988, 1994) has found evidence for split intransitive patterns in Acehnese (Austronesian) discourse, evidence from Tobelo narrative discourse studies presented here provides no indication of an activestative preferred argument structure in Tobelo. The central argument of this thesis is that the observed difference between Tobelo and Acehnese discourse patterns results directly from the lexical semantic structure of the predicates used to introduce new participants in the two languages. Both languages tend to introduce new participants as non-volitional arguments of intransitive verbs. However, due to the differing semantic bases of split intransitivity in the two languages, these non-volitional arguments receive different grammatical coding in each of the two languages. Thus, lexical semantics is responsible both for the apparent correlation between Acehnese grammatical relations and discourse patterns and for the lack of such correlation in Tobelo. This study suggests that lexical semantics plays an important role in mediating preferred argument structure in languages with split intransitive systems of grammatical relations. vi Table of Contents Acknowledgments v Abstract vi 1. Introduction. 1 2. Background. 5 2.1. Approaches to Tobelo verbal morphology. 2.2. Approaches to split intransitivity. 3. Verbal morphology. 5 11 15 3.1. Core versus oblique arguments. 19 3.2. Verb classes. 22 3.3. Animacy and the i- marker. 28 3.4. Stative verbs and the i- marker. 34 4. Grammatical relations. 38 4.1. Split intransitivity. 40 4.2. The nature of the split. 41 4.3. Comparative evidence. 52 4.4. Event-sensitivity versus participant-sensitivity. 53 5. Preferred argument structure. 54 5.1. The one full NP constraint. 57 5.2. Participant introduction strategies. 64 6. Outlook. 72 Bibliography 76 Appendix 80 vii 1 1. Introduction This paper proposes a semantic basis for the verbal cross-reference system in the Papuan language Tobelo, which has been previously described in accusative terms (cf. Van der Veen 1915; Hueting 1936). Furthermore, Tobelo discourse patterns — or more specifically “preferred argument structure” in Du Bois’ (1985) terminology — are shown to be significantly different from those in Acehnese (Austronesian), another language which bases grammatical relations on semantic distinctions (cf. Durie 1985, 1987). It is proposed here that this difference in discourse patterns between Tobelo and Acehnese directly reflects the lexical semantic structure of participant introduction strategies in each of the two languages. Both languages tend to introduce New participants as the non-volitional argument of an intransitive dynamic predicate. Grammatical relations in Tobelo are sensitive to a semantic property of the predicate — namely lexical aspect — rather than the nominal arguments of the predicate. Hence, Tobelo codes the nominal argument of such a predicate as a grammatical agent, while Acehnese codes it as a grammatical patient. As Durie (1988) has shown for Acehnese, languages which base case marking on parameters of agency may show split intransitive, or more precisely agent-patient, patterns in discourse measurements such as the distribution of full NP realizations of referents. However, evidence from Tobelo discourse presented here suggests that this is not necessarily the case for active-stative languages. 2 Languages such as Tobelo, which base grammatical relations on semantic distinctions relating to the predicate, do not exhibit split intransitive patterns in the discourse structure of nominal arguments. Durie has explained the occurrence of agent-patient patterns in Acehnese discourse in terms of “pragmatic linking”, or “the coherence in discourse of surface grammatical case categories” (1994:496). That is, Acehnese grammatical relations which are encoded in verbal crossreference morphology are mirrored in discourse patterns. Why should it be that similar pragmatic linking does not occur in Tobelo? The answer to this question can be found in a close examination of the types of predicates used for referent introduction in Tobelo and Acehnese discourse. As many cross-linguistic studies have already shown, New referents are more likely to occur as full lexical NPs. In a split intransitive system, the grammatical coding of these NPs is determined by the lexical semantics of the verb complex. As we shall see, subtle but important differences in the semantic bases of Tobelo and Acehnese grammatical relations are reflected in the lexical semantics of predicates used to introduce New participants. Thus, the apparent pragmatic linking in Acehnese actually derives from a semantic linking between grammatical relations and discourse strategies. In order to pursue this explanation for the difference between Tobelo and Acehnese discourse patterns, we must first dispel a long-standing assumption regarding the nature of grammatical relations. Traditional approaches to grammatical relations reflect a pre theoretical desire to view grammatical relations 3 as properties of entire languages rather than properties of particular constructions or grammatical systems within a language. For example, Comrie states that “in order to say that a given grammatical relation exists in a given language … a number of logically independent criteria must be established that serve to identify the grammatical relation in question as being syntactically significant in the language in question” (1989:66). This prescription has led many to downplay the significance of morphological patterns which are not also reflected in syntax or discourse. One famous example is the distinction between morphological and syntactic ergativity (Andrews 1985). There are many languages, such as the Australian language Warlpiri (Pama-Nyungan), which exhibit ergative casemarking but do not pass the syntactic litmus test for ergativity, in that syntactic patterns such as equi-NP deletion do not treat the single argument of intransitive predicates the same as the more patient-like argument of transitive predicates (Anderson 1976). Thus many authors would reserve the term ergative for the few languages, such as Dyirbal (Pama-Nyungan), which do exhibit ergative patterns in syntax (Dixon 1979). More recently this approach has been extended to discourse studies as well. For example, Acehnese marks agent-patient distinctions on pronouns; and Durie’s (1988) study of Acehnese discourse has found evidence for agent-patient patterns 4 in the distribution of full NP tokens.1 Durie notes that “what seems unusual about Acehnese is that it is so consistently active in many respects” (1987:367). The implication is that active-stative or agent-patient patterns which are reflected only in morphology do not represent grammatical relations, since they are not evidenced in syntax or discourse. In contrast, I view a grammatical relation as any grammaticized relationship between a predicate and its arguments. Such a relationship may be reflected in many parts of the grammar; but it need not be. As noted by Mithun and Chafe (1996), different relationships may be evident in different parts of the grammar, and there may indeed be principled reasons governing this distribution. In Tobelo, the active-stative relation is apparent only in verbal morphology and not in syntactic or discourse patterns. But, as I will argue below, such a state of affairs is to be expected in languages which base crossreference marking on distinctions in lexical aspect. Aspect is a property of the entire verb complex, not simply of the participants. Referent tracking in discourse reflects properties of the participants, so we should not expect discourse properties such as the distribution of full NPs to reflect active-stative distinctions in lexical aspect. In reserving the term grammatical relation to refer to patterns which 1 A note on terminology is in order here. Durie uses the term “active” in a more general sense to refer to all systems of grammatical relations based on semantic distinctions. Thus, Durie’s term “active” is not to be taken as synonymous with the term “active-stative” used here. In the terms used here, Acehnese can be said to exhibit an agent-patient system based on the semantic parameter of volition. 5 behave consistently across all domains of a language, we miss natural and motivated differences in the patterning of morphology, syntax and discourse. The remainder of this paper is organized as follows. The following section situates the problem of Tobelo verbal morphology within the context of areal linguistics and previous research. Then §3 presents an overview of Tobelo verbal morphology in particular. In §4 I describe the morphological instantiations of Tobelo grammatical relations, arguing for an active-stative interpretation. Finally, in §5 I provide a brief account of the Tobelo discourse patterns and preferred argument structure, finding no evidence for an active-stative pattern in the distribution of full NPs in Tobelo discourse. 2. Background 2.1. Approaches to Tobelo verbal morphology Tobelo is a Papuan language spoken by approximately 15,000 persons on the islands of Halmahera and Morotai in the eastern Indonesian province of Maluku.2 The verbal cross-referencing systems of Tobelo and several closely 2Almost all of the fifty or so languages spoken on Halmahera and the province of Maluku are clearly members of the Austronesian family. The historical classification of Tobelo and three related languages is less clear. Recent quantitative evidence and a somewhat older linguistic tradition maintain a historical relation between Tobelo and the West Papuan languages (e.g., 6 related languages were first rigorously described in van der Veen's (1915) dissertation, and were cited by van der Veen as the primary distinguishing factor delimiting the North Halmahera (NH) family (“Noord-Halmahera'se Taalgroep”).3 Since that time, descriptive grammars of NH languages have continued to focus on the relatively rich verbal cross-referencing morphology in these languages. Perhaps because of the status of Tobelo and the NH languages as a typological anomaly in a sea of Austronesian regularity, previous researchers — mostly area specialists — have tended to view NH verbal cross-referencing as odd, arbitrary and irregular. This problem arises due to the existence of two distinct paradigms of pronominal verbal cross-reference markers in Tobelo (and most NH languages). Maibrat) spoken on the Bird's Head of New Guinea. However, the data remain inconclusive. As noted by Foley, the more conservative term “non-Austronesian” is in many ways preferable to the more speculative term “Papuan” (1986:3). Furthermore, this negative classification captures the sociolinguistically significant fact that speakers of Tobelo interact in a basically Austronesian world through a language which is genetically unrelated to the Austronesian family. Still, for the purposes of consistency with the existing literature I retain the term Papuan here. 3Tobelo and several closely related languages, including Galela, Tabaru, Modole and Loloda, have recently been classified by Voorhoeve (1988) as dialects of the so-called Northeast Halmaheran language. This classification is based on typological similarities as well as a lexicostatistical comparison of one hundred items of basic vocabulary, with an eighty percent cognate level dividing dialect from language. All of the Northeast Halmaheran languages exhibit cognate percentages just above the eighty percent threshold. For the present, I defer debate on dialect status and continue to refer to Tobelo as a "language". This is consistent with the attitudes of Tobelo speakers, who have not recognized the term Northeast Halmaheran. The other Northeast Halmaheran dialects/languages, though typologically similar to Tobelo, tend toward less morphological complexity. 7 The choice between the two paradigms evidently depends on the particular grammatical relationship between the predicate and its arguments. In syntactically transitive clauses — i.e., those which explicitly cross-reference two distinct nominal arguments — one paradigm cross-references the semantic agent, or “Actor”, argument, while another paradigm cross-references the semantic patient, or “Undergoer” argument. I will refer to these two paradigms as grammatical agent (AGT) and grammatical patient (PAT), respectively. Traditionally, these two verbal cross-reference paradigms have been referred to as subject and object markers (cf. van der Veen 1915, Hueting 1936). A sample of the two paradigms for singular arguments is listed in (1).4 (1) person 1 2 3MASC 3FEM 3 AGT PAT tonowomoi- hiniwimia- Several straightforward morphophonemic processes operate, reflecting phonotactic constraints maintaining CV syllable structure and resulting in the elision of the cross-referencing morpheme vowel before vowel-initial verb stems (cf. Hueting 1936:356-60). Less straightforward morphological processes lead to the formation 4 The complete set of cross-referencing prefixes is listed in the appendix. 8 of portmanteau two-argument cross-referencing morphemes.5 Both AGT and PAT markers can occur on a single verb stem, but a PAT marker cannot occur without an AGT marker. As many authors have noted, the same paradigm which cross- references the Actor argument of a transitive clause also cross-references the single argument of many intransitive clauses. As evidence, Hueting cites the following examples in his grammar of Tobelo (1936:350).6 (2) to-tagi ‘I go’ (3) to-ni-gohara ‘I hit you’ Based solely on data of the type shown in (2) and (3), there does indeed appear to be a grammaticized relationship between a Tobelo predicate and its arguments, which is reflected morphologically in the choice of pronominal cross-reference markers. Intuitively, the grammatical relation which these cross-reference markers 5 Howard Sheldon describes some of these processes in Galela (1991:173 note 10). However, analogous morphophonemic processes in Tobelo appear to be much less regular. 6 The transcription here generally follows the practical orthography of Indonesian, which differs somewhat from the phonetic transcription employed in Hueting (1908a). Velar and palatal nasals are represented by digraphs <ng> and <ny>, and Hueting’s lambda is retained for the palatal lateral. This yields twenty consonants consisting of a voiceless and voiced series of stops, <p> <t> <k> and <b> <d> <g>; voiceless and voiced palatal affricates /c&/ and /j&/ (orthographic <c> and <j>); voiceless fricatives <f> <s> and <h>; a nasal series <m> <n> <ny> <ng>; alveolar and palatal laterals <l> and <¥>; velar and palatal glides <w> and <y>; and the alveolar flap <r>. Though Hueting’s transcription recognizes fourteen phonetic vowels, Tobelo vowels are here written phonemically using the standard orthography of a five vowel system: <a> <e> <i> <o> <u>. 9 encode appears to be that of “subject”. Hueting does not pursue the issue — and for good reason. It turns out that there exists a large open class of intransitive verbs which cross-reference their single nominal arguments via the PAT paradigm. Hueting cites the example in (4) and refers to this class of verbs as stative (“toestandswoorden”) (1936:352).7 (4) i-hi-daluku ‘I am drunk’ The fact that the actor-topic in (4) is cross-referenced via a different paradigm than that in (2) casts doubt on the existence of a grammatical relation of Subject encoded by the cross-reference markers. Subject may indeed play a role in other parts of the grammar, but, as I will argue below, it is not fully reified in the crossreference morphology. Up to the present time all works on NH grammar have continued to refer to the AGT and PAT cross-reference markers as Subject and Object markers, respectively. However, the terms are usually presented without justification and are usually intended simply as synonyms for the corresponding semantic roles of agent and patient (or Actor and Undergoer). As Howard Sheldon remarks, “the subject prefix of a transitive verb always refers to the actor; … the object prefix of 7 The role of the morpheme i- is discussed in §3.4. 10 a transitive verb always refers to the undergoer of the clause” (1991:164). Unfortunately, the alignment between cross-reference markers and semantic roles breaks down when one considers intransitive verbs, for there exist Tobelo intransitive verbs with semantic patient (or Undergoer) arguments which are crossreferenced via the “Subject” paradigm (e.g., honenge ‘die’), as well as Tobelo intransitive verbs with semantic agent (or Actor) arguments which are crossreferenced via the “Object” paradigm (e.g., magawe ‘be diligent’). Traditional analyses of Tobelo grammar simply split the class of intransitive verbs into two subclasses based on which cross-reference paradigm is present, tacitly or even explicitly denying a semantic basic for the subcategorization. Yet, many Tobelo intransitive verbs can occur with either paradigm of cross-reference markers depending on the context. Thus, an analysis based on such an arbitrary morphological split must posit the existence of two distinct — though homophonous — lexical forms for many intransitive verbs, one for active verbs and one for stative verbs. In what follows I present an alternate analysis which suggests that the grammatical relations reflected in Tobelo verbal crossreference markers are in fact directly determined by semantic criteria, most notably lexical aspect. Grammatical relations in Tobelo are not sensitive to the semantic roles of nominal arguments within the clause; rather, the grammatical relations encoded by Tobelo verbal cross-reference markers reflect the semantic properties of the predicate, distinguishing primarily between events and states. Thus, I argue below that the split between the two verbal cross-reference 11 paradigms patterns as an active-stative system. Accordingly, the first part of this paper can be viewed as a preliminary attempt to place Tobelo within a typological framework which includes many languages with semantically based systems of grammatical relations, such as the Austronesian language Acehnese and the American languages Guarani, Lakhota and Choctaw. 2.2. Approaches to split intransitivity The existence of a third, semantically based system of case marking or grammatical relations — often contrasted with ergative and accusative systems — has been established for quite some time. Indeed, Mithun remarks that Montoya’s grammar of Classical Guarani, an active-stative language like Tobelo, is over three hundred years old (1991:510). However, as Merlan (1985) points out, there are many more languages such as Tlingit (Na Dene), for which specious initial grammatical descriptions in terms of traditional typologies have prevented a coherent understanding of this third type of system of grammatical relations. One of my purposes here is to provide a description of Tobelo grammatical relations which is does not compartmentalize Tobelo within the confines of traditional typologies. As more and more languages are re-examined and found to have semantically based systems of grammatical relations, the internal structure of this third typological system of grammatical relations becomes more apparent. As noted by Klimov (1974), Van Valin (1990), Mithun (1991) and Palmer (1995), inter alia, semantically based systems have been described by a variety of 12 terms, including split-S, active, active-stative, active/agentive, agentive, agentpatient, and split intransitive. The “split” terminology characterizes these systems syntactically, based on the observation that these systems do not treat the single argument of an intransitive predicate (the S macro-role) consistently with respect to morphological coding. But as Durie (1987) has argued convincingly for Acehnese, there is a priori no S category in these languages, so it does not make sense to speak of a split in the S category. Mithun and Chafe argue more generally against the a priori assignment of semantico-syntactic macro-roles, warning us “to avoid reification of relations that may have no basis in any language, and direct attention instead to the precise nature of the concepts actually grammaticized in the language” (1996:18). Certainly this is true for Tobelo and Acehnese, in which the choice of cross-reference marker depends not on arbitrary syntactic properties such as valency of the predicate, but rather on semantic properties of the participant and the situation. The precise nature of the semantic parameters on which active-stative and agent-patient systems are based can vary significantly across languages and even across different domains within a language (cf. Van Valin 1990, Mithun 1991a). Until recently, this fact has been taken as evidence against elevating semantically based systems to a typological status on par with ergative and accusative systems (see §4 below). But as Mithun points out, split intransitive systems “constitute coherent, semantically motivated grammatical systems of their own” (1991a:542). The semantic characteristics of a situation and its participants comprise a complex 13 suite of features, any subset of which may serve as a possible basis for making a semantic distinction between an event and a state, or between an Actor and an Undergoer. However, even when such systems are interpreted as distinct and independently motivated, the active/agentive category has often been treated as a typological grab bag of systems which are sensitive to semantic parameters. The link between semantic roles and grammatical relations is actually quite transparent in languages with semantically based systems of grammatical relations. Semantic agents are typically coded as grammatical agents; semantic patients are typically coded as grammatical patients. But the mapping between semantic and grammatical roles falls short of isomorphism. Ironically, this near isomorphism tends to delude the researcher into believing that a simple semantic explanation exists, and then leads the researcher to reject this solution categorically when the mapping between semantics and grammatical relations breaks down. For example, if we attempt to understand active-stative patterns with an intuitive definition of eventhood, we quickly find that verbs which appear to be events in one language turn up as states in another language. To avoid this confusion we require clear, cross-linguistic definitions of the relevant semantic parameters crucial to determining grammatical relations in these languages. Thus, terms like eventhood must be defined in terms of cross-linguistic semantic universals, rather than in terms of particular lexical items. Even if we posit a universal definition of event in terms of time-stability, we cannot say categorically that a particular verb, say ‘to faint’, is an event. One language may construe the lexical item ‘to faint’ as 14 relatively dynamic, while another may construe it as relatively time-stable. This state of affairs arises indirectly as a result of the radial nature of semantic categories such as Event. An active-stative system assigns the same grammatical coding to the single argument of entire class of intransitive verbs, some of which are prototypical dynamic events, others which might appear to be less prototypical in various ways. The apparent time-stability of the lexical item ‘to faint’ will be vastly different in a language which codes ‘walk’ and ‘die’ similarly than in a language which codes ‘be sick’ and ‘die’ together. The point to be demonstrated in this paper is that not all semantic features are equivalent for the purposes of distinguishing grammatical agents from grammatical patients. Systems which are sensitive to control, performance, affectedness or some subset of these features can all be termed agent-patient, as all of these features may reasonably be considered subfeatures of the general semantic feature of agency. The fact that Lakhota is sensitive to performance, effect, instigation and control, while Central Pomo is sensitive to control and affectedness, does not prevent us from classifying both systems together under the label agent-patient. Together with Acehnese, both Lakhota and Central Pomo are sensitive to semantic properties of the participants, irrespective of the gestalt lexical aspect of the situation. These systems differ markedly from that of Tobelo, described below. Tobelo, an active-stative language, is sensitive to semantic properties of the situation. This distinction between event-sensitivity and participant-sensitivity is recognized by Van Valin, who identifies “inherent lexical 15 aspect (Aktionsart) and agentiveness as the primary semantic parameters governing split intransitivity” (1990:251-2). Mithun makes a similar distinction between active-stative systems and agent-patient systems, noting also that semantically based case marking can be “based on a variety of semantic distinctions” and “several features may even interact within a language to determine case organization” (1991:512). In many cases the surface realizations of intransitive predicates in activestative and agent-patient languages are not significantly different. Prototypical events take Actor arguments, while prototypical states take Undergoer arguments. But as I show in §4.2 below, this alignment is not complete. Significantly, the fundamental difference between the semantic bases of participant-sensitive and situation-sensitive systems turns out to have important repercussions for preferred argument structure. Predicates used for participant introduction are typically dynamic, taking a single argument which does not control, perform, effect or instigate. Hence these predicates tend to receive different grammatical coding in active-stative systems than in agent-patient systems, as a result of their inherent lexical semantic structure. Before turning to discourse studies, I begin in the following section with a description of Tobelo verbal morphology, with some reference to other NH languages. 3. Verbal morphology Hueting (1936) provides an excellent account of Tobelo verbal 16 morphology, and the account given in this section is not intended to controvert Hueting’s description. However, it is useful for the reader to have the relevant information on verbal morphology close at hand and expressed within a framework which does not presume the existence of grammatical relations such as Subject and Object. Hueting provides a purely structural description, listing an inventory of morphemes along with lexical examples containing each given morpheme. The present study aims for a more holistic description, examining verbal morphology as it is actually used rather than as it relates to predefined grammatical categories. The Tobelo lexical category of Verb is morphologically delimited as the class of lexemes which can, and must, take prefixal pronominal cross-referencing morphemes (i.e., bound pronouns) indexing at least one and as many as two core nominal arguments. No bare stem verbs occur; that is, all verbs must overtly crossreference at least one core argument. Furthermore, cross-reference markers occur only on verbs; nouns — including nominal predicates — are marked via a distinct set of prefixes. For example, the nominal predicate in (5) is marked by the noun 17 marker o,8 (5) muna o-pendeta 3FEM:PRO NM-priest ‘She is a priest’ whereas, the verb in (6) is marked with the third person feminine pronominal prefix mo-. (6) muna mo-lio 3FEMC:PRO 3FEM:AGT-depart ‘She left’ Characteristic of Tobelo and most NH languages is the existence of two distinct morphological paradigms cross-referencing distinct arguments of a 8 The following abbreviations are used in glosses of Tobelo examples: 1, 2, 3 ABL AGT ALL ASP CAUS DIR DUR ENC FEM HEST IMPF INC LOC first, second, third person ablative grammatical agent allative aspect causative directional durative exclusive plural feminine hesitation imperfective inclusive plural locative MASC NEG NM PAT PERF PL POSS PRO RDP RECIP REP RFLX RNM SEQ masculine negation noun marker grammatical patient perfective plural possessive pronoun reduplication reciprocal repetitive reflexive relational noun marker sequential 18 transitive verb. Informally, the one paradigm usually cross-references the semantic agent argument, while the other paradigm usually cross-references the patient argument, as shown in (7). 9 (7) no-mi-hi-honenge-oka 2AGT-3FEM:PAT-CAUS-die-PERF ‘You have killed her’ (Hueting 1936:329) However, the semantic correlation is not complete, thus, I reserve the terms grammatical agent and grammatical patient to refer to these two morphological paradigms. These are glossed AGT and PAT, respectively, in (7). Aspect may be (optionally) marked on the verb via a system of seven inflectional aspectual suffixes. A clause may additionally include full nominal arguments or unbound pronouns. However, only a pronominal cross-referencing morpheme prefixed to a verb stem is required to form a grammatical utterance. Several valency-changing derivational prefixes may occur between the cross-referencing markers and the verb stem, denoting reflexive, causative and reciprocal functions, among others. Thus, the templatic structure of the Tobelo verb can be described as in (8), where AGT and PAT denote the two cross-referencing paradigms described above. 9The source of Tobelo textual examples is indicated in parentheses immediately following the English gloss, identified by either the surname of the speaker or the author of the publication in which it appears. Unless otherwise noted, the balance of the Tobelo data cited in this paper consists of elicited examples from my own field notes. 19 (8) AG T PAT RECIPROCAL CAUSATIV E ROOT ASPECT NEG REFLEXIVE 3.1. Core versus oblique arguments Traditionally, a pre theoretical distinction is made between core and oblique nominal arguments in a clause. In Tobelo, the distinction between core and oblique is morphologically reified, and as used here, the terms core and oblique refer to these morphological categories. Core arguments are those which are crossreferenced on the verb via the pronominal cross-reference system, i.e., via the AGT and PAT paradigms discussed above. Oblique arguments are never cross-referenced on the verb and are often — but not always —marked with an oblique case marker. Thus, the goal argument in (9) is marked with the directional -iha ‘landward’ (see below), while that in (10) is unmarked. (9) o-takoro butanga mi-ma-togumu, NM-hour six 1INC:AGT-RFLX-rest mi-lio mia-tau-iha. 1EXC:AGT-return 1EXC:POSS-house-DIR ‘We rested for six hours and returned to our houses.’ (Kukihi) (10) y-ato i-aman-oka, 3PL:AGT-say 3AGT-safe-PERF ho ho-lio o-kapongo. thus 1INC:AGT-return NM-village ‘They said it was safe, so we returned to the village’ (Moloku) 20 Oblique case markers never occur on core arguments. In addition to -iha, oblique case makers include the instrumental/comitive preposition de, as in (11) and (12), and the locative suffix oka, as in (13) and (14). For the purposes of this exposition I make no distinction between bound and free case markers.10 (11) de ma-kakatama n-a-¥e-ino with NM-tongs 2AGT-3PAT-roll-ALL ‘Roll it up with the tongs’ (Dirsa) (12) ngohi-o to-modeke de o-Matias 1PRO-also 1AGT-agree with NM-M. ‘I also agreed with Matias’ (Paltiel) (13) o-akere-oka to-ma-ohiki NM-water-LOC 1AGT-RFLX-bathe ‘I bathed at the river’ (14) o-baiti-oka yo-wohama-uku NM-hole-LOC 3PL:AGT-enter-DOWN ‘They went down in a hole’ (Diba) In addition, like most Maluku languages, Tobelo has a highly developed directional system based on two primary orthogonal dimensions. One dimension contrasts seaward versus landward; a second dimension contrasts up and down the coast or river (defined locally), or up and down vertically. Some examples are given in (15) though (20). 10 Historically, the free morpheme de appears to be a more recent Austronesian borrowing, cognate with the Malay di. 21 (15) o-Jakarta-oko to-oiki NM-Jakarta-SEAWARD 1AGT-go ‘I’m going to Jakarta’ (16) o-¥oku-iha to-oiki NM-mountains-LANDWARD 1AGT-go ‘I’m going to the mountains’ (17) ahi-dumule-uku to-oiki 1POSS-garden-DOWN 1AGT-go ‘I’m going down the coast to my garden’ (18) o-akere-i¥e to-oiki NM-water-UP 1AGT-go ‘I’m going up the coast to the river’ (19) to-oiki 1AGT-go ‘I’m going to America’ o-Amerika-ika NM-America-ABLATIVE (20) ahi-tau-ino wo-boa 1POSS-house-ALLATIVE 3MASC:AGT-come ‘He’s coming to my house’ Nominal arguments which are marked with directional suffixes are never crossreferenced on the verb. Thus, for our purposes, the relevant feature of directionals is their use as an oblique case marker.11 11 A comprehensive analysis of the Tobelo directional system can be found in Taylor (1984). 22 It should be noted here that the obligatory nature of Tobelo pronominal cross-reference markers has an important consequence for syntactic argumentation. As mentioned in the introductory section, there is an analytical tradition which views morphological relationships between a predicate and its arguments as somehow secondary to syntactic properties. Indeed, many languages which are morphologically ergative have been claimed to exhibit accusative syntax, in that both S and A control such operations as equi-NP deletion and “raising” in complex clauses (Anderson 1976). Further evidence for accusative syntax has been sought in the possibilities for relativization, nominalization, and imperatives. These syntactic properties can be said to reflect what Palmer has termed “syntactic relations” (1994:88). Most of these tests examine the possibilities for syntactically motivated deletion of nominal arguments; but Tobelo nominal arguments occur as bound pronouns and are never omitted, hence, such tests are inconclusive for Tobelo. Mithun (1991b) demonstrates a similar situation for the agent-patient language Cayuga (Iroquoian). Thus, evidence for a system of grammatical relations in Tobelo must derive from an understanding of the patterning of Tobelo verbal morphology. 3.2. Verb classes Many authors have noted that one of the primary functional roles of 23 grammatical relations is to discriminate between nominal arguments within a clause (cf. Comrie 1989, Andrews 1985).12 Thus, it is useful to begin our investigation of grammatical relations by considering a prototypical two-argument verb, as in (21). If grammatical relations are encoded at all, they should minimally distinguish the agent and patient arguments of a highly active, telic predicate, such as gohara ‘hit’. (21) o-Matias ngohi wo-hi-gohara NM-Matias 1PRO 3MASC:AGT-1PAT-hit 'Matias hit me’ The verb in (21) denotes a performing agent acting on an affected, non-controlling patient. The agent is cross-referenced via the AGT paradigm, while the patient is cross-referenced via the PAT paradigm. The same system of grammatical marking can be extended to verbs which take less prototypical, less affected patients, as in (22). 12 Du Bois argues against a theory of grammatical relations based on the discriminatory function, noting that “transitive clauses which actually contain an overt A and an overt O are rare” (1987:8848). This objection holds more readily for systems which are encoded via nominal casemarking. Since Tobelo grammatical relations are coded via two parallel systems of obligatory pronominal cross-reference markers, the discriminatory role of grammatical relations is relevant regardless of lexical form of nominal arguments. 24 (22) ngohi-o t-i-modeke o-Matias 1PRO 1AGT-3MASC:PAT-agree NM-Matias 'I too agreed with Matias’ While all agents and human patients must necessarily be cross-referenced on the verb, indexing of non-agent non-human arguments is governed by discourse factors relating to information flow. Briefly, only referential non-agent, patientive arguments of transitive verbs are indexed on the verb. For example, the identifiable referents pine ‘rice’ and gota ‘wood’ are cross-referenced via the thirdperson singular PAT marker -a in (23) and (24). (23) (24) o-pine t-a-ija NM-rice 1AGT-3PAT-buy 'I bought the rice’ o-gota mi-a-tobiki de mi-a-bela-belaka 1INC:AGT-3PAT-break and 1INC:AGT-3PAT-RDP-split ‘We bucked and split the wood’ (Kukihi) NM-wood In contrast, the semantic patients in non-specific constructions, such as pine ‘rice’ in (25), and negative constructions, such as hilo ‘resin’ in (26), are not crossreferenced on the verb, and hence are not considered as core arguments. (25) o-pine to-ija NM-rice 1AGT-buy 'I went rice-shopping' (26) jadi ngohi to-¥e-ua o-hilo therefore 1PRO 1AGT-get-NEG NM-resin ‘Therefore I didn’t get any resin’ (Paltiel) This non-core status of non-referential patients is perhaps best demonstrated by considering a short stretch of discourse. For example, in (27) the argument gaharu 25 ‘eagle wood’ is the semanitc patient of the verb ija ‘buy’, but it is not crossreferenced as the grammatical patient on the verb because it is not referential. Here gaharu does not refer to some particular piece of wood; rather, the visitors came to buy any eagle wood which they could find. (27) o-oraha genanga, NM-time that o= -HEST o-gaharu yo-ija, NM-eagle.wood 3PL:AGT-buy yo-uti .. o-LabiLabi-ha. 3PL:AGT-arriveNM-LabiLabi-LANDWARD ‘At that time they arrive in Labi-Labi to buy eagle wood.’ (Paltiel) Later in the same narrative, the speaker and his friends decide to go in search of eagle wood. They are not in search of a specific piece of eagle wood; rather, any eagle wood will do. So again in (28) gaharu ‘eagle wood’, the semantic patient of the verb lingiri ‘look for’, is non-referential and hence treated as a non-core argument. 26 (28) dina mi-a-iha, inland 1EXC:AGT-3PAT-go.inland o-hutu hange, NM-night three mi-lingiri .. ma-gaharu. 1EXC:AGT-look.for RNM-eagle.wood ‘We went inland for three nights to look for eagle wood.’ (Paltiel) Next we consider cross-referencing on one-argument (i.e., intransitive) verbs. The distinctive character of Tobelo grammatical relations is readily apparent within this class of verbs. Some intransitive verbs cross-reference nominal arguments via the AGT paradigm, as in (29). Verbs in this category usually refer to dynamic events. (29) o-akere-iha t-oiki NM-water-landward 1AGT-go 'I went landward to the river' Other intransitive verbs in this category include both agentive and non-agentive verbs. Yet another set of Tobelo one-argument verbs cross-reference nominal arguments via the PAT paradigm, as in (30). These verbs mostly denote time-stable states. A “dummy” third person AGT paradigm marker i- occurs before the PAT paradigm marker, reflecting a morphological constraint against the occurrence of a PAT marker without an AGT marker. Such clauses exhibit morphological similarity with transitive clauses. 27 (30) i-hi-bole 3AGT-1PAT-tired 'I’m tired' Both active intransitive verbs and stative intransitive verbs constitute large open classes of lexemes. More than three hundred stative intransitive verbs have been identified. The fact that single arguments of intransitive verbs are not uniformly coded via the AGT paradigm provides evidence against the grammatical relation of subject in Tobelo. In previous descriptions, NH verbs have been divided into three morphological classes of transitivity, based primarily on the type of crossreference marking they exhibit (cf. van der Veen 1915, Hueting 1936, van Baarda 1908, Fortgens 1928, Visser and Voorhoeve 1987, inter alia). Schematically, these descriptions delimit three structural verb classes as listed in (31). (31) AGT + PAT + VERB AGT + VERB /i/ + PAT + VERB Transitive (Active) Intransitive Stative (Instransitive) It should be noted that the term Stative (or sometimes Stative Intransitive), as used in the NH linguistic literature, is not a priori a semantic category. Rather, this category is purely formal, delineated by the presence of the morpheme i-. However, as we will see in §4, there is an implicit semantic dimension to this categorization which turns out to have significant import for Tobelo grammatical relations, reflecting a primary categorization of Tobelo predicates with respect to lexical aspect. 28 For the time being, we continue the use of the verb class labels in (31) in the formal sense intended by previous authors. The class of Stative verbs exhibits morphological and syntactic similarity with both Transitive and Intransitive verbs. The question then arises, should Stative verbs be aligned with Transitive verbs or with Intransitive verbs. Morphologically, Stative verbs behave like Transitive verbs, employing the cross-reference markers from the AGT and PAT paradigms. Syntactically, Stative verbs behave like intransitive verbs in that they are monovalent. A closer look at the role of the third person AGT paradigm prefix iwill shed some light on this issue. 3.3. Animacy and the i- marker The three third-person AGT paradigm markers in (1) (repeated here as (32) for convenience) do not all have equal status. (32) person 1 2 3MASC 3FEM 3 AGT PAT tonowomoi- hiniwimia- First, they differ in terms of the animacy of their referents. The morphemes moand wo- are glossed as FEMININE and MASCULINE, respectively, and cross-reference only human arguments (or non-human referents which are anthropomorphized, as in folk tales). In contrast, the i- morpheme typically refers only to non-human referents. Thus, compare the following examples: 29 (33) o-ngoheka mo-bata NM-woman 3FEM:AGT-jump ‘The woman jumped’ (34) o-goduru wo-bata NM-young.man 3MASC:AGT-jump ‘The young man jumped’ (35) o-dodiha i-bata NM-snake 3AGT-jump ‘The snake jumped’ Though the i- morpheme is generally referred to in the literature as a nonhuman marker (cf. Hueting 1936, “zakelijke”), there are several situations in which i- can cross-reference human arguments. Based on this observation, Howard Sheldon introduces the term “fourth person” to refer to the i- marker in his description of Galela verbal cross-referencing (1991:168-70). Here Sheldon does not intend the term fourth person in the traditional sense of a switch-reference marker. Rather, the distinction between third and fourth person in NH languages is based on the two interacting parameters of animacy and transitivity. Sheldon (1991 and pers. comm.) develops an animacy hierarchy along which the distinction between third and fourth person varies depending on the verb class. For transitive verbs, the distinction is based on number: third person human singular agent arguments of transitive verbs are cross-referenced via the human AGT paradigm markers (i.e., mo- or wo-), as in (36). In contrast, third person human plural agent arguments are cross-referenced via the i- marker, as in (37), rather than the third person plural AGT paradigm marker yo-, which is used in intransitive clauses such as (38). 30 (36) wo-hi-gohara 3AGT-1PAT-hit ‘He hit me’ (Hueting 1936:351) (37) i-hi-gohara (*yo-hi-gohara) 3AGT-1PAT-hit ‘They hit me’ (Hueting 1936:351) (38) yo-lio 3PL:AGT-return.home ‘They went home’ For intransitive verbs, Howard Sheldon places the critical point distinguishing Galela third and fourth person markers between “definite” and “indefinite” referents. Whether or not this is actually the case for Galela, it is certainly not true for Tobelo.13 Single arguments of Tobelo intransitive verbs which are pragmatically non-identifiable (Sheldon’s “indefinite“) are usually cross-referenced with the third person plural AGT paradigm marker yo-. Thus, in (39) the nominal argument is identifiable, whereas, in (40) it is not. 13 A short expository text contained in Sheldon’s article actually offers conflicting evidence (1991:170). The first line of the text contains a clause which is translated as ‘we just call the village midwife’. The referent ‘midwife’ is clearly definite, as the hearer can assign unique reference via shared cultural information. Moreover, this referent appears with a definite article in the English translation. However, the referent is indexed to the verb via the fourth person prefix. Sheldon evades this problem by claiming that this referent “must be understood to be indefinite” (p. 169). 31 (39) o-nyawa wo-boa NM-person 3MASC:AGT-come ‘The man is coming’ (40) o-nyawa yo-boa NM-person 3PL:AGT-come ‘Someone is coming’ However, when the agent of a transitive clause is unknown or not relevant to the discourse, it is cross-referenced via the non-human i- marker. This yields a construction which is syntactically similar to an agentless passive or impersonal construction, as in (41); however, no passive morphology is present. (41) ngohi-o i-hi-ahoko to-karajanga 1PRO-also 3AGT-1PAT-CALL 1AGT-work ‘I was also called upon to work’ (Kukihi) (or: ‘They called me to work’) If explicit reference is made to the agent argument of a clause, then that argument must be cross-referenced with a human cross-reference marker, as in (42). (42) o-Jepang-oka NM-Japan-LOC manga-haeke, 3PL:POSS-head wo-hi-ahoko to-wohama o-Heiho. 3MASC:AGT-1PAT-call 1AGT-enter NM-indigenous.militia ‘The Japanese leader called me to enter the Heiho’ (Kukihi) Similarly, in connected discourse, the non-human cross-reference marker may be used to refer to human agents only when they are not particularly topical or locally relevant within the discourse. The non-human marker i- cannot be used to refer to 32 humans referents which are highly topical, for example, those referents which have just been mentioned in the discourse. Deidre Sheldon has analyzed a similar situation in Galela, noting that “in any stretch of discourse, only the topical participants will be marked by prefixes on the verbs” [orig. emph.] (1986:236). As example (43) demonstrates, in Tobelo non-topical human agents are crossreferenced via the non-human marker.14 (43) a. b. c. jadi, therefore ...(2.4) m-a-¥e o-kia 3FEM:AGT-3PAT-take o-gorogoro NM-taro ma-hoka. NM-leaf d. ...(2.6) e. ... o= /\ya. NM yeah f. h-ato o-gorogoro ga: 1INC:AGT-say NM-taro that g. ...(2.8) o- -NM-what NM ya ngone h-ato na: naga yeah 1INC:PRO 1INC:AGT-say here exist o= -NM o-talas NM-taro 14 Transcription follows the conventions of Du Bois et al (1993). o-kia o= -NM-what NM 33 h. i. j. k. l. ya? yeah <P ya, yeah daun talas P>. leaf taro i-hi-tamunu, 3AGT-CAUS-cover ... <P i-hi-tamunu-uku P>. 3AGT-CAUS-cover-DIR ‘Thus, she took what, a taro leaf. Yeah, what do we call it? Taro, yeah, taro leaf. She covered it over with that. She covered it over. (Tjileni) In this example, taken from a modern re-telling of a traditional folk tale, the female protagonist indexed by the third person feminine AGT paradigm marker in the second line of (43) is clearly the topical participant. She continues to appear as the active argument of clauses throughout the narrative. Yet, in lines (43k) and (43l), this character is cross-referenced via the non-human marker (i-hitamunu). This reflects the fact that no reference to this character appears in the intervening lines (43c) through (43j), so that the character has very low local topicality. In these lines, the narrator is concerned with clarifying reference to the object — a taro leaf — which the protagonist is using to cover a spring. The taro leaf is not particularly thematic; no further reference to it occurs in the narrative. Yet the taro leaf does have a certain degree of local topicality or relevance to the immediate discourse. In line (43j-k) the narrator is concerned not with the (identifiable) agent 34 of the action of covering, but rather with the instrument — the taro leaf. Thus, the agent is cross-referenced via the non-human marker i-. Clearly there are many more subtleties to the distribution of human and non-human third-person cross-reference morphemes, and a thorough explanation would require extensive discourse studies which are beyond the scope of this paper. The purpose of the foregoing discussion is merely to demonstrate that the imarker does have a role to play as an agent marker in active, indeed even transitive clauses. The precise nature of the reference of this marker depends to a large degree on the pragmatic salience of the participant which it indexes, but in any case the i- marker in these constructions is clearly referential. 3.4. Stative verbs and the i- marker There are, however, many cases in which the i- marker does not appear to be referential, particularly in stative constructions such as (44). (44) i-hi-bole 3AGT-1PAT-tired 'I’m tired' I have glossed i- here as the third person AGT paradigm marker, but it differs in several respects from other AGT paradigm markers, or even from the i- marker referring to agents of transitive predicates as in (37). First, it cannot be anaphoric. In other words, constructions such as (44) are semantically intransitive, lacking an agent argument. Second, only the i- marker occurs in such constructions; no other AGT paradigm markers may occur in stative constructions. It seems likely that this 35 i- morpheme has indeed grammaticized from a third person AGT paradigm marker. But then what is its synchronic function? Describing a similar construction in the Papuan language Marind, Boelaars (1950) argues that the AGT marker actually refers to an inanimate causer, as in (45) (square brackets in the example mark an infix). (45) a-huya[na]v 3AGT-tremble[1PAT] ‘I shiver’ Much early research in NH languages seems to implicitly assume Boelaars explanation, referring to these stative constructions as passives (cf. van Baarda 1908:80-81). Van der Veen notes in resignation that “the effect of this suffix [sic] i cannot be determined from the data” (1915:47). Clearly i- cannot be a true passive marker, as in a passive construction the valency change would at a minimum be signaled by a concomitant change in cross-referencing paradigm. Yet single arguments of Tobelo stative verbs exhibit cross-referencing via the PAT paradigm, not the AGT paradigm as with (active) intransitive verbs. Perhaps the strongest and most obvious evidence against the passive interpretation of stative verbs is the observation that stative verbs are not derived constructions. Most stative verbs have no corresponding “active” counterpart (cf. (44)). Further insight into the role of the i- morpheme can be gained by considering comparative evidence. Data from other NH languages suggest that the third person singular AGT paradigm morpheme i- which occurs on Tobelo stative 36 verbs may be merely a syntactic residue of the aforementioned constraint against verbs with initial PAT paradigm cross-referencing morphemes (see §3.2). In the closely related language Tabaru, the occurrence of the i- morpheme is governed by the person of the nominal argument.15 The i- morpheme occurs only with first person (singular and plural) arguments, as in (46). (46) i-na-tootasa 3AGT-1INC:PAT-angry 'We’re angry' (Fortgens 1928:362) (47) ni-tootasa 2PL:PAT-angry 'Y’all are angry' (Fortgens 1928:362) In contrast, plural arguments of Tabaru stative verbs are cross-referenced via a PAT marker which is not preceded by the i- marker, as in (47). In Galela, the constraint against the occurrence of PAT paradigm markers without AGT paradigm markers appears to have weakened even further in recent times. Both van Baarda (1891) and van der Veen (1915) list Galela stative verbs which contain the initial i- morpheme. More recent work by Sheldon (1991; pers. comm.) finds no evidence for such a constraint. In fact, use of the i- morpheme on stative verbs is ill-formed in modern Galela. Even in Tobelo I have found that the 15According to a recent survey by Voorhoeve (1988), nearly ninety percent of Tabaru basic vocabulary is cognate with Tobelo, making Tabaru the most closely related to Tobelo of all NH languages. 37 initial i- frequently elided from stative verbs in fast speech, indicating that Tobelo may soon follow Galela in this change. All of these observations are consistent with recent apharesis, which often applies irregularly within morphological paradigms (Hock 1991). From this we conclude that it is misleading to interpret the initial i- on stative verbs as either a cross-referencing marker or a de-transitivizing marker. Thus, to answer the question posed at the end of §3.3, stative verbs should be aligned with intransitive verbs to form a class of one-argument verbs, morphosyntactically opposed to the class of two-argument (transitive) verbs. This permits a coherent analysis of NH grammatical relations, both for languages which require the i- marker and for those that do not. In addition, it allows stative verbs in languages like Tabaru to be analyzed as a single class, irrespective of the occurrence of the i- morpheme. This completes our summary of Tobelo verbal morphology. The description of the synchronic function of the i- marker must remain incomplete for the present. However, it should be clear that stative verbs marked with the i- prefix are not transitive or passive, but should rather be interpreted as intransitive verbs which cross-reference a single nominal argument via the PAT paradigm. We now turn to the question of what determines the division of the category of oneargument verbs into intransitive verbs and stative verbs. 38 4. Grammatical relations Previous researchers have attempted to explain the “split” between NH active intransitive and stative intransitive verbs in terms of traditional linguistic structures. In his description of Tobelo’s southern neighbor Tabaru, Fortgens views the split in terms of lexical categories, referring to intransitive verbs which denote states as verbal adjectives (1928:362). He notes that the stative verbs in Tabaru are precisely those which correspond to Dutch participial adjectives, “formed with the helping verbs have and be” (1928:362). Indeed, some Tobelo stative verbs do correspond to Dutch (or even English) participials, such as ‘fallen’ in (48). (48) i-na-otaka .. dau o-wai-uku 3AGT-1INC:PAT-fall down NM-valley-DIR ‘We have fallen down into the valley’ (Paltiel) However, not all stative verb constructions in Tobelo correspond to participials. For example, stative verbs denoting property concepts may also code attributive functions, as with timino ‘old’ in (49). (49) o-nyawa i-wi-timono wo-boa-iha NM-person 3AGT-3MASC:PAT-old 3AGT-come-landward ‘an old man came in’ Furthermore, stative verbs need not denote property concepts, but may also refer to actions which are construed as states, such as mongoro ‘to faint’ in (50). (50) i-mi-mongoro-oka 3AGT:3FEM:PAT-faint-PERF 'She has fainted’ 39 Thus, we cannot distinguish the class of stative verbs as an adjective subcategory of verbs. Current explanations for the split in the intransitive category focus on the non-uniform morphological treatment of S macro-role. In a recent article, Howard Sheldon (1991) describes the verbal cross-referencing system of Galela. As noted above, Galela differs from Tobelo in allowing case markers from the PAT paradigm to occur without the AGT paradigm markers. However, in other respects the Galela cross-referencing system is almost identical to that of Tobelo. For example, consider the following Galela intransitive verbs. (51) wo-sone-ka 3MASC:AGT-die-PERF 'He died' (van Baarda 1891) (52) si-bole 1PAT-tired 'I’m tired' (van Baarda 1891) Sheldon notes that Galela one-argument verbs — like those in Tobelo — do not receive uniform treatment with respect to cross-reference markers. Observing the syntactic patterning of the two paradigms of cross-reference markers, Sheldon classifies the Galela verbal cross-referencing system as split-S, drawing on 40 terminology from Dixon (1979).16 Nominal arguments of some Galela intransitive verbs, such as sone ‘to die’, are treated in the same manner as agent arguments of transitive verbs (AGT paradigm). On the other hand, nominal arguments of other intransitive verbs, such as bole ‘be tired’ are treated like non-agent arguments of transitive verbs (PAT paradigm). This “split” in the semantico-syntactic role of subject-of-intransitive (S) motivates the split-S terminology, although as Palmer points out, this choice of terminology is rather unfortunate “because the term ‘split’ is used in a rather different sense in ‘split ergativity’” (1994:66). 4.1. Split intransitivity The split which Sheldon and others refer to is actually a semantic split between active and stative verbs. This in itself is not a new observation. Even van der Veen (1915) recognized a category of “active” intransitive verbs. However, as noted by Dixon (1979), there exists quite a bit of cross-linguistic variability with respect to the precise characteristics of the split in so-called split-S or split intransitive type systems. In describing such “unaccusative” phenomena in a number of languages, many authors have taken this variability to argue against a 16Sheldon actually uses the phrase “split S-marking type of ergative language” to refer to Galela, though it is difficult to see what motivates his use of the term “ergative” in this respect. In fact, Dixon cautions against the use of the term ‘ergative’ in this sense. Furthermore, Dixon’s “fluid-S” might be more appropriate, because, like Eastern Pomo and other examples cited by Dixon, many Tobelo intransitive verbs may occur with either active or stative morphology. 41 semantic explanation for the active-stative distinction (cf. Perlmutter 1978, Rosen 1984). However, the active-stative distinction in Tobelo is much more robust than the parallel accusative/unaccusative distinction in languages such as Italian. In Tobelo the distinction is reflected in obligatory morphological marking; in Italian the distinction is apparent only in restricted syntactic contexts. Recent work by DeLancey (1985), Durie (1985, 1988), Van Valin (1990) and especially Mithun (1991a) suggests a more coherent analysis of split-S and fluid-S marking languages. All such languages distinguish grammatical case-roles based on semantic, rather than syntactic criteria. The precise details of the distinguishing criteria vary cross-linguistically in terms of the relative importance of various semantic parameters of activity, including aspect, agency, control and affectedness. However, within a particular language, the division between more active and more stative verbs tends to be highly regular and systematic. Thus, closer examination of the particular lexical items involved in the putative split reveals very natural semantic criteria which govern the split. Certainly, this is true of Tobelo, for which the distinction between active and non-active verbs appears to be correlated with the use of different verbal cross-referencing paradigms. 4.2. The nature of the split As Mithun (1991a) points out, the precise semantic criteria on which the active-stative distinction is based need not be stable across languages, so that verbs which are coded as active in one language need not be active in another language, 42 and vice-versa (cf. also Van Valin 1990). Such variation has been taken by some authors as evidence against a semantic basis for the active-stative distinction. For example, in some split intransitive languages (e.g., Guarani, Choctaw, Tobelo), the verb ‘die’ occurs with active morphology, while in other languages (e.g., Lakhota, Central Pomo) ‘die’ occurs with stative morphology (Mithun 1991a). Indeed, Howard Sheldon notes that the class of Galela active verbs “is not defined by strict semantic criteria” (1991:162). In support of this claim he cites the Galela verbs sone ‘die’, cawaro ‘be smart’ and dodo ‘be clear’, all of which are coded via active (AGT paradigm) cross-reference morphology (cf. Tobelo cognates honenge, cawaro and dodono, respectively). Based on some putative universal semantic parameters of activity, the categorization of these particular verbs as active does seem a bit odd. In particular, the act of dying does not typically involve control or agency. Yet, a closer look at the nature of the semantic split between active and stative verbs reveals that the Tobelo active-stative distinction is not strictly sensitive to semantic parameters relating to agency. That is, the division between active and stative morphology is sensitive to the semantics of the verb complex, rather than the semantic roles of the participants. Intransitive verbs which refer to events always employ active (AGT ) cross-reference morphology, regardless of whether the single nominal argument is agentive. In order to demonstrate this, it is useful to separate the various different (though partially overlapping) semantic 43 parameters of agency. Mithun (1991a) distinguishes four such parameters, namely eventhood, performance/effect/instigation (P/E/I), control and affectedness. The parameter of eventhood distinguishes between dynamic, telic events and time-stable, atelic states. Thus, eventhood captures the semantic dimension of time-stablity. Distinguishing between P/E/I and control effectively teases apart the notion of ‘actor’ which Foley and Van Valin define as “the participant which performs, effects, instigates, controls the situation denoted by the predicate” (1984:29). The general notion of actor casts too wide a net to capture the relevant semantic parameters of split intransitive systems. All systems seem to be equally sensitive to performance, effect and instigation, in the sense that if a system is sensitive to any one of these three parameters, then it is also sensitive to the others. This justifies lumping P/E/I as a single semantic parameter (which is often referred to simply as “agency”). However, many split intransitive systems are independently sensitive to the two parameters of P/E/I and control. For example, Acehnese codes the non-controlling (i.e., non-volitional) argument of an intransitive verb as a grammatical patient, regardless of whether that argument is performing, effecting or instigating. The affectedness parameter reflects whether the event/state has a significant effect on one or more of the participants. This last parameter is somewhat more ad hoc than the others, but it has been shown to crucial in at least one language. For example, Central Pomo codes participants as grammatical patients only when they are both out of control and significantly 44 affected, so that many atelic predicates, such as ‘be tall’ take grammatical agent arguments (Mithun 1991a:520-1). Combinations of values for these four binary parameters give rise to sixteen logically possible predication types; however, not all of these are realized due to conditional implications between the various parameters. For example, a participant who controls a situation (event/state) must necessarily also perform, effect or instigate that event, though the converse does not hold. Also, events always affect a participant therein, as do states with a controlling participant. Accounting for these various implicatures reduces the number of predication types from sixteen to seven. The table in (53) lists examples of Tobelo intransitive verbs representing each of these seven predication types, together with the grammatical marking associated with the cross-reference morpheme. 45 (53) Semantic parameters of split intransitivity. Semantic Parameter affected Tobelo dynamic P/E/I contro Verb l ness bata + + + + hangeru + + + honenge + + gogama + + kioko + puturungu bole + English Gloss AGT/ PAT jump sneeze die shivering be asleep strong tired AGT AGT AGT PAT PAT PAT PAT The prototypical semantic agent is a controlling participant in an temporally dynamic event who also performs, effects or instigates the event, as with the Tobelo verb bata ‘jump’. And indeed, the single participant of bata is coded as a grammatical agent. Other examples of Tobelo verbs which belong to this class are given in (54). (54) Controlling, performing participants of intransitive events hioru hoho oara ohiki o¥omo temo dumunu toimi ‘paddle’ ‘fly’ ‘run’ ‘bathe’ ‘eat’ ‘speak’ ‘dive’ ‘shoot’ But this need not be true of all grammatical agents. As the table in (53) crucially demonstrates, the single participant of a Tobelo intransitive predicate may be 46 grammatically encoded as an agent, even if it is not controlling, as with hangeru ‘sneeze’, or not performing/effecting/instigating, as with honenge ‘die’. Other verbs which behave like hangeru are listed in (55). (55) Performing, non-controlling participants of intransitive events adono ari gegoto gehanga gogere guroko hangeru iete wunenge ‘reach’ ‘cry’ ‘worry’ ‘yawn’ ‘live’, ‘dwell’ ‘snore’ ‘sneeze’ ‘laugh’ ‘vomit’ Verbs which behave like honenge are less frequent, but one common example is ¥ahini ‘drift away’, as in the following example. (56) de i-sobo-oli i-¥ahini, and 3AGT-depart-REP 3AGT-float ka yo-dotoaka o-iwi ma-pako. because 3PL:AGT-break.off NM-rattan REL-large ‘and they floated away again, because the big rattan lines broke’ (Tjileni) An additional implication (53) is that participants can be grammatically encoded as patients, even when they are semantically performing, effecting or instigating, as with gogama ‘shivering’ and kioko ‘asleep’. For example, consider magawe ‘diligent’ in (57). 47 (57) i-wi-magawe una. 3AGT-3MASC:PAT-diligent 3MASC:PRO de ai-hininga i-rahai. and 3MASC:POSS 3AGT-good ‘He was diligent, and his heart was good.’ (Paltiel) Moreover, even for those states for which the participant is not performing, effecting or instigating, the participant is grammatically coded as a patient regardless of whether it is affected, as in (58), or not, as in (59). (58) Stative intransitive verbs with affected participants darato gogama mongoro pehaka tilibu (59) ‘storm-bound’ ‘shivering’ ‘become faint’ ‘wet’ ‘pregnant’ Stative intransitive verbs with unaffected particpants dapa¥u hahini hauku kuata modongo omu ‘depressed’ ‘hungry’ ‘hot’ ‘strong’ ‘angry’ ‘jealous’ For Tobelo eventhood is the single salient parameter which distinguishes between grammatical agents and grammatical patients. Still other intransitive verbs, such as eluku ‘lie, deceive’ may crossreference nominal arguments via either paradigm, as in (60) and (61). 48 (60) wo-eluku-oka 3MASC:AGT-lie-PERF ‘he lied’ (61) i-wi-eluku 3AGT-3MASC:PAT-lie ‘he is a liar’ The division here, like that between (29) and (30) above, is based on aspect. In example (60), eluku ‘lie’ is construed as a single telic action. In example (61), this verb is construed as an atelic state, without a well-defined end point. Thus, (61) might be equally well glossed as ‘he continuously lies’ or ‘he is in a state of lying’. Many Tobelo intransitive verbs behave like eluku, in that they may be construed as actions or states, and the choice of cross-reference morphology reflects this construal. Some additional examples are given in (62). (62) birahi hihanga kioko lihiti modongo ngamo tikiti tohata AGT PAT ‘rejoice’ ‘go astray’ ‘go to sleep’ ‘sprain’ ‘fear’ ‘quarrel’ ‘cough’ ‘angry’ ‘be happy’ ‘be lost’ ‘be asleep’ ‘have a sprain’ ‘be afraid’ ‘be quarrelsome’ ‘cough continuously’ ‘evil’ Similar observations have been made for other active-stative languages. For example, the Guarani verb karú means ‘to dine’ in the active form and ‘to be a glutton’ in the stative form (Mithun 1991a:513). Like Tobelo, the division is based on aspect. The first sense of karú refers to a telic action, while the latter sense refers to a durative state. 49 Furthermore, for some intransitive verbs, this difference in construal distinguishes inchoative meanings, as in (63). The inchoative form is construed as an action and cross-references nominal arguments via the AGT paradigm. The stative form is construed as a state and cross-references nominal arguments via the PAT paradigm. (63) AGT PAT wo-daluku-ohi 3MASC:AGT-intoxicated- i-wi-daluku 3AGT-3MASC:PAT-drunk DUR ‘He getting drunk’ ‘He’s drunk’ mo-hiri 3FEM:AGT-sick ‘She’s getting sick’ i-mi-hiri 3AGT-3FEM:PAT-sick ‘She’s sick’ uha no-lepa NEG 2AGT-muddy ‘Don’t get muddy!’ i-ni-lepa 3AGT-2PAT-muddy ‘You’re muddy!’ This can be viewed as a special case of the distinction in lexical aspect between grammatical agents and grammatical patients. The inchoative sense of the verbs in (63) clearly have a more dynamic, telic connotation and are naturally classified as events. A similar alternation is found in Central Pomo, which marks agent-patient distinctions based on the dual semantic parameters of control and affectedness. However, in Central Pomo the inchoative form receives PAT case. Mithun remarks that “the coming into being of a state is viewed as affecting a participant more than simply being in a state”, thus “basic adjectives appear with agent-case arguments, while their inchoative counterparts appear with patient-case arguments” 50 (1991a:521). Thus, the use of grammatical case marking to discriminate between inchoative and stative senses appears to an important cross-linguistic feature of semantically based systems of grammatical relations. Tobelo grammatical relations distinguish “active” verbs, regardless of whether they have volitional agents (cf. (54)) or not (cf. (55)). For languages such as Guarani and Tobelo which make active-stative distinctions based on lexical aspect, it is not unusual for the verb ‘die’ to be construed as an activity. Of course, since some Tobelo one-argument verbs may occur with either AGT paradigm or PAT paradigm markers, we might expect to find a stative, atelic form of honenge with the meaning ‘to be dead’. However, in Tobelo this meaning is conveyed via overt aspectual marking, as in (65), and no stative form such as (66) exists. (64) wo-honenge 3MASC:AGT-die ‘He is dying’ (65) wo-honenge-oka 3MASC:AGT-die-PERF ‘He is dead’/’He has already died’ (66) *i-wi-honenge 3AGT-3MASC:PAT-die Intended meaning: 'He is dead' These examples should not be taken as evidence for a constraint against the occurrence of overt aspectual morphology on stative intransitive verbs. Indeed, stative verbs occur freely with aspectual marking, including the PERFECTIVE suffix -oka, as in (67). 51 (67) ... ma ngohi, but 1PRO ... t-ato t-oik-oka-ua 1AGT-say 1AGT-go-PERF-NEG i-hi-bole-oka. 3AGT-1PAT-tired-PERF ‘But I said I’m not going. I’m already tired.’ (Paltiel) The nonexistence of a form such as (66) can readily be viewed as an historical accident. In general, intransitive verbs which occur as both active and stative form may also take overt aspectual marking in both forms. But in the particular case of honenge ‘die’, the perfective, time-stable sense is already conventionalized in the form (65) which is marked with an overt PERFECTIVE suffix. Thus, the corresponding atelic, stative form is not present, its function having been usurped by the conventionalized honenge-oka ‘be dead’. In any case, the verb ‘die’ straddles the active-stative distinction, being neither a prototypical dynamic event nor a prototypical atelic state, and taking a single argument which is neither a prototypical semantic agent nor a prototypical semantic patient. It is precisely for such verbs that we find a high degree of cross-linguistic variation with respect to active verbal cross-referencing. The semantic nature of the active-stative division is radial, whereas the morphological constraints on verbal cross-referencing require a binary distinction between active and stative. That the categorization of some verbs which are neither prototypical states nor prototypical events should appear arbitrary is to be expected. 52 4.3. Comparative evidence Similar arguments could be made regarding the classification of cawaro ‘be smart’ and dodo ‘be clear’, but alternate explanations are available as well. The verb cawaro is borrowed from the Ternate verb waro ‘to know’. Ternate, though related to Galela and Tobelo, does not distinguish two paradigms of case markers, but rather uses one set of case markers, which are cognate with the Tobelo case markers from the AGT paradigm, i.e., the active cross-reference markers, as in (68). (68) Ternate Tobelo Galela fo-waro ho-cawaro po-cawaro ‘we know’ (de Clercq 1890) ‘we are smart’ ‘we are smart’ (van Baarda 1891) Thus, it is not unexpected to find the verb cawaro borrowed into Tobelo and Galela as an active verb. With respect to dodo we note the importance of looking beyond synchronic explanations, as emphasized by Mithun (1991a). It is possible that dodo, which now functions as a property concept word, derives from a lexeme which had a more active inchoative sense, as in ‘to become clear’. Semantic categorization, when supported by morphological marking, tends to be rather resistant to change, thus the strictness of semantic criteria can be obscured by diachronic factors such as lexicalization, and the possible influence of such factors cannot be ignored when the “naturalness” of the active-stative distinction is considered. As Mithun remarks, “case selection may be made on some principled basis when a predicate becomes established in the language; but the combination of predicate plus case 53 can become lexicalized and subsequently learned and used as a unit…. The basic meaning of verbs themselves may be extended or shift over time, while the case remains intact” (1991a:540). Nor is such a caveat any less applicable to ergative or accusative systems. For example, the existence of “dative subjects” in Germanic languages does not constitute evidence against the nominative/accusative pattern of grammatical relations in those languages. This is especially true for Tobelo, since little or no historical data are yet available with which to make judgments regarding the possible influence of grammaticalization and lexicalization on the semantic categorization of verbs. 4.4. Event-sensitivity versus participant-sensitivity The important point demonstrated in this section is not only that the split between Tobelo stative and intransitive verbs is semantically based, but more importantly that this split is based on the semantics of the verb. The choice of Tobelo pronominal cross-reference marker is not based on agency, or any subparameter of agency. Rather, the choice of cross-reference paradigm reflects an aspectual characterization of the verb. Intransitive verbs which are construed as events cross-reference nominal arguments via the AGT paradigm; verbs which are construed as states cross-reference nominal arguments via the PAT paradigm. Grammatical relations encoded via Tobelo verbal morphology are sensitive to semantic properties of the situation rather than the participants. Thus, Tobelo pronominal cross-reference markers pattern in an active-stative system of 54 grammatical relations. In the following section we turn our attention to the relationship between Tobelo grammatical relations and discourse structure. 5. Preferred argument structure In the past decade, several authors have discovered correlations between the discourse pattern of core nominal arguments and the grammatical relations coded on those arguments. Not only do such studies have potential implications for our understanding of the diachronic motivation of systems of grammatical relations, but they also point to a third dimension of grammatical relations, apart from the morphological and syntactic dimensions. Du Bois’ (1985) seminal study of Sacapultec (Mayan) discourse, an ergative language, finds evidence for a “preferred argument structure” which treats the semantico-syntactic macro-roles S and O alike, in that each clause tends to have no more than one full NP and this NP tends to be either S or O. In contrast, A tends not to be realized as a full NP. Thus, not only do Sacapultec S and O share the absolutive morphological marking, they also pattern together in discourse. In regard to split intransitive systems, it is natural to ask to what extent the morphological distinction between active and stative arguments is reflected in the discourse patterning of those arguments. Durie has addressed just this question for Acehnese, concluding that in Acehnese discourse, grammatical agent (“Actor”) arguments are mostly null, while grammatical patient (“Undergoer”) arguments are mostly full NPs (1988:11). Thus, at first glance, Acehnese discourse can be 55 said to pattern as an agent-patient system.17 For Tobelo quite a different picture emerges. Like Acehnese, Tobelo provides clear morphological evidence for a semantically based system of grammatical relations and no persuasive evidence for an alternate system of syntactic relations. Yet, in contrast to Acehnese, the activestative pattern which is so prominent in Tobelo morphology is not reflected in discourse patterns. This observation emerges from a short study of Tobelo narrative texts. The data on which this study is based are derived from two sets of Tobelo texts. The first set consists of a selection of twenty-five short folk tales collected by the Dutch missionary linguist Anton Hueting early this century (Hueting 1908b). These texts were originally dictated to one of Hueting's Tobelo assistants, and Hueting claims to have made no significant changes to the texts (1908b:10): De taalkundige lezer vindt die van zelf, want ik heb niets genivelleerd, maar alles gelaten zooals het opgeschreven is. Mijne correctie heeft zich bepaald tot het verbeteren van wat bepaaldelijk foutief was opgeschreven. Ik veranderde echter niet een vorm. 17 Durie actually refers to an “active” system, though the split between grammatical agent (Durie’s “Actor”) and grammatical patient (Durie’s “Undergoer”) is actually based on agency, specifically volitionality, as in an agent-patient system. As Durie notes, “[Acehnese] Actors are always volitional” (1988:6). 56 The linguistic reader will find for himself that I have leveled nothing, and everything has been left as it was written down. My corrections have been limited to improving particular mistakes in the transcription. I have not changed any forms. The second set of data is comprised of six spoken narrative texts collected by the author in 1995, comprising a corpus of roughly five thousand intonation units of spoken Tobelo. These texts were transcribed from tape recordings of six speakers in the same dialect area of Halmahera in which Hueting's tales were recorded. Each referring noun phrase in the corpus was coded both for grammatical macrorole (A, O, Active S and Inactive S), for lexical form (full NP, pronoun or zero) and for activation state (Given, New). For this study, a lexical zero is considered to be any pronominal cross-reference marker for which no co-referential pronoun or noun phrase occurs in the same clause.18 Activation state is taken in the sense of Chafe (1994), broadly interpreted as a binary division between Given and New. Both Chafe and Durie make finer distinctions along the continuum between Given and New . Thus Given, as used here, collapses Chafe’s categories of Given and Accessible, and Durie’s categories of Mentioned, Evoked, Accessible and Inferable. The category of New used here corresponds to Durie’s category of Brand New. The finer graduations used by these authors are not relevant to the 18 Recall from §3 that Tobelo pronominal cross-reference markers are obligatory but corresponding co-referential full NPs are not. 57 present discussion, which focuses on referent introduction, as captured in New mentions. This approach permits cross-linguistic comparison of Tobelo discourse patterns with patterns found in previous studies of Acehnese (Durie 1988) and Sacapultec Mayan (Du Bois 1987). It should be noted, however, that there is some danger in interpreting the results of such a comparison too liberally. In comparing discourse patterns across these languages the variable of discourse genre is not necessarily held constant. The Sacapultec data consist of narratives elicited in response to fixed visual stimuli; the Acehnese data consist of epic narratives and folk tales; and the Tobelo data consist of personal and folk narratives. Each of these three genres may employ referent tracking devices for genre-specific purposes, and hence any measure of grammatical or lexical form of noun phrases in these texts may reflect these genre-specific motivations. Still, all three genres do represent narrative, and cross-linguistic variation in the discourse patterning of noun phrases across these three languages is likely to be at least partially indicative of differing systems of preferred argument structure among the languages. 5.1. The one full NP constraint As discussed in Du Bois (1987), studies of connected discourse in typologically diverse languages have revealed a universal tendency toward avoidance of more than one lexical argument per clause. As noted by Durie (1988) for Acehnese and as shown in (69) and (70) for Tobelo, languages with 58 semantically based systems of grammatical relations are no exception to this one NP constraint. (69) Tokens of Tobelo clauses. Valency Transitive Intransitiv e Total (70) Number of full NP arguments 0 1 2 93 39 5 23 14 0 116 53 5 Total 137 37 174 Proportion of Tobelo full NP core arguments in clause. U se W ord 6. 0c or later to view Macintosh picture. However, the means by which this constraint is met need not be constant across languages. In fact, Durie notes striking and typologically motivated differences in the means by which Sacapultec and Acehnese meet the one NP constraint: “Sacapultec does it ‘ergatively’, Acehnese does it ‘actively’” (1988:14). In these terms, one could say that Tobelo does it ‘accusatively’. As shown in (71) and (72), only the Tobelo transitive PAT arguments (O macro role) occur as full NP arguments with any significant frequency. 59 (71) Tokens of Tobelo nominal arguments by macro-role. Macro-Role Transitive AGT Intransitive AGT Transitive PAT Intransitive PAT Total (72) NP 7 18 22 0 47 Lexical Form PRO ZERO 16 33 16 91 0 16 3 15 35 155 Total 56 125 38 18 237 Proportion of Tobelo full NP arguments by macro-role. U se W ord 6. 0c or later to view Macintosh picture. This result contrasts markedly with the distribution of full NP arguments in Sacapultec and Acehnese. As shown in (73) and (74), Sacapultec discourse groups S and O together, as opposed to A. Thus, the frequency of full NP realizations of Sacapultec nominal arguments can be said to embody an ergative discourse relation. 60 (73) Tokens of Sacapultec nominal arguments by macro-role. Macro-Role A O S Total (74) Lexical Form NP PRO ZERO 11 13 156 81 2 94 126 12 124 218 27 374 Total 180 177 262 619 Sacapultec (after Du Bois 1987:822) U se Word 6. 0c or later to view Macintosh picture. As noted by Durie (1988) and shown in (75), Acehnese discourse groups some S with A and other S with O. That is, Acehnese grammatical patients — both transitive and intransitive — tend to occur as full lexical NPs, while grammatical agent arguments do not. 61 (75) Acehnese (after Durie 1988:15) U se W ord 6. 0c or later to view Macintosh picture. Moreover, the discourse distribution of full NP realizations of S arguments parallels the morphological distinction between active and inactive (i.e., stative) verbs. That is, in Acehnese discourse active arguments (macro roles A and Active S) pattern together in opposition to non-active arguments (macro roles O and Inactive S). So, the distribution of full NP realizations of Acehnese nominal arguments describes an active discourse relation. Thus, Tobelo differs from both Sacapultec and Acehnese in that the morphological system of grammatical relations in Tobelo is not reflected in preferred argument structure in discourse. Tobelo clearly exhibits a morphological active relation via an aspectual distinction between prototypical actions and prototypical states, yet the distinction is not found in the patterning of full NP 62 mentions in discourse.19 Further insight into the distribution of Tobelo full NP arguments shown in (71) can be gained by considering the animacy of each core argument. The proportion of human arguments for each macro-role in the Tobelo corpus is listed in (76). As one would expect, active arguments — those crossreferenced via the AGT paradigm — are nearly all human, whether the clause is transitive (A) or intransitive (Active S). In contrast, the animacy of non-active arguments — those cross-referenced via the PAT paradigm — differs markedly depending on the transitivity of the clause. Single arguments of stative verbs (i.e., Inactive S) are all human, while very few O arguments are human. 19 One possible explanation for this difference, not considered here, lies in the nature of the cross-referencing systems in the two languages. With a very few exceptions, as noted in §3.2, Tobelo core nominal arguments are obligatorily cross-referenced on the verb. In contrast, Acehnese active arguments are usually cross-referenced on the verb, while inactive cross-reference markers are usually absent (Durie 1988:3). Since both languages code grammatical relations via these pronominal cross-reference markers, Tobelo can be seen to code grammatical relations much more consistently than Acehnese. This presents little problem for Acehnese, for the morphological system of grammatical relations is supported by the agent-patient distinctions in discourse. Because of the strong alignment of full NP mentions with the inactive arguments, Acehnese grammatical relations are recoverable even when the cross-reference morphology is absent. 63 (76) Proportion of Tobelo core arguments which are human. Transitive AGT (A) Intransitive AGT (Active S) Transitive PAT (O) Intransitive PAT (Inactive S) 100% 95% 16% 100% As expected, Sacapultec and Acehnese differ little in this respect. Like Tobelo, both Sacapultec and Acehnese discourse exhibit low frequencies of human referents in O position. This is due to a probably universal tendency for human referents to have greater topic continuity, or prominence, within narrative discourse describing human activities. In other words, human referents often server as “starting points” in the sense of Chafe (1994). Thus in Tobelo narrative discourse the pressure to code these starting points outcompetes the parallel discourse pressure to code the New information. This observation is especially relevant for Tobelo, because Tobelo conforms to the one NP constraint via an accusative system of preferred argument structure based on topicality, or topic continuity, which avoids full NP mentions of core arguments in other than the O macro role. Sacapultec and Acehnese do not ignore the grammar of topicality; they simply conform to the one NP constraint by other methods. Sacapultec avoids full NP mentions of A arguments, while Acehnese avoids full NP mentions of Inactive arguments. 64 5.2. Participant introduction strategies Though the precise motivations remain moot, it is at least plausible that the patterning of preferred argument structure in an ergative language such as Sacapultec should differ greatly from that of languages with split intransitive systems such as Tobelo and Acehnese. Yet it is not at all obvious why Tobelo and Acehnese should behave so differently from one another in this respect. In particular, why should it be that Acehnese permits full NP mentions of Inactive S arguments while Tobelo does not? To address this question we must examine the role of the Inactive S argument in each of the two languages. Durie’s study finds that 97% of New mentions are coded as grammatical patient (“Undergoer”) arguments (1988:18). This contrasts with Tobelo for which only one quarter (13 of 52 tokens) of New mentions are coded as grammatical patient arguments. Thus, while Intransitive S is the preferred argument for referent introduction in Acehnese, this does not appear to be true of Tobelo, in which New information shows an overwhelming tendency to occur as a transitive PAT (O), an oblique or an unmarked non-core argument, as shown in (77). 65 (77) Tobelo activation state by syntactic role. Macro-Role Transitive AGT Intransitive AGT Transitive PAT Intransitive PAT Oblique Total Tokens Given New 55 1 112 12 23 15 17 1 24 17 231 46 % New 2 10 37 6 41 Durie also finds a statistically significant difference between the frequency of New mentions for Active S and Inactive S arguments, which he explains by observing that “in Acehnese the intransitive Undergoer is the argument in which animate, subsequently thematic referents are introduced” (1988:18). In order to better understand the different patterning of preferred argument structure in Tobelo and Acehnese, we must closely examine the semantic nature of the particular predicates used to introduce New participants in each of the two languages. Tobelo employs a much wider semantic range of predicates for participant introduction. First, in Tobelo narrative New participants are often introduced using the existential-presentative construction with the particle naga ‘exist’, as in (78). This is a particularly common strategy with which to open folk tales. 66 (78) naga exist o-nyawa wo-ma-tengo NM-person 3MASC:AGT-RFLX-alone ai-tapihuma o-karianga 3MASC:POSS-mask NM-lizard ‘There was a man who was disguised as a lizard.’ (Hueting 1908b:128) Tobelo naga cannot take verbal morphology, and hence it is not possible to determine the grammatical status (AGT or PAT) of the nominal “argument” of naga. In other words, naga is not a verb in the sense of §3. It may be possible on an ad hoc basis to classify naga as syntactically intransitive, taking a semantic patient argument. But it does not make sense to speak of naga as taking a grammatical patient argument. Thus I do not include such constructions in the counts in (69) and (77) above. According to Durie, Acehnese also introduces New participants with an existential-presentative construction, namely the verb na ‘to exist’. However, Acehnese na is clearly an intransitive verb taking a non-volitional (PAT) argument. Yet there do appear to be three preferred participant introduction strategies which are shared by both Tobelo and Acehnese. In both languages, New information tends to be introduced in either the oblique role, the O macro-role or the S macrorole. Each of the first two strategies are fairly similar in the two languages. However, the third strategy is effected quite differently, in accordance with the differences in how the S splits in each language. We consider each strategy in turn. 67 The first strategy for introducing animate participants in Tobelo narrative is not explicitly mentioned by Durie (1988), though the occurrence of New information as oblique arguments is found to be quite common in a later study (Durie 1994). As shown in (77), Tobelo often introduces New information as an oblique argument. For example, consider the folk tale which opens with the clause in (79) above. The second major participant in the narrative, bereki ‘old woman’, is introduced as an oblique argument (marked with the directional suffix -ika) of the intransitive verb oiki ‘go’. (79) o-wange moi-uku NM-day one-down de ai-tapihuma wo-ma-hinoa and 3MASC:POSS-mask 3MASC:AGT-RFLX-put.on de w-oiki o-bereki-ka mo-ma-tengo and 3MASC:AGT-go NM-old.woman-DIR 3FEM:AGT-RFLX-alone ‘One day he put on his mask, and he went to an old woman’ (Hueting 1908b:128) It seems likely that this use of oblique arguments may be an important crosslinguistic strategy for participant introduction. Durie’s later study of Acehnese narrative and conversation finds that one quarter (14 of 67 tokens) of New 68 mentions are coded as oblique (Durie 1994:27-28).20 So clearly oblique argument coding represents an important participant introduction strategy for both Acehnese and Tobelo. A second common strategy in both languages is the introduction of a participant as the grammatical patient argument of a transitive clause, i.e., in the O macro-role. For example, consider again the Tobelo folk tale referred to in examples (78) and (79) above. The two major characters, the young man and the old woman, are introduced at the beginning of the tale. Later in (80), a third important character, the King of the East’s daughter, is introduced as the grammatical patient of the transitive verb gahoko ‘call on’. (80) de ma-bereki genanga mo-mi-gahoko and RNM-old.woman that 3FEM:AGT-3FEM:PAT-call.on o-wange ma-nyonyie ngo-ai-ora NM-day RNM-east FEM-3MASC:POSS-daughter ‘And the old woman called on the King of the East’s daugher’ (Hueting 1908b:128) 20 The figure quoted here represents the aggregate for the three texts in Durie (1994). Durie also lists counts for nominal predicates and free nominals, but in order to facilitate comparision with Tobelo data, I have included in my total only the counts for core arguments and obliques. 69 Similarly, in the story of the turtle and the heron, the turtle is first introduced with the existential presentative naga and only later appears as the grammatical agent of the verb adono ‘approach’ in (81). (81) de i-a-adono-li o-totaleo moi and 3AGT-3PAT-approach-REP NM-bird one ‘And he again approached a bird’ (Hueting 1908b:111) The totaleo ‘bird’ turns out to be the blue heron, one of the main characters in the narrative. Here it is introduced as the grammatical patient of a transitive clause (O). This strategy appears to be even more common in Acehnese, for which fully 45% of all New mentions (30 of 67 tokens) occur as transitive Undergoers (Durie 1994:27-8). The final common participant introduction strategy codes New information as the single core argument of an intransitive verb. As Du Bois has noticed for Sacapultec, the intransitive verb used for participant introduction is often a “relatively neutral verb like ‘come’, ‘arrive’, ‘appear’” (1987:831). In particular these verbs tend to show motion toward the speaker’s deictic center (cf. Cumming 1994). For example, in the Tobelo example (82) the referent o-Jepang mangatentara ‘Japanese soldiers’ is introduced as the active argument of the intransitive verb uti ‘arrive’. 70 (82) tahun seribu sembilan ratus empat puluh tiga, year thousand nine hundred four ten three Januari ma-mede, January RNM-month o-Jepang manga-tentara y-a-muruono, NM-Japan 3PL:POSS-soldier 3PL:AGT-PAT-many yo-uti o-Miti ma-nuhu-iha. 3PL:AGT-arrive NM-Meti RNM-island-landward ‘In January of 1943 many Japanese soldiers landed on Meti Island’ (Kukihi) Crucially, the verb uti ‘arrive' in (82) takes a single argument which is coded via the AGT paradigm (Active S). As shown in (77), the Active S argument is a common position for New information in Tobelo. Yet Acehnese narrative contains no tokens of New information coded as a grammatical agent (Active S) argument.21 Recall that Acehnese distinguishes grammatical agents from grammatical patients based on the semantic parameter of volitionality. Thus, we must conclude that Acehnese does not use intransitive verbs with volitional agents to introduce New participants. Indeed, Durie remarks that the “typically presentative verbs are always intransitive, taking an Undergoer, such as na ‘be, exist’ tinggay ‘dwell’ and lahé ‘be born’” (1988:19). In essence, this statement 21 Durie (1994) does find occurences of New information as Active S arguments in the conversational text (6%, 2 of 34 tokens). 71 represents a claim about the semantic structure of participant introduction in Acehnese narrative. Thus, in Acehnese narrative participants are introduced as non-volitional arguments of intransitive verbs and therefore coded as grammatical patients. The difference between Tobelo and Acehnese preferred argument structure stems from subtle but important differences in the semantic distinctions to which Tobelo and Acehnese grammatical relations are sensitive. As noted earlier, Acehnese verbal morphology reflects an agent-patient system; that is, it distinguishes grammatical agent and grammatical patient based on parameters of agency, namely volitionality — a property of the participant. In contrast, Tobelo verbal cross-referencing makes active-stative distinctions between grammatical agent and grammatical patient based on lexical aspect, which is a holistic property of the entire situation. Herein lies the key to the differing discourse patterns observed in the two languages. As we have seen for both Acehnese and Tobelo, when New participants are introduced as core arguments, the tend to occur as nonvolitional arguments of intransitive predicates. Because grammatical relations in Acehnese and Tobelo differ in their sensitivity to the semantic parameter of volitionality, Acehnese and Tobelo assign different grammatical coding to such New participants. This difference in grammatical coding is reflected in discourse measurements which correlate activation state and lexical density with grammatical status. This in turn results in the apparent difference in preferred argument structure between Acehnese and Tobelo. 72 6. Outlook A conception of Tobelo grammatical relations as an active-stative system provides a coherent account of Tobelo verbal cross-referencing which has been overlooked in previous discussions of NH grammar. Cross-linguistically, descriptions of active-stative systems remain relatively rare. However, this may be merely an effect of specious grammatical descriptions which misinterpret distinctions in lexical aspect as distinctions between nominative and accusative case. The active-stative system in Tobelo distinguishes events from states, regardless of transitivity. Unlike some other languages with semantically based systems of case marking and grammatical relations, verbal cross-referencing in Tobelo is not directly sensitive to semantic features of the participants. For example, a Tobelo intransitive verb may cross-reference its single argument via the PAT paradigm, even when that argument is performing, effecting or instigating (as in gogama ‘shiver’), or even when that argument is visibly affected (as in bole ‘tired’). For Tobelo, lexical aspect, as captured in time-stability, is the relevant semantic parameter which governs the choice of verbal cross-reference morphology. In this sense, the pronominal cross-reference system in Tobelo is very different from agent-patient systems in languages such as Acehnese, in which case marking for intransitive predicates is governed by semantic parameters relating to agency. Case marking in Acehnese is sensitive to the volitionality of the intransitive argument; there is no internal distinction in lexical aspect in Acehnese. 73 In other words, Tobelo is sensitive to semantic properties of the predicate, while Acehnese is sensitive to semantic properties of the participants. A cursory examination of discourse patterns suggests that this distinction between situationsensitivity and participant-sensitivity is reflected in Tobelo and Acehnese discourse patterns of information flow. Acehnese, which makes agent-patient distinctions in pronominal cross-referencing, also exhibits agent-patient distinctions in information flow patterns such as the distribution of full NP mentions. In contrast, Tobelo, which makes active-stative distinctions in pronominal cross-referencing, exhibits no semantically based preferred argument structure. Yet, as we have seen in this study, the existence of such a correlation between grammatical relations and discourse patterns does not arbitrary pragmatic linking, but rather is motivated by the lexical semantic structure of participant introduction strategies. Indeed, there is no compelling reason to believe that the semantic properties relevant to split intransitivity should be reflected in discourse patterns. Referent tracking, as indexed by such measures as the occurrence of full NP and New arguments, reflects global properties of discourse participants. On the other hand, semantic properties such as volition, performance and eventhood are local properties of the event/state which are not continuous throughout the discourse, and as such are not directly reflected in the distribution of discourse participants. However, if particular semantic features are observed to cluster about a particular discourse function, then the grammatical coding associated with such semantic 74 features (i.e., AGT or PAT) may indeed exhibit a measurable discourse correlate. This is clearly the case for Acehnese and Tobelo, in which the referent introduction function is often achieved via a non-volitional intransitive verb. Thus, it is the grammatical coding of the non-volitional arguments of such verbs which is captured by the discourse measurements discussed in the previous section. In particular, preferred argument structure does not directly reflect grammatical relations in either Acehnese or Tobelo. What is significant about this study is the observation put forth in §5, that the difference in the discourse behavior between active-stative systems (e.g., Tobelo) and agent-patient systems (e.g., Acehnese) can be traced to the lexical semantic structure of the predicates used for participant introduction in Tobelo and Acehnese. Thus, the discourse pressures which are responsible for preferred argument structure in Tobelo and Acehnese are moderated by the same semantic pressures which govern grammatical relations in the two languages. This observation presents an interesting avenue for future research. Much recent work on the relationship between grammatical relations and discourse patterns has tended to focus on the explanatory value of such correlations, leading Durie to conclude that “grammatical strategies have discourse motivations” (1988:24). However, the mere existence of such discourse correlates, or pragmatic linking, in Acehnese does not imply a causal relationship. The observed pattern of information flow in Acehnese can in fact be directly related to the semantic classification of the intransitive verbs used to introduce New participants. Since 75 such verbs tend to be non-volitional, they take a grammatical patient (or Undergoer) argument. Therefore, New information tends to be coded as a grammatical patient. Thus, the relationship between discourse and grammar in Tobelo and Acehnese is indirect, mediated by relatively transparent verbal semantics. Of course, the existence of such semantic motivations for discourse patterns does not deny the importance of discourse pressure on information flow. In effect, the need for semantic transparency competes with both the need to mark New information and the parallel need to mark thematic arguments. Yet in split intransitive systems the pressure toward semantic transparency prevails. 76 Bibliography Andersen, Stephen. 1976. On the notion of subject in ergative languages. Subject and Topic, ed. by Charles N. Li, 1-23. New York: Academic Press. Andrews, Avery. 1985. The major functions of the noun phrase. Language Typology and Syntactic Description, vol. I, ed. by Timothy Shopen, 62154. Cambridge: Cambridge University Press. Baarda, M.J. van. 1891. Beknopte spraakkunst van de Galillareesche taal. Utrecht: Kemink. Baarda, M.J. van. 1908. Leidraad bij het bestudeeren van 't Galela'sch dialekt, op het eiland Halmahera. 's-Gravenhage: Martinus Nijhoff. Boelaars, J.H.M.C. 1950. The Linguistic Position of South-Western New Guinea. (Orientalia Rheno-Tariectina vol. 30.) Leiden: Brill. Clercq, F.S.A. de. 1890. Bijdragen tot de kennis der Residente Ternate. Leiden: E.J. Brill. Comrie, Bernard. 1989. Language Universals and Linguistic Typology, second edition. Chicago: University of Chicago Press. Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press. Cumming, Susanna. 1994. Functional categories in the lexicon: referent introduction in Indonesian novels. Text 14.465-94. DeLancey, Scott. 1985. On active typology and the nature of ergativity. Relational Typology, ed. by Frans Plank, 47-60. Berlin: Mouton. Dixon, R.M.W. 1979. Ergativity. Language 55.59-138. Du Bois, John W. 1985. Competing motivations. Iconicity in Syntax, ed. by John Haiman, 343-65. Amsterdam: John Benjamins. 77 Du Bois, John W. 1987. The discourse basis of ergativity. Language 63.805-855. Du Bois, John W., Stephan Schuetze-Coburn, Susanna Cumming and Danae Paolino. 1993. Outline of discourse transcription. Talking data: Transcription and coding in discourse research, ed. by Jane A. Edwards and Martin D. Lampert, 45-90. Hillsdale, NJ: Lawrence Erlbaum Associates. Durie, Mark. 1985. A Grammar of Acehnese on the Basis of a Dialect of North Aceh. (Verhandelingen van het Koninklijk Instituut voor Taal-, Land- en Volkenkunde 112.) Dordrecht: Foris. Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11.365-99. Durie, Mark. 1988. Preferred argument structure in an active language. Lingua 74.1-25. Durie, Mark. 1994. Pragmatic linking in Acehnese. Text 14.495-529. Foley, William A. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge University Press. Foley, William A., and Robert D. Van Valin. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Fortgens, J. 1928. Grammatikale aantekeningen van het Tabaroesch, Tabaroesch volksverhalen en raadsels. Bijdragen tot de Taal-, Land- en Volkenkunde 84.300-544. Hock, Hans Heinrich. 1991. Principles of Historical Linguistics, second edition. Berlin: Mouton de Gruyter. Hueting, Anton. 1908a. Tobeloreesch-Hollandsch woordenboek met HollandschTobeloreesch inhoudsopgave. 's-Gravenhage: Nijhoff. Hueting, Anton. 1908b. O Tobelohoka manga totoade: verhalen en vertellingen in de Tobeloreesche taal. Bijdragen tot de Taal-, Land- en Volkenkunde 61.1-318. 78 Hueting, Anton. 1936. Iets over de spaakkunst van de Tobeloreesche taal. Bijdragen tot de Taal-, Land- en Volkenkunde 94.295-407. Klimov, G.A.1984. On the character of languages of active typology. Linguistics 131.11-25. Merlan, Franscesca. 1985. Split intransitivity: Functional oppositions in intransitive inflection. Grammar Inside and Outside the Clause, ed. by Johanna Nichols and Anthony C. Woodbury, 324-62. Cambridge: Cambridge University Press. Mithun, Marianne. 1991a. Active/agentive case marking and its motivations. Language 67.510-46. Mithun, Marianne. 1991b. The role of motivation in the emergence of grammatical categories: The grammaticization of subjects. Approaches to Grammaticization, vol. II, ed. by E. Traugott and C. Heine, 159-85. Amsterdam: John Benjamins. Mithun, Marianne and Wallace Chafe. 1996. What are S, A, and O? ms. Palmer, F.R. 1994. Grammatical Roles and Relations. Cambridge: Cambridge University Press. Perlmutter, David. 1978. Impersonal passives and the unaccusative hypothesis. Berkely Linguistics Society 4.157-89. Rosen, Carol. 1984. The interface between semantic roles and initial grammatical relations. Studies in Relational Grammar 2, ed. by David Perlmutter and Carol Rosen, 38-77. Chicago: Chicago University Press. Sheldon, Diedre. 1986. Topical and non-topical participants in Galela narrative discourse. Papers in New Guinea Linguistics, no. 25, 233-48. (Pacific Linguistics A-74.) Canberra: Australian National University. Sheldon, Howard. 1991. Galela pronominal verb prefixes. Papers in Papuan Linguistics no. 1, ed. by Tom Dutton, 161-75. (Pacific Linguistics A-73.) Canberra: Australian National University. Taylor, Paul Michael. 1984. Tobelorese deixis. Anthropological Linguistics 26.102-22. 79 Van Valin, Robert D. 1990. Semantic parameters of split intransitivity. Language 66.221-60. Veen, Hendrik van der. 1915. De Noord-Halmahera'se Taalgroep tegenover de Austronesiese talen. Leiden: Van Nifterik. Visser, L.E. and C.L. Voorhoeve. 1987. Sahu-Indonesian-English Dictionary and Sahu Grammar Sketch. (Verhandelingen van het Koninklijk Instituut voor Taal-, Land- en Volkenkunde 126.) Dordrecht: Foris. Voorhoeve, C.L. 1988. The languages of the northern Halmaheran stock. Papers in New Guinea Linguistics, no. 26, 181-209. (Pacific Linguistics A-76.) Canberra: Australian National University. 80 Appendix Ø tomihononiwo- 1EXC:PAT no-mini-miwo-mi- 1INC:PAT no-nani-nawo-na- 2PL:PAT ti-nimi-niwi-ni- 3PL:PAT ta:mia:ha:na:nia:wa:- 3FEM:AGT 3PL:AGT 3AGT moyoi- mo-nai-mii-mi mo-nai-nai-na- mi-nii-nii-ni- ma:ya:ya:- 1AGT 1EXC:AGT 1INC:AGT 2AGT 2PL:AGT 3MASC:AG 1PAT no-hini-hiwo-hi- 2PAT to-nimi-niwo-ni- 3MASC:PAT timi-wihinini:wi- 3FEM:PAT to-mimi-miho-mino-mini-miwo-mi- 3PAT tamiahananiawa- mo-hii-hii-hi- mo-nii-nii-ni- mo-wii-wii-wi- mo-mii-mii-mi- mayaya- 1AGT 1EXC:AGT 1INC:AGT 2AGT 2PL:AGT 3MASC:AG T T 3FEM:AGT 3PL:AGT 3AGT Table 1. Tobelo pronominal cross-reference markers