Ensaio Teórico Cross-Linguistic Patterns of Linking VAN VALIN Jr Robert D revistadaabralin@gmail.com State University of New York at Buffalo 4 1/2 CROSS-LINGUISTIC PATTERNS OF LINKING 143 170 143-170 http://creativecommons.org/licenses/by/4.0/

This paper will present a number of cross-linguistic generalizations about argument linking framed within a conceptually based, rather than structurally based, theory of linking, Role and Reference Grammar [RRG] (Van Valin 1993, 2005, Van Valin & LaPolla 1997). It will be shown that there are a universal aspects of linking which are readily captured in this framework and that the language-specific variation is quite limited and falls within a narrow range of possibilities. The linking phenomena to be discussed in the paper include: voice alternations, direct vs. inverse verbs (e.g. German haben ‘have’ vs. gehören ‘belong to’) and applicative constructions in syntactically accusative and ergative languages. It will be argued that it is necessary to make a distinction between lexical and syntactic aspects of linking. The lexical phase of the linking is virtually universal. The syntactic phase is the locus of the contrast between ergative, accusative and other syntactic systems. Properties of different voice constructions, e.g. Japanese ‘plain’ vs. adversative passives, will be shown to be a function of whether they are part of the lexical or the syntactic phase of the linking. It will also be shown that applicative constructions in syntactically ergative languages work the same way as their counterparts in syntactically accusative languages.

Este artigo apresentará um número de generalizações inter-lingüísticas sobre a conexão de argumentos, tomando por base uma teoria conceptualmente baseada na gramática de papéis, referência e conexão [linking, Role and Reference Grammar [RRG], embora não estruturalmente baseada. Será mostrado que há aspectos universais da conexão que são prontamente capturados neste arcabouço e que a variação específica de cada língua é bastante limitada e restringe-se a um conjunto limitado de possibilidades. O fenômeno da conexão (dos argumentos) a ser discutido neste artigo inclui: alternações de voz, verbos com estrutura direta (direct verbs) e com estrutura inversa (inverse verbs) (ex. Ger man haben ‘ter’ vs. gehören ‘pertencer a’) e construções aplicativas em línguas acusativas e ergativas. Argumentar-se-á que é necessário fazer uma distinção entre aspectos sintáticos e lexicais da conexão (de argumentos). A fase lexical da conexão é virtualmente universal. A fase sintática é o lócus do contraste entre sistemas ergativo-absolutivos, nominativo-acusativos e outros sistemas. Mostrar-se-á que as propriedades das diferentes construções de voz, – ex. passivas “simples’ vs passivas adversativas, como por exemplo no japonês – são funções condicionadas pelo fato de essas propriedades pertencerem ora a uma fase lexical ora a uma fase sintática da conexão. Será também mostrado que as construções aplicativas nas línguas sintaticamente ergativas funcionam do mesmo modo que em suas contrapartes sintaticamente acusativas.

 Applicative constructions syntactic ergativity role and reference grammar linking voice constructions semantic roles lexical decomposition Introduction

The relationship between the semantic argument structure of a verb (or other predicate) and the syntactic realization of those arguments is a major topic of interest in contemporary linguistic theory. This paper presents a number of cross-linguistic generalizations about argument linking within a conceptually based, rather than structurally based, theory of linking, Role and Reference Grammar [RRG] (Van Valin, 2005; Van Valin and LaPolla, 1997)[1,2]. It will be shown that there are universal aspects of linking which are readily captured in this framework and that the language-specific variation is quite limited and falls within a narrow range of possibilities.

The discussion will proceed as follows. In section 2 the relevant aspects of RRG will be summarized. In section 3, an analysis of voice in both ergative and accusative syntactic systems will be given. In section 4 universal vs. language-specific aspects of linking will be investigated. Conclusions will be presented in section 5.

_

1. The Linking System in Role and Reference Grammar

The general organization of RRG is presented in (1).

RRG posits a direct mapping between the semantic representation of a sentence and its syntactic representation; there are no intermediate levels of representation such as ‘D-structure’ or syntactic argument structure. It is a truly ‘minimalist’ theory. The main focus in this paper will be on the semantic representation and the linking algorithm; the nature of the syntactic representation and the role of discourse-pragmatics in linking not be discussed. See Van Valin (2005)[1] for a detailed presentation.

_

1.1. The semantic representation of sentences

The semantic representation of a sentence is based on the lexical representation of the verb or other predicating element. It is a decompositional representation based on Vendler’s (1967)[3] theory of Aktionsart. The four basic classes (state, achievement, accomplishment and activity) are augmented by two additional classes, semelfactives (punctual events) and active accomplishments (telic uses of activity verbs e.g. run to the store) and by causative versions of each. Examples of the six classes are given in (2), and sentences illustrating the six classes plus their causative counterparts are given in (3).

Syntactic and semantic tests determine the Aktionsart of a clause (see VV §2.2.1). As the sentences in (3e-f´) show, a single verb, e.g. walk, can have more than one Aktionsart interpretation. This verb would be listed in the lexicon as an activity verb, and lexical rules would derive the other uses from the basic activity use (see Van Valin (in press a))[4].

The system of lexical decomposition builds on the one proposed in Dowty (1979)[5]. Unlike Dowty’s scheme, the RRG system treats both state and activity predicates as basic. The lexical representation of a verb or other predicate is termed its logical structure [LS ]. State predicates are represented simply as predicate´, while all activity predicates contain do´. Accomplishments, which are durative, are distinguished from achievements, which are punctual. Accomplishment LSs contain BECOME, while achievement LSs contain IN GR, which is short for ‘ingressive’. Semelfactives contain SEML. In addition, causation is treated as an independent parameter which crosscuts the six basic and derived Aktionsart classes, hence the twelve classes in (3). It is represented by CAUSE in LSs. The lexical representations for each type of verb in (3) are given in (4).

Examples of simple English sentences with the LS of the predicate are presented in (5).

Full semantic representations of sentences also contain lexical representations of the NPs, adjuncts, and grammatical operators like tense and aspect; see Van Valin (2005, §2.2, 2.3)[1].

_

 1.2 Semantic macroroles and lexical entries for verbs

The semantic interpretation of an argument is a function of its position in the LS of the predicate, and, as will be seen below, the linking system refers to an element’s LS position. Thematic relations as such play no role in the theory; the traditional thematic role labels are used only as mnemonics for the LS argument positions, e.g. ‘theme’ is the mnemonic for the second position (y) in a two-place locational LS like be-at´ (x, y). RRG posits two generalized semantic roles or SEMANTIC MACROROLES, which play a crucial role in the linking system. The two macroroles are ACTOR and UNDERGOER, and they are the two primary arguments of a transitive predication; the single argument of an intransitive predicate can be either an actor or an undergoer, depending upon the semantic properties of the predicate. The basic distinction is illustrated in the following Brazilian Portuguese examples.

In (6a), María is the actor and a janela ‘the window’ is the undergoer of the transitive verb fechar ‘sell’; in the sentences with intransitive verbs, María is an actor with the activity verb cantar ‘sing’ and an undergoer with the accomplishment verb morrer ‘die’. Actor is not equivalent to syntactic subject, nor is undergoer equivalent to syntactic direct object, as the examples in (6c) and crucially (6d) show: in both of these sentences the syntactic subject is an undergoer, and in the passive sentence in (6d) the actor is an oblique adjunct. In an English clause with an active voice transitive verb, the actor is the initial NP (the traditional subject) and the undergoer is the direct NP immediately following the verb.

Actor and undergoer are generalizations across specific semantic argument types, as defined by LS positions. This is illustrated in (7).

The x argument of all of these verbs functions as the actor, regardless of whether it is the first argument of the generalized activity verb do´ (conventionally labeled ‘effector’), as with kill, put and present, or the first argument of a two-place state predicate, as with see. With two-place transitive verbs like kill and see, the y argument is the undergoer. With three-place verbs like put and present, the situation is potentially more complex. Put allows only the z argument to be undergoer (in (5a, e) with locational predicates the first argument is the location and the second argument is the located entity), as in Chris [x, actor] put the book [z, undergoer] on the table [y]. Present, on the other hand, permits either the y or the z argument to function as undergoer; these two possibilities are given in (8); (7) depicts (8b), with the y argument as undergoer.

The relationship between LS argument positions and macroroles is captured in the Actor-Undergoer Hierarchy [AUH] in (9).1

The basic idea of the AUH is that in a LS the leftmost argument in terms of the hierarchy will be the actor and the rightmost will be the undergoer. This was true for kill, see and put in (7). It was not true for present, however, and this reflects a fundamental asymmetry in the AUH: the leftmost argument in a LS (in terms of the AUH) is always the actor, but the rightmost argument is only the default choice for undergoer. There are languages in which undergoer selection follows a different principle, so-called ‘primary object languages’; in such languages, the second highest ranking argument is chosen as the undergoer. See Guerrero and Van Valin (2004)[6] for detailed discussion.

Transitivity in RRG is defined semantically in terms of the number of macroroles a predicate takes. This is termed ‘M-transitivity’ in RRG, following Narasimhan (1998)[7], in order to distinguish it from the number of syntactic arguments a predicate takes, its ‘S-transitivity. The three M-transitivity possibilities are: transitive (2 macroroles), intransitive (1 macrorole), and atransitive (0 macroroles). The principles determining the M-transitivity of verbs are given in (10).

These principles correctly predict the transitivity of the verbs from Brazilian Portuguese and English discussed above. (The LSs for the Brazilian Portuguese verbs are [do´ (x, Ø)] CAUSE [BECOME closed´ (y)] for fechar ‘close’, do´ (x, [sing´ (x)]) for cantar ‘sing’, and BECOME dead´ (x) for morrer ‘die’.) Fechar ‘close’, like English kill, see, put and present, has at least two arguments in its LS, and therefore it is transitive, following (10a1). Cantar ‘sing’ and morrer ‘die’ both have only one argument in their LSs, and accordingly they are intransitive, following (10a2). If a verb takes only one macrorole, then the principles in (10b) come into play. Cantar ‘sing’ has an activity predicate in its LS, and therefore its macrorole is actor, following (10b1). Morrer ‘die’, on the other hand, has no activity predicate in its LS, and consequently its macrorole is undergoer, following (10b2). If a verb is irregular and has exceptional transitivity, it will be indicated in its lexical entry by ‘[MR<]’, where ‘<’ is a variable for the number of macroroles. Examples of lexical entries for some English verbs are given in (11).

A major claim in RRG is that no syntactic subcategorization information of any kind is required in the lexical entries for verbs. For regular verbs, all that is required is the LS and nothing more, as in all except (11d). For most irregular verbs, only the macrorole number needs to be specified2. The prepositions that mark oblique arguments with verbs like show are predictable from general principles and need not be listed in the lexical entry (see Jolly, 1993;[8] Van Valin and LaPolla (1997)[2], §7.3.2). All of the major morphosyntactic properties of verbs and other predicates follow from their LS together with the linking system.

_

 1.3 Syntactic functions, case and agreement

The linking between semantics and syntax depicted in (1) has two phases: first, the determination of semantic macroroles based on the LS of verb or other predicate in the clause, and second, the mapping of the macroroles and other arguments into syntactic functions. The traditional grammatical relations have no theoretical status in RRG; rather, RRG posits a single, construction-specific grammatical relation, which is termed the PRIVILEGED SYNTACTIC ARGUMENT [PSA] of the construction.3 It subsumes the notion of ‘syntactic pivot’ used in Dixon (1972, 1979, 1994)[9-11] and in earlier work in RRG. The non-PSA syntactic arguments in the clause are referred to as direct or oblique core arguments. The PSA for most (but not all) English constructions is the traditional subject. Languages have selection hierarchies to determine the PSA; the ones for syntactically accusative and ergative languages are given in (12).

For a language like English, (12a) captures the fact that in an active voice clause with a transitive verb, the actor is the PSA, whereas for a language like Dyirbal (Dixon 1972)[9], in an active voice clause with a transitive verb the undergoer is the PSA, following (12b). Languages also differ with respect to whether the PSA must be a macrorole: German, Italian, Dyirbal, Jakaltak (Mayan) and Sama (Philippine) restrict PSA selection to actors and undergoers only, while Icelandic, Georgian, Japanese, Korean and Kinyarwanda allow non-macrorole direct core arguments to function as PSA (see Van Valin, 1991;[12] Van Valin (2005)[1], §4.2). The linking system is summarized in (13).

The technical details of the linking algorithm are developed in Van Valin (2005)[1], chapter 5 presents the linking algorithm for simple sentences, while chapter 9 presents the one for complex sentences. Both (1) and (13) contain double-headed arrows; this means that the linking system not only maps semantic representations into syntactic representations, but it also maps syntactic representations into semantic representations. This is, after all, part of what language users must do when they are producing and comprehending speech4. The emphasis in this paper is solely on the semantics to syntax mapping.

Case and agreement rules are formulated with reference to the linking system. Examples of case and agreement rules are given in (14) and (15).

The operation of these rules can be illustrated using the following examples from German in (16a) and Dyirbal in (16b); the LS for verbs meaning ‘give’ is presented in (16c).5

In the German example, x = Frauen ‘women’, y = Mann ‘man’, and z = Hut ‘hat’; in the Dyirbal sentence, x = dyugumbil ‘woman’, y = ya≠a ‘man’, and z = mirañ ‘beans’. The first step in the linking is to assign macroroles. The x argument is the leftmost in terms of the AUH, and therefore the nouns meaning ‘woman/women’ will be the actor in each clause. The z argument is the rightmost, and therefore Hut ‘hat’ and mirañ ‘beans’ will be the undergoer. The third argument, the y argument meaning ‘man’, will be a non-macrorole core argument. Following the PSA selection hierarchy in (12), the actor Frauen ‘women’ will be the PSA in German, since it is a syntactically accusative language, and the undergoer mirañ ‘beans’ will be the PSA in Dyirbal, since it is a syntactically ergative language. The case rules in (14) are relevant now. For German, the rules in (14a) require that the actor, the highest ranking core macrorole, appear in the nominative case, that the undergoer, the other macrorole, appear in the accusative case, and that the non-macrorole core argument, Mann ‘man’, appear in the dative case. For Dyirbal, the rules in (14b1) are relevant, because all of the arguments are third person. The lowest ranking macrorole, the undergoer, must appear in the absolutive case, while the other macrorole, the actor, appears in the ergative case; finally, the third argument, ya≠a ‘man’, which is not a macrorole, appears in the dative case. The single macrorole argument of an intransitive verb is both the highest ranking and the lowest ranking, and therefore it will appear in the nominative case in German and in the absolutive case in Dyirbal. German, but not Dyirbal, has finite verb agreement, and the rule in (15a) states that the finite verb must agree with the highest ranking macrorole, which is the actor in this sentence. Thus, the basic morphosyntactic properties of the German and Dyirbal sentences in (16) are readily captured by the lexical representation of the verbs in (16c) and the principles in (12)-(15) from the RRG linking system.

An important problem is posed by verbs which are very similar in meaning or fall into the same semantic class but which exhibit very different morphosyntactic properties, e.g. English own (e.g. I own the house) vs. belong to (e.g. The house belongs to me), or German mögen ‘like’ (e.g. Ich mag das Haus ‘I [NOM] like the house [ACC]’) vs. gefallen ‘please’ (e.g. Das Haus gefällt mir ‘The house [NOM] pleases me [DAT]’). The RRG analysis of belong (to) and gefallen ‘please’ is that they are M-intransitive; they are exceptional in that despite having two arguments in their LS, they take only one macrorole. Hence this would have to be marked in their lexical entry, as in (11c) for belong (to). This is all that needs to be stated regarding these verbs, as all of their other properties fall out from the linking system. In order to illustrate this, comparable examples from will be used; they involve the verbs telja ‘believe’ and óykja ‘think, consider’, as in (17).

Telja ‘believe’ is a regular transitive verb with the LS believe´ (x, y); the x argument is the actor, and the y argument is the undergoer. The actor is the PSA, following (12a), and it appears in the nominative case, following (14a); the finite verb also agrees with it, following (15a). The undergoer, the other macrorole, appears in the accusative case. In contrast, óykja ‘consider’ is M-intransitive and has the LS consider´ (x, y) [MR 1]. This specification says only that there is one macrorole but not which one it is; that follows from (10b), which assigns undergoer, due to the lack of an activity predicate in the LS. The y argument is the undergoer, following the AUH, leaving the x argument as a non-macrorole core argument. Icelandic is one of the languages which does permits non-macrorole core arguments to function as PSA, and following (12a), the x argument is the PSA. This is the correct assignment, because the dative NP in (17b) is the argument that can be ‘raised’, can control a reflexive, etc.; the nominative NP cannot. With respect to case and agreement, the undergoer is the only (and therefore the highest ranking) macrorole, and consequently it appears in the nominative case; the non-macrorole core argument appears in the dative case, following (14a3). The finite verb agrees with the highest ranking macrorole, hence with óeir ‘they’, and the predicate adjective agrees with the undergoer, which the nominative NP, according to (15b). The linking in these two sentences is presented in a simplified form in (18).

Hence all of the morphosyntactic properties of (17b) follow from the LS plus the linking system; verbs like óykja ‘consider’ in Icelandic, gefallen ‘please’ in German and belong (to) in English are completely regular except for their transitivity.6

_

 2. Voice

All of the discussion in the previous section concerned active voice clauses with transitive verbs, and the PSA selection hierarchies in (12) specify only the default selections. In many languages (probably most), the default linkings are the only linkings possible, but in some languages the default can be overridden and a different argument can be selected to function as PSA. In an accusative language like German, it is possible for the other macrorole argument of a transitive verb, the undergoer, to function as PSA in a passive construction; similarly, in a syntactically ergative language like Dyirbal, it is possible for the actor of a transitive verb to function as PSA in an antipassive construction. This is illustrated in (19).

In (19) the default PSA selections have been overridden, yielding passive and antipassive clauses, respectively. The case and agreement rules apply as before. In these constructions there is only one core macrorole, and therefore it is the highest ranking (German) or the lowest ranking (Dyirbal). Hence the undergoer Hut ‘hat’ appears in the nominative in German, and the actor dyugumbil ‘woman’ appears in the absolutive in Dyirbal. The case marking of ‘man’ is the same as in (16), because it is still a non-macrorole core argument. The default choices for PSA are treated as oblique adjuncts in these constructions, appearing in a PP headed by von in German or in the instrumental case in Dyirbal. The finite verb agrees with Hut ‘hat’ in German, as it is the highest ranking core macrorole, following the rule in (15a). The situation with respect to PSA selection and voice in these two systems is summarized in (20).7

_

 3. Universal vs. Language-specific Aspects of Linking

The summary of the RRG linking system in (13) is divided into two parts, one of which is labeled ‘universal’ and the other ‘language-specific’. A major claim of RRG is that what may be called the ‘lexical’ phase of the linking, the determination of macrorole assignments based on the LS of the verb, is virtually universal and subject to only very limited cross-linguistic variation, while the second phase, which may be termed the ‘syntactic’ phase, the mapping of the macroroles and other arguments into the syntax, is subject to considerable cross-linguistic variation, e.g. the contrast between accusative and ergative syntactic systems. The variation in the lexical phase concerns (i) whether languages allow variable undergoer assignment of the type discussed in §2.2, and (ii) the role of animacy in macrorole assignment (see Van Valin and LaPolla (1997)[2], §7.4.1).

Most linking theories go directly from thematic relations to grammatical relations and posit universal constraints on the linking like ‘highest ranking thematic relation links to subject’. Approaches such as this ignore ergative systems and split-intransitive systems like Acehnese.8 The RRG system, in contrast, has two steps, (1) LS to macroroles and (2) macroroles to syntactic functions.9 By factoring the linking in this way, it is possible to capture significant generalizations that the first approach cannot. In order to illustrate the kind of generalization that the RRG approach can capture, applicative constructions in syntactically accusative and syntactically ergative languages will be examined. Applicative constructions have been traditionally characterized as involving the appearance as direct object of an argument that would normally appear as an indirect or oblique object. Applicatives in two accusative languages, Indonesian (Dardjowidjojo, 1971)[13] and Swahili (Driever, 1976; Vitale 1981)[14,15] are presented in (21) and (22); analogous constructions from two syntactically ergative languages, Sama (Walton, 1986)[16] and Dyirbal, are given in (23) and (24).

10

11

Standard (i.e. thematic relations map directly to grammatical relations) analyses of applicative constructions postulate that they involve an alternative choice for direct object. The theme is the default choice in (21a) and (22a), and only the direct object of an active verb can function as the subject in a passive construction in these languages, as shown in (21a´, a´´) and (22a´). The recipient is the direct object in the applicative constructions in (21b) and (22b), and it, not the theme, can serve as the subject of the passive versions of these sentences, as shown in (21b´, b´´) and (22b´, b´´). The RRG analysis in terms of macroroles would claim that what is involved here is not an alternative choice for ‘direct object’ but rather an alternative choice for undergoer. In these languages, undergoer is the default choice for ‘direct object’ and is the only argument that can serve as the subject of a passive construction.

With respect to syntactically accusative languages like Indonesian and Swahili, these two analyses seem to be equivalent, but they make very different claims with regard to syntactically ergative languages like Sama and Dyirbal. (In Sama only pronouns are case-marked; full NPs are marked as oblique by the preposition ma.)

The absolutive NP is the syntactic subject for many constructions in these languages,12 and the addition of the applicative morpheme (-an in Sama13 and -mal in Dyirbal) results in an alternative choice for the absolutive NP, hence an alternative choice for subject, not direct object as in (21) and (22). For the thematic-relations-to-grammatical-relations theories, it appears that two accounts are required for the languages in (21)-(24): in syntactically accusative languages, applicative constructions yield an alternative choice for direct object, while in syntactically ergative languages, they result in an alternative choice for subject. The RRG analysis, on the other hand, applies equally to all four languages. Applicative constructions involve an alternative undergoer choice (or, in the case of Dyirbal, adding an undergoer argument to the clause), and this is true of each of the languages. The differences among them arise from the linking between macroroles and syntactic functions summarized in (20): in Indonesian and Swahili the undergoer is the ‘direct object’ in an active-voice clause and PSA in a passive clause, whereas in Sama and Dyirbal it is the PSA in an active-voice (i.e. non-antipassive) clause. Thus, the RRG analysis captures an important generalization about applicative constructions that analyses positing only thematic relations and grammatical relations miss.

It should be noted the Dyirbal also has an alternation like that in (8); in addition to the pattern in (16b) with wugal ‘give’, which is analogous to (8a) in that the theme is the undergoer, it is also possible to have the recipient as the undergoer, as in (25), which is analogous to (8b).

In terms of the LS in (16c), ya≠a ‘man’ is the y argument and is the undergoer; because (25) is active voice, the undergoer is the PSA. In RRG terms, (25) is like (8b) and English dative shift in that there is a marked linking to undergoer in terms of the AUH. Where English and Dyirbal differ, of course, is in the mapping of macroroles to syntactic functions. The identity of the alternation in the two languages would be hidden in the standard analysis, since the grammatical relation of the recipient in (25) would not be the same as the grammatical relation as in (8b) or as in English dative shift. Here again a significant generalization is captured in the RRG analysis but not in the standard account.

There is a principled reason for why the lexical phase of the linking is universal. The Aktionsart distinctions underlying the lexical decomposition are universal; all languages exhibit them. The notions of actor and undergoer are likewise valid for all languages, and the relationship between macroroles and LSs is governed by the principles in (10) and the AUH, which are likewise universally valid. Thus, all of the components of this phase of the linking are universal. It is in the syntactic phase that considerable cross-linguistic variation is found, and it is only because the linking has been factored into two phases, one lexical and the other syntactic, that the truly universal aspects can be identified.

_

 4. Conclusion

In this paper it has been shown that universal patterns of linking can be identified using Role and Reference Grammar, a framework in which the linking between semantics and syntax is factored into two phases, a lexical phase and a syntactic phase. The universal patterns are found in the lexical phase of the linking, while the major cross-linguistic variation is located in the syntactic phase.14 These generalizations could not be readily captured in frameworks which posit a direct mapping from thematic relations to grammatical functions.

The RRG linking system also demonstrates how it is possible to dispense completely with syntactic subcategorization information in the lexical entries for verbs. Given the LS of the verb and the linking principles, it is possible to derive the morphosyntactic properties of most verbs without any further information. In the case of irregular verbs, the M-transitivity must be specified in the lexical entry, but this is semantic, not syntactic information in RRG. As the discussion of pairs of verbs like own and belong (to) in English and mögen ‘like’ and gefallen ‘please’ in German showed, quite striking differences in morphosyntactic behavior can follow from a simple difference in M-transitivity. Thus, RRG provides a theoretical and descriptive framework in which linguistically significant generalizations about the relationship between the semantic argument structure of the verb and the morphosyntactic realization of the arguments can be readily captured and explained.

'Do, represents the lexicalization of agentivity with verbs, Hence its first argument is the traditional agent. With many verbs, agentivity is an implicature rather than a lexical property; see Holisky (1987)[17] , Van Valin and Wilkins (1996)[18] and Van Valin and LaPolla (1997)[2], §3.2.3.2 for detailed discussion.

When systmatic exceptions to the principles in (10) exist, e.g. activity verbs are universally M-intransitive as a default (Van Valin, 1990;[19] Van Valin and LaPolla (1997)[2], §3.2.3.3), then this can be stated as a general principle, and individual lexical entries would not be marked with this feature. See Van Valin and LaPolla (1997)[2], §4.2 for detailed discussion. It should also be noted that syntactic subcategorization information is a stipulation of transitivity, and most theories include this information for all verbs. No such information is included in the lexical entries in (11), and the macrorole number is specified only for irregular verbs which are not subsumable under the kind of generalization mentioned above.

See Van Valin (2005)[1], Chapter 4, for a datailed presentation of the RRG theory of grammatical relations

See Van Valin (in press b)[20] for discussion of how RRG's linking system funcions in neurocognitive models of language processing.

Abbreviations used: ACT 'actor', ABS 'absolutive', ACC 'accusative', ACV ‘active voice’, ANTI ‘antipassive’, APL ‘applicative’, DAT ‘dative’, DCA ‘(non-macrorole) direct core argument’, DEF ‘definite’, ERG ‘ergative’, IND ‘indicative’, IN ST ‘instrumental’, LOC ‘locative’, LS ‘logical structure’, NM ‘noun marker’, OBL ‘oblique’, PASS ‘passive’, PAST ‘past tense’, pl ‘plural’, PM ‘proper noun marker’, PRES ‘present tense’, PSA ‘privileged syntactic argument’, sg ‘singular’, TNS ‘tense’, UND ‘undergoer’, 1, 9 ‘Swahili noun classes’.

For a detailed comparison of this analysis with other accounts, see Van Valin (1991)[12], Van Valin and LaPolla (1997)[2], §7.3.1.1.

For analyses of the other systems as well as a typologi of voice functions, see Valin and LaPolla (1997)[2], §§6.3-4, 7.3.1. See also Roberts (1995)[21].

Acehnese lacks syntactic grammatical relations and operates with a semantically-based system in which syntactic rules and restrictions are stated in terms of actor and undergoer rather than subject and direct object; see Durie (1987)[22], Van Valin (1993)[23] and Van Valin (2005)[1], §4.1 for detailed arguments.

Manning (1996)[24] follows an RRG- type approach but operates with a different factorization: (1) thematic relations to argument structure [ARG-S] , and (2) ARG-S to syntactic relations. The highest ranking thematic relation is the ARG-S subject, the other argument the ARG-S object. Languages differ in terms of how ARG-S relations map into syntactic relations: in accusative systems, the ARG-S subject = syntactic subject, while in ergative systems, the ARG-S object = syntactic subject. While this appears to be similar to the RRG system, ARG-S is explicitly claimed to be syntactic, not semantic; hence ARG-S subject is not equivalent to actor and ARG-S object is not equivalent to undergoer. This difference manifests itself clearly with respect to languages like Acehnese, which Manning does not discuss; it is not at all obvious that his approach can deal with a language of this type, while it presents no problems for RRG (see references in footnote 8)

Strictly speaking, RRG does not employ the notion of direct object, but it will be used here to facilitate the comparison between the two analyses. See Van Valin (1993)[23], Van Valin (2005)[1], §4.5 for discussion of how ‘direct object’ phenomena are handled in RRG.

'%' means that the PP kwa Juma is not acceptable to all Swahili speakers.

Bittner and Hale (1996)[25] argue that the actor is the subject in Diyrbal based on a single construction. This conclusion is dubious, for two reasons. First, no language is completely ergative syntactically, and consequently the existence of non-ergative constructions is to be expected. The majority of constructions in Dyirbal do operate ergatively, and this cannot be overlooked. Second, the construction in question, manner adverbials which are verbal in nature and require the same actor as the verb they modify, can be given a semantic analysis, unlike conjunction reduction (topic chaining) and relativization, the major ergative constructions, and hence it is irrelevant to the determination of grammatical relations in Dyirbal. See Dixon (1979, 1994)[10,11] for arguments as to why such semantically determined constructions are irrelevant to the syntactic type of a language.

The applicative merkar -an in Sama is cognate with the applicative marker in Indonesian, -kan

For an account of this variation, and in particular why certain linking patterns are more frequent than others, see Van Valin and LaPolla (1997)[2], §6.5

 531 604 27 1996 BITTNER Maria HALE Ken Linguistic Inquiry Ergativity: toward a theory of a heterogeneous class. 71 84 2 1971 DARDJOWIDJOJO S Philippine Journal of Linguistics The men-, men-kan and men-i verbs in Indonesian Cambridge Cambridge University Press, 1972 DIXON R M W The Dyirbal language of north Queensland 59 138 55 1979 DIXON R M W Language Ergativity Cambridge Cambridge University Press 1994 DIXON R M W Ergativity Dordrecht Reidel 1979 DOWTY David Word meaning and Montague Grammar Hamburg Helmut Buske Verlag 1976 DRIEVER Dorothea Aspects of a case grammar of Mombasa Swahili 365 399 11 1987 DURIE Mark Studies in Language Grammatical relations in Acehnese 290 319 70 2004 GUERRERO Lilián VAN VALIN Robert D. Jr. International Journal of American Linguistics Yaqui and the analysis of primary object languages 103 132 71 1987 HOLISKY Dee A Lengua The case of the intransitive subject in Tsova-Tush (Batsbi) 275 310 Amsterdam/Philadelphia John Benjamins 1993 JOLLY Julia VAN VALIN R Advances in role and reference grammar Preposition assignment in English Stanford CSLI 1996 MANNING Christopher Ergativity: argument structure and grammatical relations 48 76 52 1998 NARASIMHAN Bhuvana Studia Lingüística A lexical semantic explanation for ‘quirky’ case marking in Hindi 157 214 15 1995 ROBERTS Linda Australian Journal of Linguistics Pivots, voice and macroroles: from Germanic to universal grammar 221 260 66 1990 VAN VALIN Robert D. Jr. Language Semantic parameters of split intransitivity 145 194 9 1991 VAN VALIN Robert D. Jr. Natural Language and Linguistic Theory Another look at Icelandic case marking and grammatical relations 1 164 Amsterdam/Philadelphia John Benjamins 1993 VAN VALIN Robert D. Jr. ________ Advances in role and reference grammar A synopsis of role and reference grammar Cambridge Cambridge University Press 2005 VAN VALIN Robert D. Jr. Exploring the syntax-semantics interface Dordrecht Kluwer (in press a) VAN VALIN Robert D Jr PUSTEJOVSKY James New developments in the generative lexicon Lexical representation, co-composition, and linking syntax and semantics Berlin Mouton de Gruyter (in press b) VAN VALIN Robert D., Jr BORNKESSEL SCHLESEWSKY Matthias Semantic role universals and argument linking: Theoretical, typological and psycho/neurolinguistic perspectives Semantic macroroles and language processing Cambridge Cambridge University Press 1997 VAN VALIN Robert D., Jr. LAPOLLA Randy J Syntax: structure, meaning & function 289 322 Oxford Oxford University press 1996 VAN VALIN Robert D., Jr WILKINS David P SHIBATANI M THOMPSON S Grammatical constructions The case for ‘effector’: Case roles, agents and agency revisited Ithaca, NY Cornell University Press 1967 VENDLER Zeno Linguistics in philosophy Dordrecht Foris 1981 VITALE Anthony Swahili syntax Manila Linguistic Society of the Philippines 1986 WALTON Charles Sama verbal semantics: classification, derivation and inflection