On Certain Clause Types [a 1984 working paper on ergativity]

University of Melbourne

Linguistics Section

Faculty of Arts

Working Papers in Linguistics

No. 10, December 1984: 57-68

On Certain Clause Types

Probal Dasgupta

Deccan College, Pune 411006, India

1. This is very much a working paper. I will assume some of the small clause theory developed by Kayne, Stowell, Pesetsky and other GB syntacticians, and make proposals about certain types of clause structure within a variant of the GB framework. The point is to explore certain elaborations of GB in a way which, I hope, will seem fruitful and therefore will provoke criticism and correction. My proposals in their present form are almost certainly wrong, and therefore I have compressed this presentation as much as possible, to avoid taking up too much of your time.

2. Schwartz (see the bibliography for all references) has claimed that there is a significant difference between the ergative and accusative clause types: while the latter may be VSO, SVO, or SOV, the former may only be VSO or SOV, never SVO. (With apologies to Hixkaryana, I will stick to the major types and ignore VOS, OVS, and OSV.) I am told by as eminent an ergativologist as Dixon that, indeed, there are no well-known SVO languages with syntactic ergativity; let us assume that there are none at all, and proceed. Since languages are not terribly rigid about favouring this or that clause type, let us speak of clause types and not of language types. Now, why is Schwartz’s generalization real? How do SVO order and syntactic ergativity clash? I will offer what amounts to an answer to this question.

[p 58:] 3. This answer makes no sense without a wider setting. What goes on in a clause? Let us, to be concrete, propose an exact representation:

(1)               [_S”NP-topic [_S’NP-focus [_S’Comp [_SNP-subject [_PredPInfl VP]]]]]

I will assume, with Kayne and others, that PredP is Infl’ and that S is Infl”, and that S’ is Comp’. I am adding the assumption that S” is Comp”. The assumption that a focus is adjoined to S’ is made in Baltin’s thesis and related work. By parity, we may propose that some languages are going to have a focus adjoined to PredP which, being Infl’, is parallel to Comp’; this proposal makes sense in the predominantly SOV languages of South Asia, at least, but will not be pursued here. I am merely focusing on the point that Infl and Comp seem to be deeply similar. If they are, the Topic GF, [NP, S”], is on a par with the Subject GF, [NP, S]. I would like to propose that these are not D-structure GFs. D-Structure is a pure representation of Theta Roles. Notions like Topic and Focus involve Iota States, where iota stands for the system of Introduction, introducing New material in relation to Given material in a discourse, just as theta stands for the THematic system. Assume that Subject, being structurally parallel to Topic on our assumptions, also involves Iota States. If Move Alpha in syntax maps theta representations onto mixed theta-iota representations, it follows that all material in Subject, Focus or Topic positions at S-structure must have come there by movement. Not only does this assimilate Move NP (to Subject) and Move Wh (to Focus or Topic); this also says that ordinary subjects of clauses, without raising or passive, become subjects by moving into subject position. Subject-Infl agreement, Williams-coindexing between subject and predicate, Nominative Case Assignment – all this must take place or be checked at S-structure.

[p 59:]4. Where do subjects come from? What is their canonical D-structure position? Initially, let us make the standard GB assumption that subjects, unlike objects, are indirectly theta-marked and thus are outside the ‘real’ VP comprising the V and its complements. Suppose I postulate:

(2)                [_PredPInfl [_VPNP VP]

Let the NP in (2) be the D-structure subject. Given this configuration, we may propose that, in a typical VSO language, V moves to Infl, accounting for the surface facts, and that, in an SVO or SOV language, NP moves to what (1) calls the subject position and antecedent-governs its trace, which is made possible by the fact that the ‘broad VP’ (which contains NP and the ‘narrow VP’) is not a maximal projection.

5. The broad VP of (2) is technically a small clause with NP as its ‘subject’ in the sense of the work of Stowell, Kayne, etc.  I am not proposing a fresh modification of X’ theory.

6. The narrow VP, under orders from its V, assigns the Agent theta role to its sister NP, or tells this NP to be an expletive pro. The relevant sense of Agent is problematic (in John understands the problem, is John an Agent or an Experiencer?), but that is an unsolved problem of long standing.

7. So much for one clause type. (2) is the basic story for accusative clauses, and for such variants as unaccusative clauses like The stick broke, in which we need to say something about what theta role the V assigns to its deep object and S-structure subject; I will take all that as done. Let us proceed to syntactically ergative clauses, for which (2) fails. Let us try (3):

(3)               [_VP[_NP3NP₁ NP₂] V]

[p 60:] Assume that only VP can assign the Agent theta role. In (3), V can only assign Theme, which it assigns to the small clause NP₃ and thus to its head NP₂. If we take NP₃ V to be an ‘unaccusative’ structure in that V assigns no Case to NP₃, then NP₃ must move:

(4)               [_SNP₃ PredP]

The Infl head of PredP will assign Absolutive to NP₃ and thus to its head NP₂. This still leaves NP₁ unroled and uncased. NP₁ must therefore take care of itself. It must bear Ergative Case. We may say that NP₁ is really, throughout, a PP with a special kind of P, an Ergative Case desinence, which expresses Agent theta role, and, like Locative and other ‘concrete’ Case desinences (cf. the work of Kurylowicz on ‘concrete’ Cases, and the discussion of this work in my 1985 paper), is ‘self-appointed’. Another option is to move NP₁ and form (5),

(5)               [_SNP₁ S]

and then to let NP₁ in this new position receive Ergative Case, either by invoking a rule like the Genitive Rule or by letting the S sister of the NP₁ position in (5) assign this Case. If Ergative Case always has an Agent reading, the Logical Form will then come out right.

8. How do we get the Ergative phrase to receive Agent theta role? Surely we don’t want to stipulate that Ergative Case expresses this role. Since agentive verbs like hit and read in general are indirect assigners of the Agent theta role, we expect this lexical property to play a role in ergative clauses too. In an accusative clause with an agentive V, the VP projected by this V theta-marks its NP sister in the broad VP shown in (2). But, in an ergative clause, VP has no NP sister. Recall that NP₃ moves to the position shown in (4), where PredP contains only Infl and [p 61] the VP (3). Thus the usual mechanism of indirect theta-marking through V is rendered inoperative. Assume that in such a case VP passes on its indirect theta-marking ability to the next maximal projection, S. Thus, S, on V’s behalf, theta-marks its sister NP₁ in (5). It was suggested in paragraph 7 (as a second possibility) that S also Case-marks NP₁; perhaps, specifically, an S can Case-mark such an NP sister if it theta-marks this NP. Hence the association of Agent theta role and Ergative Case. No stipulation is needed.

9. The proposal of paragraph 7 says that NP₁ receives no theta role in its D-structure position in (3), but is theta-marked only in its S-structure position shown in (5). If this is rejected as conceptually unsound, we may propose instead that in (3) the NP₂ head of NP₃ and the V compositionally theta-mark NP₁; cf. the standard account of joint theta-marking by V and P in [V [P NP]]. Then NP₁ in (5) receives Ergative Case by other means, e.g. structural Case assignment, the first possibility suggested in paragraph 7. This solution would be more in keeping with the GB style than that of paragraph 8. While I will stick to paragraph 8 here, the question remains open.

10. In syntactically ergative clauses, NP₂ has some subject properties, e.g. the ability to control the PRO subject of an adverbial clause. How does this come about? Consider (4). NP₂ heads NP₃, which, as the daughter of S in (4), is supposed to have some subject properties.

11. R. Kayne has pointed out that my account needs to be supplemented by an answer to the following question: How is it that no ergative language has S-structures such as (6)? If I cannot answer this question, I have, in effect, said nothing about Schwartz’s generalization.

(6)               [NP₁ Infl [_VPV [_NP3 t₁ NP₂]]]

[p 62:] Try some elaborations of (6). It is clear that (7) is out:

(7)               [_SNP₁ [_PredP Infl [_VPV [_NP3 t₁ NP₂]]]

We want NP₁ and NP₂ to be Ergative and Absolutive, respectively, and in (7) Infl doesn’t govern NP₂ and does govern NP₁. Thus (7) is not a possible ‘SVO clause’ representation. Try (8) now:

(8)               [_SNP₁ [_S[_PredPInfl [_VPV t₃]][_NP3 t₁ NP₂]]]

One should be able to make this work; NP₂ agrees with, and is assigned Absolutive Case by, Infl; NP₁ is the Ergative phrase; theta-marking is as in my account of (3)-(5). Why is there, apparently, no ergative language like this?

12. In English, where a lexical (or non-lexical) head precedes its complements, forming H’, the specifier, if any, precedes H’, forming H”, as noted in Remarks on nominalization. If PredP is Infl’ and S is Infl”, the subject is the specifier of Infl’ and should precede Infl’. This canon is violated by (8), where the specifier NP₃ follows the Infl’ (the PredP). This is not a theoretical answer, of course, but it is a starting point. There are several moves one can imagine making. One would be to say: English is a head-first language; Kayne and others have taken part in a process of turning specifier positions into head positions (e.g. Infl and Comp used to be regarded as specifiers), helping to reduce ‘Specifier precedes Head-bar’ to a case of ‘Head precedes Complement’; to take the process further, some people have proposed that the Topic position heads S”; going back to (1), since Subject parallels Topic, perhaps the Subject position heads S, to extend the innovative idea of a position type (rather than a category) heading a projection (the idea that Topic is the head of S” is due, I seem to recall, to Guéron and May), not an implausible idea, for the notion of Infl and Comp heads already extends head theory to permit a function to head a [p 63:] projection (notice that Infl and Comp, as functors, are not ‘categories’ like NP as distinct from functions like ‘Subject’ in the sense of the classic discussion of this matter in Aspects), and the idea of Topic heading S” or Subject heading S only makes fuller use of this already proposed innovation. This move, or series of moves, entails the interesting consequence that (8) violates the Empty Category Principle. Consider NP₃ and its trace t₃. The lexical head V, which governs t₃, starts a g-projection set, in the sense of Kayne’s recent (and still current) work on connectedness. Infl canonically governs VP in this head-first language, permitting the g-projection set to proceed to PredP, the maximal projection of Infl on the new assumption that S is a projection of subject and not of Infl. Now, NP₃ is on the incorrect branch: it fails to canonically govern PredP; thus, the g-projection set initiated by the lexical governor of t₃ does not contain the antecedent of t₃; the ECP is violated. In contrast, take (9), the result of packing (3) and Infl into (4) and (4) into (5):

(9)[_SNP₁ [_S[_NP3 t₁ NP₂][_PredP[_VP t₃ V] Infl]]]

This respects the Kayne version of the ECP, as the reader can verify, as far as t₃ is concerned. There is a problem with t₁, for NP₂ is not a lexical head and Infl on my assumptions cannot govern NP₃ and thus cannot govern into it; but my hunch is that the ‘theta route’ from V to the narrow S (which, on V’s behalf, theta-marks its sister NP₁) saves t₁. Assume that V hands the theta job over to VP, which passes it on to PredP, which passes it on to the narrow S; this gives Infl, or even V, an access to (and into) NP₃ which it otherwise would not have. Alternatively, NP₁ in its new position antecedent-governs and thus sanctions t₁ by governing across the NP₃ boundary and ignoring the S boundary if we may assume that the narrow S and the broad S are ‘the same projection’ and that therefore NP₁ (a) heads it and (b) does not see the narrow S as a [p 64:] boundary; if this is licit, we don’t need g-projections for t₁.

13. One may ask a very different question. Why not assume that (7) is the representation of English John loves Mary? Why start with (2) and not (10) as the D-structure level in English?

(10) [_PredPInfl [_VPV [_NP3 NP₁ NP₂]]]

Once we are willing, on the basis of the considerations outlined above, to accept (9) as an ergative S-structure, we may consider giving up our initial assumption that accusative clauses have the quite different D-structure (2). Perhaps all D-structures have a VP consisting of V and, if there are two arguments, a small clause comprising these arguments. An S-structure like (7) will ensue. Let us work this out.

14. The issues are slightly sharper in a language like Hindi, with SOV order and split morphological ergativity: the Theme has no syntactic subject properties (does not control into adverbial clauses, bind reflexives, etc.) but controls Infl agreement when the V is a past participle, forcing the Agent into the ‘morphologically ergative’ or Agentive Case (marked by the Postposition ne); when the V is in some other form, we have a straightforward accusative clause. Let us, then, explore the idea that (11) is the D-structure and (12) the S-structure for Hindi ra:m ciTThi: likhega: “Ram letter will.write” ‘Ram will write a letter’.

(11) [_PredP[_VP[_NP3 [_NP1 ra:m][_NP2 ciTThi:]][_Vlikhe]]ga:]]

(12) [_S[_NP1 ra:m][_PredP [_NP3 t₁[_NP2  ciTThi:]][_Vlikhe]][ga:]]]

15. While retaining the GB idea of Infl and VP as a leading idea, I have implemented it somewhat unusually: in [p 65:] (11) and (12) I show the V node as dominating the verb stem likh and the mood marker e, leaving the unlabelled node after it (you may call it Infl, but only if you remember that I am cutting the pie differently) dominating only the tense-agreement marker ga:. As Infl heads PredP, it can, being partly nominal, impose Agr on its complement VP, explaining the occurrence of Agr features on V; this is analogous, on my assumptions, to the way the Subject, head of S, imposes its Agr features on PredP in general and on PredP’s head Infl in particular. In my view, different ‘moods of V’ (subjunctive dekhe, imperfect dekhta:, perfect dekha:) are slightly different categories within the broad archicategory ‘V’. This will have to be left for proper treatment elsewhere.

16. Assuming, then, that ga: is Infl in (11)-(12), let us note that V assigns Objective Case and Theme theta role to NP₃ (and its head NP₂) but lends its agentivity to the VP, which, having no sister, lends to the PredP this job of assigning Agent theta role to an Agent NP. This is how ra:m gets its theta role. Assume than an element E can theta-mark (on its own or on some other element’s behalf) a constituent C iff E Case-marks C. Assume that ‘lending’ the ability to assign a particular theta role is a real phenomenon, and that an element E will assign Case in accordance with its own nature (PredP assigns Nominative/ Absolutive; S assigns Agentive/ Ergative; V assigns Objective). Assume that, in a syntactically ergative language, V does not assign Case (and therefore must lend its theta roles), and that, if a V lends two theta-marking abilities to VP, i.e. Theme and Agent, then the usual inner-outer distribution with Theme inside and Agent outside will be respected (for reasons not yet understood), i.e. in an ergative clause the PredP will end up assigning the Theme theta role and the S, the Agent theta role. On these assumptions, many problems dissolve, and we can even derive as a theorem a version of Burzio’s generalization (cf. Chomsky 1981) that a V which does not [p 66:] Case-mark its object cannot indirectly theta-mark its subject position: the point, recast in this framework, is that an ‘unaccusative’ V, which takes an NP but does not Case-mark (and therefore does not theta-mark) this NP, will either take an NP₃, a small clause, or just an NP₂, a bare theme. In the former case, the V will, in our terms, lend its Theme and Agent theta-marking abilities upwards, with the consequences we have seen. In the latter case, which Burzio was concerned with, the V’s theta grid has just a Theme, no Agent.

17. Returning to (12), we see, then, that ra:m gets Nominative Case and Agent theta role from PredP, while ciTThi: is Case-marked and theta-marked by V. Consider now a morphologically ergative structure:

(13) [_S[_NP1 ra:m ne][_S[_NP3 t₁ [_NP2 ciTThi:]][_PredP[_VP t₃ [_Vlikhi:]][_Infl hai]]]]

Ram AGENTIVE       letter.Fem        written.Fem     is

‘Ram has written a letter’

The perfect mood of the V, as mentioned in paragraph 15, is a specific category (although called ‘V’), like the subjunctive dekhe of (12), with a categorial feature bundle of its own (for some proposals about an adequate theory of categorial features, see my forthcoming Deccan College Bulletin paper); a perfect mood V, unlike a subjunctive mood V, cannot assign Case. Therefore, on our assumptions, it cannot do its own theta-marking either, but must lend both the Theme and the Agent of its theta grid to VP, which lends them to PredP, which takes the lower member of the hierarchy (the Theme), ciTThi:, assigns Theme theta role and Nominative Case, and lends the Agent-marking job to S, which theta-marks and Case-marks its NP₁ sister; hence the Agent theta role and Agentive Case of ra:m ne.

[p 67:] 18. On this account, morphological ergativity involves a particular ‘form of a V’ (a particular category in the V family) being unable to assign Case, and syntactic ergativity, we may assume, involves an inability on the part of V in general, as an archicategory, to assign Case. This is the parameter. If V in an ergative system never assigns Case, canonical alignments change, so that PredP canonically assigns Absolutive Case and Theme theta role, making it a regular pattern for S to assign Ergative Case and Agent theta role. Arguably, the canonical position of NP sister of PredP is locus of classical ‘subject properties’; hence the facts of Dyirbal or Central Arctic Eskimo. As a matter of fact, ergativologists say that all ergative languages have accusative enclaves. Perhaps these are like Hindi’s ergative enclave. Just as some moods in the Hindi V ‘switch Case off’, perhaps some regions in the V system of a syntactically ergative language ‘switch Case on’, producing a superficial deviance from the patterns that are canonical in the language.

19. All this will have to pass the test of the passive and anti-passive patterns carefully studied by relational grammarians. I don’t yet have enough to say about these constructions. As is indicated by my calling hai ‘is’ in (13) an Infl, I assume that auxiliary elements start out in Infl; perhaps a passive PredP is of the form [_PredPPredP Infl] in Hindi and correspondingly [_PredPInfl PredP] in English; perhaps John is good has a D-structure like [_PredP[_Inflis][John good]], where [John good] is a small clause. More work is obviously needed so that this model can be fleshed out and criticized seriously. Meanwhile, it is surely a straightforward task to work out the consequences of my assumptions for the analysis of well-studied languages like English and French, which are not all that different from Hindi. And surely linguists who are familiar with ergative languages can tell, as grammarians, whether what I am saying ‘sounds right’.

[p 68:]Bibliography

Baltin, Mark R. 1978. Towards a theory of movement rules. MIT Ph.D. dissertation.

Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.

Dasgupta, Probal. In press (1985). On Bangla nouns. Indian Linguistics 46:1-2.

Dasgupta, Probal. Forthcoming (1986). Gerunds, participles and the theory of categories. Bulletin of Deccan College Research Institute.

Kayne, Richard S. 1984. Connectedness and binary branching. Dordrecht: Foris.

Kuryłowicz, Jerzy. 1960. Esquisses linguistiques. Wroclaw and Krakow: Wydawnictwo Polskiej Akademii Nauki. Especially pp 131-50, Le problème du classement des cas.

Pesetsky, David. 1982. Paths and categories. MIT Ph.D. dissertation.

Schwartz, Arthur. 1972. The VP constituent of SVO languages. In John P. Kimball, ed., Syntax and Semantics Vol. 1. New York: Seminar Press.

Stowell, Tim. 1983. Subjects across categories. The Linguistic Review 2.285-312.

Colophon

Some stimulating discussion with Anoop Mahajan and Gyanam Dave was of great help in putting this paper together. So was some significant work by Paul Postal on antipassives; although my thinking has not yet reached a stage where I can tackle his facts, I should make my personal debt to him – for sending me his work and for discussing many issues with me – clear. My debt to Richie Kayne is clear from the text, surely.