This paper is, in the words of the Call for Papers of this conference, “in the spirit of LFG.” The claims that I will be making were originally conceived of in a lexically-oriented version of GB. I have not made the effort to see how it would work in the formalism of LFG. However, the ideas are independent of the theoretical framework, and I believe that they are very congenial to the LFG perspective, under which grammatical sentences are ones that meet conditions at distinct types of representations simultaneously.

In Case, as with everything else, the goal of a generative theory is to determine what raw materials Universal Grammar provides for individual languages to use. This requires a broad typological perspective; focusing on a narrow class of languages will often fail to provide the proper basis. It is therefore unfortunate that much theorizing about Case is based almost exclusively on nominative-accusative languages, with ergative languages, if considered at all, added on as an afterthought. A theory of Case must be built from the ground up to consider facts in both nominative-accusative and ergative languages.

My point of departure is the phenomenon known as split ergativity: the phenomenon under which ergative languages display some patterning that looks nominative-accusative. The phenomenon is pervasive among ergative languages, but its importance has, in my opinion, not been fully appreciated. The phenomenon of split ergativity shows that, contrary to the usual description, the only real difference between nominative-accusative languages and ergative languages is the existence of ergative Case. Both language types have accusative Case. However, accusative and ergative Case do not always surface; under some conditions a form unmarked for Case (traditionally called the nominative, or, in ergative languages, the absolutive) appears instead. Following more-or-less traditional views on the subject, and contrary to most modern theoretical approaches, I assume that the unmarked Case is literally unmarked; that is, that nominative and absolutive nominals are actually lacking in Case marking.1 This view of the moral of split ergativity contradicts the conventional wisdom, according to which one argument of a transitive clause must be marked with morphological Case to avoid ambiguity. Under this conventional view, the difference between nominative-accusative languages and ergative languages lies in which argument they choose to mark: nominative-accusative languages choose the OBJ while ergative languages choose the SUBJ. Such an approach has its origins in the work of typologists such as Dixon (1994). It is translated into a theoretical approach in several recent studies, most notably Bittner and Hale (1996a), in which Case marking is a consequence of Case competition. Unconventional though it may be, my approach seems unavoidable if the syntax of the language is to account for all the Cases that are available in the language. It is reinforced by the existence of so-called three-way languages, in which ergative and accusative Cases are always  present in transitive clauses.

If my approach is correct, the crucial question becomes what controls the appearance of ergative and accusative Case. I will argue that it is not syntax, but discourse grammar. Overt morphological Case is thus an example of a superficially syntactic phenomenon that turns out to be based on constraints in parallel levels of representation.

1. Split Ergativity Exemplified

2. The Discourse Component

2.1. Ergative Languages

An answer to this question is forthcoming by examining the cross-linguistic distribution of ergativity splits. Typological studies, such as Dixon (1994) and references cited there, point to several factors involved. Perhaps the most important is animacy: the more animate a nominal the more likely it is to be marked with accusative Case and the less likely to be marked ergative. To a lesser extent, definiteness/specificity3 is also involved.  The effect of animacy can be clearly seen in Dyirbal. As noted earlier, pronouns (which can only refer to animate beings) display a nominative-accusative contrast. In other words, pronoun OBJs are accusative (unlike ordinary nominals, which are “absolutive”) and pronoun SUBJs are not ergative (again, unlike ordinary nominals). This corresponds to the general trend referred to above. In addition, as noted earlier, animate nominals in OBJ position can optionally be marked with an accusative suffix, and ‘who’ obligatorily displays a three-way contrast. Dixon (following earlier work by Silverstein) proposes the following animacy hierarchy.
 

(3)	1st person pronoun  2nd person pronoun  3rd person pronouns; demonstratives  proper nouns  human common nouns  animate common nouns  inanimate common nouns

 
Other ergative languages display a similar pattern. For example, in Nyawaygi (Dixon 1983), accusative Case shows up on animate pronouns, and sporadically on animate nouns.
 

(4)	Nad,a	ñaNga	muyma-	ña	yu:Rimayi.
	I.ERG	3SG.ABS	boy-	ACC	grow.up.COMIT.PERF
	‘I brought up the boy.’

 
In Djapu (Morphy 1983) high animacy nominals in OBJ position are marked accusative, while low animacy nominals are not. Waga-Waga (Dixon 1994) marks all transitive SUBJSs ergative, but it only marks OBJs accusative if they are pronominal, proper, or common with human reference. In Dhalanji (Dixon 1994), first person pronouns do not have ergative marking (all other nominals are overtly three-way);  Animacy effects are not limited to Australia, either. Eastern Pomo (McLendon 1978) has an ergative system for nouns and a more-or-less accusative system for pronouns4. In other words, the (less animate) nouns have ergative Case but not accusative and the (more animate) pronouns have accusative but not ergative. In Hindi and related languages, OBJs are marked with the accusative postposition when animate and/or definite;5 otherwise, they are unmarked.6
 

(5)	a.	laRkiyõ	ne	rooTii	khaaii.
		girls	ERG	bread(F.SG)	ate.F.SG
		‘The girls ate the bread.’

 

	b.	laRkiyõ	ne	rooTii	ko	khaayaa.
		girls	ERG	bread	ACC	ate.M.SG
		‘The girls ate (the) bread.’

In the Papuan language Fore, ergative Case appears only if the SUBJ is lower on the animacy hierarchy than the OBJ (Blake 1994: 123).
 

(6)	a.	Yagaa	wá	aegúye.
		pig	man	3SG.kill.3SG
		‘The man kills the pig.’

 

	b.	Yagaa-	wama	wá	aegúye.
		pig-	ERG	man	3SG.kill.3SG
		‘The pig kills the man.’

In the Maku language Nadëb (Dixon 1994), only first person pronouns have accusative Case, while all nominals have ergative Case.

Animacy also sometimes affects the nature of agreement. For example in the Australian language Rembarnga, the OBJ agreement marker is followed by an accusative marker only if the OBJ is higher on the animacy hierarchy than the SUBJ (Mallinson and Blake 1981).
 

(7)	a.	Na-	n-	pa-	na.
		me-	ACC-	they-	saw
		‘They saw me.’

 

	b.	pa-	Na-	na.
		they-	me-	saw
		‘I saw them.’

In Ndjebbana, first person singular agreement is nominative-accusative while second person singular is ergative (Dixon 1994). Again, similar effects are evident in languages spoken outside of Australia. In the Straits Salish language Lummi (Jelinek 1993), sentences with third person SUBJs and OBJs exhibit an ergative agreement (or incorporated pronoun) structure: OBJ agreement is on the aspectual marker outside the VP and SUBJ agreement is on the verb. Sentences with non-third person SUBJs display a nominative-accusative system, with SUBJ agreement outside the VP and OBJ agreement on the verb.
 

(8)	a.	leN-	t-	s	lE’-	.Æ.
		see-	TRANS-	3ERG	PST-	3ABS
		‘He saw him.’

 

	b.	leN-	t-	oNEL	lE’-	sxw.
		see-	TRANS-	1plACC	PST-	2sgNOM
		‘You saw us.’

An interesting case that has been mentioned in the literature is the Sahaptian language Nez Perce (Rude 1986; Woolford 1997). What makes it interesting is that, unlike in the languages discussed above, in Nez Perce either ergative and accusative Case both surface or neither surfaces. In a sentence where accusative Case does not surface, the OBJ also does not control agreement. This is illustrated in the following examples from Woolford.
 

(9)	a.	háama-	nm	pée-	wiye	wewúkiye-	ne.
		man-	ERG	3SUBJ.3OBJ-	shot	elk-	ACC
		‘The man shot an elk.’

 

	b.	háama	hi-	wiye	wewúkiye.
		man	3SUBJ-	shot	elk
		‘The man shot an elk.’

A sentence in which one of the Cases is overt and the other is not is ungrammatical. This state of affairs has been analyzed in different ways in the literature. Rude analyzes the form with no overt Case as an antipassive, but as Woolford points out this is a problematic analysis. The Caseless sentence has no antipassive morpheme and the OBJ is not marked with an oblique marker. A more straightforward analysis is to simply consider the b. example to be a sentence in which neither Case is realized overtly (i.e. both SUBJ and OBJ are nominative). Woolford's own analysis is that this is an example of a four-way Case contrast. She calls the Case on the OBJ in (9a) objective and the (phonologically null) one in (9b) accusative. However, nowhere does she give any evidence that the null marking on the OBJ is distinct from the null marking on the SUBJ.7 Woolford also argues for her analysis from the ungrammaticality of combining an ergative subject with an “objective” (overt accusative) object. She claims that this derives from a version of Burzio's Generalization that she develops. However, other languages we have discussed raise serious doubts about a UG account like the one she proposes. In Dyirbal, for example, all four possible Case marking combinations for SUBJ and OBJ are grammatical.  We will not propose an account of the mutual dependence of ergative and accusative Case. The crucial point from our perspective is a fact that Woolford relegates to a footnote: “Rude argues that there is a difference in discourse function [between the two constructions].” According to Rude (1986), in the construction with marked arguments and OBJ agreement the OBJ is much more specific and topical than in the construction with unmarked arguments and no OBJ agreement. The construction with marked arguments can also be used if the SUBJ is nonreferential (and thus very low in specificity/definiteness). This correlates with the general typological observation.

Further examples of the effects of animacy (and definiteness) abound in the literature. There are two potential ways to account for these animacy- and definiteness-based ergativity splits: syntactic and extra-syntactic (discourse). In the typological literature there is a consensus around a discourse grammar explanation. For example, Dixon (1979, 1994) attributes the animacy effect to the discourse nature of SUBJs and OBJs. Transitive sentences in discourse are typically about people performing actions on things. Thus, the prototypical SUBJ is animate and the prototypical OBJ is inanimate. (See also Givón's 1995 discussion of the prototypical transitive event.) In addition, the SUBJ is usually old information while new discourse entities are often introduced as OBJs. The prototypical SUBJ is therefore definite/specific while the prototypical OBJ is indefinite/nonspecific. There are other types of considerations that lead to the same basic conclusions about prototypical SUBJs and OBJs. For example, in Jackendoff's (1990) theory of Conceptual Semantics, transitive verbs have an Actor-Patient semantic structure and Actors are prototypically volitional. Volitionality implies animacy. From whatever angle one approaches it, the idea is that SUBJs and OBJs that are unmarked for Case (i.e. absolutive or nominative) are relatively canonical SUBJs and OBJs. The ones that are marked (in languages that do not mark all) deviate from the prototypical pattern.

Jelinek (1993) objects to the discourse theoretic account of animacy- and definiteness-based splits, and offers a syntactic account instead. Her objections to the discourse (or “pragmatic,” as she calls it) account is twofold. In the first place, she claims that such an analysis cannot explain the fact that these splits are found only in languages with rich agreement. Second, if a pragmatic account is based on a notion of a scale or hierarchy there should be gradations of judgments rather than the sharp grammatical judgments that one finds. Alternatively, if the pragmatic account is based on general cognition, all languages should display the same split. Neither of these objections is convincing. The alleged typological generalization concerning agreement is not true; Dyirbal, for example, has no agreement, yet it displays exactly this kind of split. The claim about pragmatics in general is also not true; if discourse grammar is a legitimate part of the grammar, then it will have formal structure just as syntax does and will allow for parametric variation. The fact that discourse grammar has not yet been formalized to the same degree as sentence grammar (syntax) is a fact about linguistics and linguists, not about language. It is of no principled importance in linguistic analysis.8

Jelinek's explanation of animacy- and definiteness-based splits is based on the theory of Diesing (1992), according to which at LF definite nominals are outside VP and indefinite nominals are inside. Assuming that nominative and absolutive nominals are outside VP, she argues that since first and second persons are always definite, they must be raised out of the VP by LF. In languages with an ergativity split, this must occur before S-structure (or Spell-out).

The empirical evidence casts doubt on Jelinek's account and is more conducive to a discourse grammar account. First, as the examples above show, the split is not always between first and second person on one hand and third person on the other.9 It can be between first and second person (as in Dhalanji, Nadëb, and Ndjebbana), and it can distinguish between third person nominals on grounds of animacy (Djapu, Eastern Pomo, Fore, Rembarnga, Waga-Waga), definiteness (Nez Perce), or both (Hindi). In addition, since third person nominals can be definite, it is not clear why, in her account, definite third person nominals can stay in VP at S-structure. Also, as we can see in some of the above examples, the application of the animacy hierarchy to ergatives and accusatives is different, and some languages have an overlapping area in the middle of the hierarchy where both ergative and accusative Case are possible. This suggests the correctness of a hierarchy-based approach. We conclude that Jelinek's syntactic explanation of the split is inadequate, and her arguments against a discourse account are not valid.

There are other conditions as well that can be implicated in split ergativity. One such factor, about which I will have nothing to say here, is aspect. Yet another circumstance in which ergative Case may be missing morphologically is exemplified by the New Guinea Austronesian language Motu and the Australian language Murinypata, also discussed by Dixon (1994: 58–59). The ergative Case marker is present in these languages only when the sentence would be otherwise ambiguous; i.e. if there is no other way (relative animacy, real-world knowledge, agreement markers) to tell which nominal is the SUBJ and which is the OBJ. This can be seen as a discourse grammar condition that disallows overt Case marking up to the limit of ambiguity.

The conclusion to be drawn is that syntax and discourse grammar jointly determine the morphological Case that surfaces. The syntax specifies the Case that an argument has, but the morphological realization is subject to licensing by discourse grammar. There are various ways that this could be achieved in a formal theory of syntax; in a theory that recognizes functional categories, for example, Case marked noun phrases could be distinguished categorially from non-Case marked noun phrases (KP vs. DP/NP). The Case marking mechanism would specify Case without regard to the category of the nominal. In LFG, this could be done the standard way with functional equations such as the following.

(10) (­ OBJ CASE) = ACC

An OBJ would then be syntactically accusative whether it is morphologically accusative (a c-structure KP) or absolutive (a c-structure DP or NP). It will be principles of discourse grammar (or the discourse-syntax mapping) that will then determine whether a KP or DP/NP is grammatical in that position, perhaps using Optimality Theory.

This approach has as one of its consequences the conclusion that objs are specified for accusative Case even if the accusative does not surface. This makes an interesting prediction. Consider a language that allows “discontinuous NPs.”10. Presumably, under anyone's analysis of such languages Case is specified uniformly for all parts of the NP. However, in a split ergative language the different parts can surface with different Cases, as in the following example from the Jiwarli (from Austin and Bresnan 1996).
 

(11)

Juru-

ngku

ngatha-

nha

kulypa-

jipa-

minyja

parna.

sun-

ERG

I-

ACC

be.sore-

TRANS-

PST

head.ABS

‘The sun made my head sore.’

The OBJ here consists of ‘I’ and ‘head’, which differ in overt morphological marking; under our proposal, ‘I’ is a KP and ‘head’ is an NP. If Case is specified uniformly for the various parts of the OBJ, both ‘I’ and ‘head’ must be specified syntactically for accusative Case.

Similarly, the SUBJs of transitive clauses in ergative languages are ergative even if the ergative does not appear explicitly. Note the ergative Case marking on the secondary predicate in the following example from Dyirbal (Bittner and Hale 1996b).11
 

(12)

Midi-

gu

aja

palan

yibi

bura-

n.

small-

ERG

I.NOM

that

woman

see-

NFUT

‘When I was little, I saw that woman.’

A unification-based theory, such as LFG, provides a natural way to describe this duality of Case marking on NPs of this nature.

2.2 Nominative-Accusative Languages

In fact, examples of nominative-accusative languages in which accusative Case is overt only when needed by the discourse grammar are not hard to come by. However, they have generally not been considered in the context of similarity to ergative languages. For example, only specific (or definite) OBJs are marked accusative in Turkish ((13) from Blake 1994), Mongolian ((14) from Mallinson and Blake 1981), and Hebrew (15).
 

(13)	a.	Hasan	öküz-	ü	aldi.
		Hasan	ox-	ACC	bought
		‘Hasan bought the ox.’

 

	b.	Hasan	bir	öküz	aldi.
		Hasan	a	ox	bought
		‘Hasan bought an ox.’ (non-specific)

 

	c.	Hasan	bir	öküz-	ü	aldi.
		Hasan	a	ox-	ACC	bought
		‘Hasan bought an ox.’ (non-specific)

 

(14)	a.	bagši	dorj(i)-	iig	wZEbE.
		teacher	Dorji-	ACC	saw
		‘The teacher saw Dorji.’

 

	b.	Dorji	ZurEg	ZurEbE.
		Dorji	picture	painted
		‘Dorji painted a picture.’

 

(15)	a.	karati	et	ha-	sefer.
		I.read	ACC	the-	book
		‘I read the book.’

 

	b.	karati	sefer.
		I.read	book
		‘I read a book.’

 
There are also nominative-accusative languages in which only animate objs are marked with accusative Case. One such language is Malayalam (Mohanan 1982).
 

(16)	a.	Puuca	eliy-	e	/	*eli	tinnu.
		cat.NOM	rat-	ACC	/	*rat.NOM	ate
		‘The cat ate the rat.’

 

	b.	Puuca	roTi	/	*roTiy-	e	tinnu.
		cat.NOM	bread	/	*bread-	ACC	ate
		‘The cat ate the bread.’

There are other languages that can be analyzed as exemplifying this, but are not generally analyzed as such. For example, the well-known appearance of a before animate/specific direct objs in Spanish (and their ability in some dialects to be doubled by a clitic) can be plausibly assimilated to this kind of analysis if we suppose that a, instead of being an alternative Case assigner, as often assumed, is instead the morphological realization of accusative Case in Spanish (example from Blake 1994.)
 

(17)	a.	Deseo	un	empleado.
		I.want	an	employee
		‘I want an employee.’			[Anyone will do.]

 

	b.	Deseo	a	un	empleado.
		I.want	ACC	an	employee
		‘I want an employee.’					[I can't think of his name at the moment.]

Even English, both Old and Modern, seems to restrict accusative Case to high animacy nominals. In Old English, neuter nouns and pronouns did not have accusative forms.12 In Modern English, accusative Case only surfaces in pronouns (which are higher on the animacy hierarchy than nouns), and not on the inanimate pronoun it.13

A more complicated pattern involving both definiteness and animacy is apparent in some of the Bantu languages. Wald (1979) examines the situation in the Mombassa, Kenya, dialect of Swahili, and suggests that discourse “distinctiveness” is involved, a concept involving unexpectedness and topicality.

A related phenomenon can be found in Hungarian. The SUBJ agreement marker in Hungarian is taken from one of two sets, traditionally called the indefinite and definite conjugations. The indefinite conjugation marker is used if the verb is intransitive, or if the verb is transitive and has an indefinite OBJ. The definite conjugation is used if the verb has a definite OBJ (Marácz 1989).14
 

(18)	a.	Lát-	ok	egy	lány-	t.
		see	1SG.INDEF	a	girl-	ACC
		‘I see a girl.’

 

	b.	Lát-	om	a	lány-	t.
		see	1SG.DEF	the	girl-	ACC
		‘I see the girl.’

The indefinite and definite agreement suffixes might be more accurately called intransitive and transitive respectively, with the special transitive form being used only when the obj is definite.

There are also nominative-accusative languages in which OBJs are nominative when there is no nominative SUBJ. Dixon (1994) mentions Finnish (in 1st and 2nd person imperatives, where the subj is unexpressed) and other Balto-Finnic languages, Australian nominative-accusative languages (such as Ngarluma, Lardil, and Kayardild), and Southern Paiute and other Uto-Aztecan languages. In Finnish, for example, there is no overt nominative SUBJ in first and second person imperatives. In sentences of this type, OBJs are nominative (Comrie 1978).
 

(19)	a.	Maija	tuli.
		Maija.NOM	came
		‘Maija came.’

 

	b.	Maija	söi	kala-	n.
		Maija.NOM	ate	fish-	ACC
		‘Maija ate fish.’

 

	c.	Syö	kala!
		eat.IMP	fish.NOM
		‘Eat fish!’

A striking example is Icelandic. In Icelandic, SUBJs sometimes exhibit inherent (“quirky”) Case. In such a case, the OBJ (or ECM subj) is nominative (Zaenen and Maling 1982, Yip, Maling and Jackendoff 1987).
 

(20)	a.	Barninu	batnaði	veikin.
		the.child.DAT	recovered	the.disease.NOM
		‘The child recovered from the disease.’

 

	b.	Barninu	finnst	mjólk	góð.
		the.child.DAT	finds	milk	good
		‘The child likes milk.’

 

	c.	Henni	hefur	alltaf	þótt	Ólafur	leiðinglegur.
		her.DAT	has	always	thought	Ólaf.NOM	boring
		‘She has always thought Ólaf boring.’

In a ditransitive clause with a dative object and an accusative object, if the dative becomes SUBJ the retained object surfaces as nominative, not accusative.
 

(21)	a.	Jón	gaf	barninu	bókina.
		John	gave	the.child.DAT	the.book.ACC
		‘John gave the child the book.’

 

b.

Barninu

var

gefin

bókin

(af

Jóni).

the.child.DAT

was

given

the.book.NOM

(by

John.DAT)

‘The child was given the book by John.’

This kind of situation is analogous to those languages that do not mark Case unless there is a danger of ambiguity.

Thus, evidence from nominative-accusative languages confirms the basic outline of our approach to ergative languages. Specifically, we see evidence that the realization or nonrealization of accusative Case can be dependent on nonsyntactic (specifically discourse) conditions.
 

3. Derivationalism

I believe that it is important to examine Jelinek's opposition to a discourse-based explanation of split ergativity. Jelinek is not alone in seeking a purely syntactic account of the distribution of Case. Almost every account of Case that has been proposed in the GB and Minimalist frameworks has been purely syntactic in nature. (A recent example is Bittner and Hale 1996a,b.) This seems odd, since there is no reason in principle why a GB theory of Case should be based purely on syntax: as a modular theory, GB should have no problem with the phenomenon of Case being partly dependent on a nonsyntactic module.

Here is where the conceptual distinction between a derivational theory and a parallel level constraint-based theory becomes crucial. A derivational theory like GB leads one to view all kinds of syntactic, and perhaps linguistic, information as being of the same kind; multiple strata of the same kind of representation connected by some derivational procedure (Move   in GB). While such an approach is not necessary, a derivational theory is conceptually conducive to it. The theory of Generative Semantics was an early attempt to view nonsyntactic information as part of the syntactic derivational system. Examples of this kind of approach in GB outside of the realm of Case include the subsumption of morphology under syntax by Baker (1988) and the approach to argument structure proposed by Hale and Keyser (1993).

On the other hand, a theory that is designed to view syntax as consisting of distinct autonomous levels, each with its own principles of organization, and in which a grammatical structure must meet the constraints on each of these levels simultaneously, is one that leads the theoretician to search for the domain-specific properties that lie behind the data. Nonderivational parallel constraint-based theories are therefore, to quote Chomsky (1995), although in a different context, “therapeutic.”

In conclusion, while an approach to the distribution of morphological Case in ergative languages that refers to notions like animacy hierarchies is not in fact inimical to GB, it feels unnatural to most theoreticians working in such a framework. On the other hand, a theory that recognizes distinct parallel structures with different kinds of constraints applicable at each can more comfortably accommodate the kind of approach to split ergativity that I am proposing. LFG is, of course, such a theory.
 

References

Baker, Mark (1988) Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press.

Bittner, Maria, and Kenneth Hale (1996a) “The Structural Determination of Case and Agreement.” Linguistic Inquiry 27: 1–68.

Bittner, Maria, and Kenneth Hale (1996b) “Ergativity: Towards a Theory of a Heterogeneous Class.” Linguistic Inquiry 27: 531–604.

Blake, Barry J. (1994) Case. Cambridge: Cambridge University Press.

Chomsky, Noam (1995) The Minimalist Program. Cambridge, Mass.: MIT Press.

Comrie, Bernard (1978) “Ergativity.” in Winfred P. Lehmann, ed., Syntactic Typology: Studies in the Phenomenology of Language. Austin: University of Texas Press. 329–394.

Comrie, Bernard (1984) “Reflections on Verb Agreement in Hindi and Related Languages.” Linguistics 22: 857—864.

Diesing, Molly (1992) Indefinites. Cambridge, Mass.: MIT Press.

Dixon, R. M. W. (1979) “Ergativity.” Language 55: 59–138.

Dixon, R. M. W. (1983) “Nyawaygi.” in R. M. W. Dixon and Barry J. Blake, eds., Handbook of Australian Languages, Volume 3. Canberra: The Australian National University Press. 431–525.

Dixon, R. M. W. (1994) Ergativity. Cambridge: Cambridge University Press.

Falk, Yehuda N. (1991) “Case: Abstract and Morphological.” Linguistics 29: 197–230.

Givón, T. (1995) Functionalism and Grammar. Amsterdam: John Benjamins.

Hale, Kenneth, and Samuel Jay Keyser (1993) “On Argument Structure and the Lexical Expression of Syntactic Relations.” in Kenneth Hale and Samuel Jay Keyser, eds., The View From Building 20: Essays In Honor Of Sylvain Bromberger. Cambridge, Mass.: MIT Press. 53–109.

Jackendoff, Ray (1990) Semantic Structures. Cambridge, Mass.: MIT Press.

Jelinek, Eloise (1993) “Ergative ‘Splits’ and Argument Type.” in Jonathan D. Bobaljik and Colin Phillips, eds., MIT Working Papers in Linguistics 18: Papers on Case and Agreement I. 15–42.

McGregor, R. S. (1972) Outline of Hindi Grammar, Second Edition. Delhi: Oxford University Press.

McLendon, Sally (1978) “Ergativity, Case, and Transitivity in Eastern Pomo.” International Journal of American Linguistics 44: 1–9.

Mahajan, Anoop (1994) “Oblique Subjects and Burzio's Generalization.” Paper presented at the Workshop on Burzio's Generalization, Utrecht University, 3–4 June 1994.

Mallinson, Graham, and Barry J. Blake (1981) Language Typology: Cross-linguistic Studies in Syntax. Amsterdam: North-Holland.

Marácz, László (1989) Asymmetries in Hungarian. Doctoral dissertation, Rijksuniversiteit Groningen.

Mohanan, K.P. (1982) “Grammatical Relations and Clause Structure in Malayalam.” in Joan Bresnan, ed., The Mental Representation of Grammatical Relations. Cambridge, Mass.: MIT Press. 504–589.

Morphy, Frances (1983) “Djapu, a Yolngu Dialect.” in R. M. W. Dixon and Barry J. Blake, eds., Handbook of Australian Languages, Volume 3. Canberra: The Australian National University Press. 1–188.

Nash, Léa (1995) Portée argumentale et marquage casuel dans les langues SOV et dans les langues ergatives: L'exemple du géorgien. Doctoral dissertation, Université de Paris VIII.

Rude, Noel (1986) “Topicality, Transitivity, and the Direct Object in Nez Perce.” International Journal of American Linguistics 52: 124–153.

Wald, Benji (1979) “The Development of the Swahili Object Marker: A Study of the Interaction of Syntax and Discourse.” in Talmy Givón, ed., Syntax and Semantics, Volume 12: Discourse and Syntax. New York: Academic Press. 505–524.

Woolford, Ellen (1997) “Four-way Case Systems: Ergative, Nominative, Objective and Accusative.” Natural Language and Linguistic Theory 181–227.

Yip, Moira, Joan Maling, and Ray Jackendoff (1987) “Case in Tiers.” Language 63: 217–250.

Zaenen, Annie, and Joan Maling (1982) “Passive and Oblique Case.” in Lori Levin, Malka Rappaport, and Annie Zaenen, eds., Papers in Lexical-Functional Grammar. Bloomington, Ind.: IULC. 159–191.

Footnotes

2. As will be made clear, we reject this as an accurate description of the situation in Dyirbal. In particular, the nominative part of the description is not entirely accurate. What is crucial here is the presence of accusative Case on pronouns. [Back to text]

3. We will not distinguish here between definiteness and specificity. [Back to text]

4. More accurately an accusative active system, in which the sole argument of an unaccusative verb gets accusative Case. See McLendon (1978) and Bittner and Hale (1996a). [Back to text]

5. McGregor (1972: 49), in a pedagogical grammar, states that the accusative postposition ko is used “with direct objects which are individualized to some extent, and to which a degree of contextual importance is thus attached; hence usually where direct objects refer to human beings, and certain animals, and quite frequently where they refer to inanimate objects.” As for unmarked (absolutive/nominative) direct objects, they are “not of any individual importance in a given context  In practice, words [sic] used in the direct [i.e. unmarked] case in this way usually have inanimate reference, but not invariably so.” The examples in the text come from Comrie (1984). [Back to text]

6. Mahajan (1994) states “It is sometimes claimed that -ko is an accusative marker in Hindi. Since the presence or absence of -ko is related to the specificity of the object rather than any relevant property of the verb, it is not entirely clear how it can be construed as an accusative marker.” It should be clear from the typological observations here that such behavior is typical of accusative morphology and is therefore not evidence against such an analysis of Hindi ko. [Back to text]

7. Both Rude and Woolford use the lack of object agreement in the Caseless construction as arguments for their analyses. For Rude, this is evidence that the Caseless construction is intransitive, and in particular that the Patient argument is not an object. For Woolford, it is evidence that the object has not raised to [SPEC, AGR_OP]. However, we see from other languages that object agreement is subject to the same animacy and definiteness/specificity conditions as overt accusative Case marking, so there is no reason to see the agreement facts in Nez Perce as a separate phenomenon to be accounted for. [Back to text]

8. The basing of the explanation of some linguistic phenomenon on a scale or hierarchy does not imply scalar judgments. It is generally accepted that syllable structure is based on a scale of sonority, yet in any given language a particular sequence of phonemes either is or is not a well-formed syllable. There are no scalar judgments. [Back to text]

9. Nash (1995) also bases a syntactic account of animacy splits on referential differences between first and second persons on one hand and third person on the other. While some languages do display this split, not all animacy (or person) splits are at the boundary between second and third persons. [Back to text]

10. The reason for the scare quotes is that I am not taking a position on whether these are really NPs at some level of representation or not. On the analysis of languages like these, see Austin and Bresnan (1996). [Back to text]

11. Bittner and Hale gloss the pronoun as ergative, although the form is identical to the one used for intransitive SUBJs. They state that distinction between nominative and ergative is “masked.” Within the context of the approach being proposed here, this notion of “masked” gets a precise definition: the argument is marked for ergative Case in the syntax, but it is realized as a bare DP instead of a KP (or alternatively, it is not morphologically marked for Case). The fact that the form derives historically from an ancestral ergative is, of course, just as irrelevant to the synchronic analysis of Dyirbal as the fact that most of the English accusative pronouns derive from Old English datives is to the synchronic analysis of English. [Back to text]

12. As Mallinson and Blake (1981) point out, this is true in many Indo-European languages. [Back to text]

13. Accusative Case also does not show up on the second person pronoun you either. This may also have a basis in discourse grammar. Alternatively, as Cynthia Allen has suggested to me (p.c.) it may be just an accident of the history of English. [Back to text]

14. Note, however, that the OBJ is overtly accusative even when it is indefinite. [Back to text]