A Note on 'Principle C' in Ellipsis Constructions

Jan-Wouter Zwart

NWO/University of Groningen

1998

When Noam Chomsky discussed his 'minimalist program' with students, staff, and visitors in the Fall of 1991, it became clear that the familiar LF-movement operation of Quantifier Raising (May 1985) is not easily described in terms of a morphosyntactic 'feature checking' requirement. "But I never believed in that anyway," Noam said. When asked how to avoid the infinite regress in antecedent contained deletion constructions, he bounced the question back. I never thought about it much after that, but a recent flurry of activity in the domain of ellipsis research made me take another look at the problem. I'd like to offer this note to Noam in recognition of the extraordinary pleasure I always get out of the problems he inspires us to address.

This note starts from an apparent exception to Principle C of the Binding Theory (Chomsky 1981:188), illustrated in (1) (cf. Fiengo & May 1994:265, Fox 1995):

(1)
I bought him_i everything John_i wanted me to

Principle C of the Binding Theory states that R-expressions must be free (2), i.e. not bound, where being bound (binding) is defined as in (3):

(2) Principle C
R-expressions are free

(3) Binding
A binds B iff A c-commands B and A and B are coindexed

Principle C of the Binding Theory explains the general impossibility of sentences like (4):

(4)
* He_i saw John_i

The coindexing in (4) (and (1)) expresses that he and John are intended as referring to the same (notion of a) person. In both (4), the general case, and (1), the exception, is the R-expression John c-commanded by a coindexed pronoun (him/he), in violation of Principle C of the Binding Theory. The question to be addressed is why (1) is not ungrammatical under the indexing shown.

(1) contains an empty Verb Phrase (VP Ellipsis) in the complement of to, interpreted as in (5):

(5)
I bought him_i everything John_i wanted me to buy him_i

As noted in the literature, Principle C is not lifted in the absence of ellipsis:

(6)
* I bought him_i everything John_i wanted

Moreover, the pronoun c-commanding the R-expression must be contained in the VP functioning as the 'antecedent' for the VP ellipsis (i.e. bought him everything John wanted me to in (1)):

(7)
* He_i bought me everything John_i said he_i would (buy me)

As noted, the antecedent for the empty VP in (1) is bought him everything John wanted me to. That is, the deletion site in the complement of to is contained within the antecedent for the deleted VP ('antecedent contained deletion').

It has been proposed (by May 1985, Fiengo & May 1994, Fox 1995, Merchant 1998) that the interpretation of (1) as (5) (at the level of Logical Form) involves 'reconstruction' of the antecedent VP in the position of the empty VP. This reconstruction leads to an infinite regress, as the empty VP is contained within the reconstructed antecedent VP bought him everything John wanted me to [e]. The correct reconstruction needs to involve buy him (with a morphosyntactic adaptation of bought from the antecedent VP, cf. Bouton 1970, Potsdam 1997) and a wh-trace to be bound by the operator associated with the relative clause everything John wanted me to (assuming the analysis of relative clauses as involving wh-movement of Chomsky 1977, illustrated in (8):

(8)
everything OP_i John wanted me to buy him t_i

). The reconstructed VP therefore should be as in (9):

(9)
buy him [t]

In the approach to the interpretation of VP-ellipsis involving reconstruction at the level of Logical Form (following May 1985), the antecedent VP is modified to the form in (10), the trace resulting from leftward movement of the quantified noun phrase everything John wanted me to at the level of Logical Form (Quantifier Raising):

(10)
bought him [t]

Quantifier Raising yields the structure in (11):

(11)
[everything John wanted me to [_VP e ] ]_i I [_VP bought him [t]_i ]

From (11), reconstruction of the antecedent VP in (10) in the position of the empty VP inside the quantified noun phrase is straightforward, yielding (9) and (5).

It should be noted at this point that the Logical Form operation Quantifier Raising solves a problem (the problem of infinite regress under reconstruction) which is generated solely by the assumption that the interpretation of VP-ellipsis involves another Logical Form operation, reconstruction. I believe that neither of the two operations is well motivated. Quantifier Raising being the more contentious of the two (cf. Vanden Wyngaerd & Zwart 1991), the assumption that interpretation involves reconstruction has gone largely unchallenged (see Tancredi 1992 for an exception).

The Quantifier Raising analysis is ill equiped to deal with the problem of the unexpected grammaticality of (1). Principle C, like all principles of the Binding Theory, involves relations between A-positions (otherwise, every A'-bound wh-trace should be in violation of Principle C of the Binding Theory). In earlier work, this was expressed in terms of rule ordering (reflexivization taking place before wh-movement, cf. Postal 1971:77) or by restricting the Binding Theory to a relatively early level of representation (S-structure, cf. Chomsky and Lasnik 1995:106). The Binding Principles cannot be applied to the output of the rule of Quantifier Raising, because otherwise (6) should also be grammatical (after Quantifier Raising of everything John wanted).

Fiengo & May (1994:pp) propose that precisely in ellipsis contexts, the Binding Principles should apply at Logical Form rather than at S-structure, because the Logical Form operation of reconstruction introduces an element (him in (5) and (9)) that is subject to the Binding Theory. Note that this solution again presupposes the operation reconstruction, which I will argue below is superfluous.

In Chomsky's Minimalist Program, all syntactic conditions apply at the interface levels Logical Form and Phonetic Form (Chomsky 1993, 1995). There is no question, then, that the principles of the Binding Theory apply at Logical Form. But since binding is a relation between A-positions, wh-movement (and Quantifier Raising) are argued to leave behind full copies instead of empty traces. The copies are ignored at Phonetic Form, but are relevant to interpretational processes at Logical Form.

The 'copy theory of movement' plays havoc with the Quantifier Raising analysis of both antecedent contained deletion in general and the Principle C-lifting effect in (1). If Quantifier Raising does not leave behind a trace, as in (11), but a full copy, as in (12), the infinite regress under reconstruction cannot be avoided (copies are notated between angled brackets):

(12)
[everything John wanted me to [e] ] I bought him <everything John wanted me to [e]>

Moreover, even without reconstruction, Principle C would still be violated by the presence of John inside the copy of the raised quantified noun phrase.

Fox (1995) therefore proposes to minimize the copy of raised quantified noun phrases (a possibility left open by the discussion in Chomsky 1993), essentially stipulating that the copy theory of movement does not apply to Quantifier Raising. This proposal is motivated by the consideration that (1) would never be interpretable otherwise (because of the infinite regress). Since this is only correct assuming Quantifier Raising and reconstruction, the proposal is ad hoc.

The problems sketched conspire with the entire approach to syntactic derivation in Chomsky (1995) to suggest a different analysis of antecedent contained deletion in general, and of the construction in (1) in particular.

Chomsky (1995) proposes to describe syntactic derivations as built up from the bottom up, by application of subsequent 'merger' operations of two terminal elements or phrases. This replaces the earlier view in which structures are generated by context free rewrite rules, to be fleshed out by 'lexical insertion' of terminals at a relatively early point in the derivation (D-structure).

In the earlier view, VP-ellipsis can be described as involving a VP node, generated by the rewrite rules, which is not filled up with terminal elements during lexical insertion. Reconstruction at Logical Form is then needed to assign the required interpretation to the empty VP. The present approach does not acknowledge this distinction between phrase structure rules (rewrite rules) and lexical insertion. Consequently, no empty VP can be present at the earlier stages of the derivation, and VP ellipsis must be the result of deletion.

As Tancredi (1992) argues, VP-deletion is a variant of a more general process of deaccenting, which can be located in Phonetic Form. The syntactic derivation of (1), then, never involves a stage at which there is an empty VP to be fleshed out by reconstruction.

In other words, (5) is presumably a decent approximation of the structure of (1) both in overt syntax and at the level of Logical Form (ignoring aspects relating to the syntax of relative clauses). Needless to say that the problem of infinite regress does not occur (as there is no reconstruction) and that Quantifier Raising is not needed for the purpose of avoiding it.

(Note at this point that antecedent contained deletion is presented by May 1985 and Fiengo & May 1994 as part of an argumentation in defense of the operation of Quantifier Raising. If the approach to antecedent contained deletion involving merger and deletion is correct, this argumentation appears to be seriously affected. See Vanden Wyngaerd & Zwart 1991, 1998 for further discussion of the relevance of Quantifier Raising to antecedent contained deletion and vice versa.)

Now to return to (1) and Principle C of the Binding Theory.

Principle C is one of a set of three principles, describing the interpretation of anaphora (Principle A), pronouns (Principle B), and R-expressions (Principle C):

(13) The Binding Theory
Principle A: Anaphors are bound in their local domain
Principle B: Pronouns are free in their local domain
Principle C: R-expressions are free

(The definition of 'local domain' is irrelevant at this point.)

The Binding Theory essentially describes a morphological skewing of pronominal forms, depending on whether they are locally bound or free. We might take the Binding Theory to remain silent on nonpronominal forms, such as R-expressions. If binding is about assigning (or acquiring) reference, R-expressions are free because they are referential by themselves. Principle C, then, is a non-principle, added to the Binding Theory for the sake of completeness, but without independent theoretical status.

This trivial point has interesting consequences for the sentence in (1). If Principle C does not exist, the indexing merely indicates that him and John are accidentally taken to refer to the same (notion of a) person. Crucially, no principle of the grammar is violated. This makes the exceptional case (1) considerably less problematic, but it raises the question why he and John in (4) cannot be taken to accidentally refer to the same person.

Let us take Principle C to express that referential expressions are by definition obviative to any c-commanding noun phrase (Lasnik 1976). Non-c-commanding noun phrases are not relevant to the interpretation of referential expressions, as (14) shows:

(14)
His_i mother loves John_i

As Epstein (1995) has pointed out, the difference between an element A, c-commanding an element D, and an element B, not c-commanding an element D, is that A, but not B, is merged with a projection containing D. We may conclude from this that obviation is a function of the operation Merge (still assuming the bottom-up structure building process of Chomsky 1995). More generally, we may conjecture that interpretation of noun phrases always proceeds in tandem with the operation Merge, i.e. from the bottom up (see also De Vries 1997, Grozeva 1998, Zwart 1998).

In (14), then, his is contained within the noun phrase his mother, and therefore is not merged with a projection containing John. The interpretation of his therefore is entirely free. We may take the cataphoric interpretation indicated in (14) as an instance of accidental coreference. In (4), on the other hand, he is merged with the I' projection containing John, and must therefore be interpreted as obviative with respect to John. This makes (4) ungrammatical under the indexing shown.

By the same token, we would expect (1) to be ungrammatical. However, there is a difference between him in (1) and he in (4). In (4), he is introduced as an entirely new element in the structure; apart from the obviation requirement (Principle C), we are free to assign it any interpretation we want. In (1), on the other hand, him is introduced as an 'echo' of him in the elided VP buy him (cf. (5), repeated here for convenience).

(5)
I bought him everything John wanted me to buy him

The intonation pattern, indicated in (15), has high pitch on wanted and low flat pitch on buy him:

(15)
I bought him EVerything John WANted me to buy him

If either instance of him is replaced by another noun phrase, and the intonation is not adapted accordingly, the resulting sentence is ungrammatical:

(16)
a. * I bought Mary EVerything John WANted me to buy him
b. I bought MAry everything John wanted me to buy HIM

(17)
a. * I bought him EVerything John WANted me to buy Mary
b. I bought him everything John wanted me to buy MAry

As Tancredi (1992) has shown, antecedent contained deletion is just the result of a further reduction of the deaccented part buy him in (15). Precisely in this intonation pattern, then, is the first instance of him not free in its reference: it must be understood as a repeat of the deaccented instance of him (just like bought must be understood as a repeat of deaccented buy).

Still taking interpretation of noun phrases to proceed in tandem with the operation Merge, i.e. from the bottom up, we may describe the interpretation of him in (1) as 'bound' by the interpretation of him in the elided VP. Since there is no choice as to the reference assigned to him in this case, the obviation principle (Principle C) does not apply.

More precisely, the interpretation of him as referring to the same person as John is forced by the more accidental coreference relation between John and him in the elided VP. (I prefer to describe the relation between John and him in the elided VP as coreference rather than binding, for reasons irrelevant to the present discussion.) As is expressed by Principle B of the Binding Theory, pronouns are only subject to a local obviation requirement:

(18)
a. * John_i loves him_i
b. John_i thinks you love him_i

In (15), John and the second instance of him are not in the local domain relevant for the Binding Theory (not demonstrated here), so that nothing prevents us from interpreting John at the moment of its merger with the I' containing (the second instance of) him as referring to the same person as him. If John is in fact interpreted as such, the coindexing in (1) follows straightforwardly after deaccenting or deletion of the VP buy him.

It is intuitively clear that the exception to Principle C in (1) is made possible by the parallelism existing between the 'antecedent' VP and the deleted VP. But if reference is assigned to noun phrases in a string from left to right, or in a structure from top to bottom, it is not easy to make the parallelism have the desired effect. More exactly, at the moment that John is assigned an index (working from left to right or from top to bottom) the parallelism has not yet been established, and Principle C would flatly prohibit the indexing indicated in (1). More seriously, perhaps, the function of intonation and parallelism in VP-deletion contexts would remain obscure. If noun phrase interpretation is taken to be a function of the bottom-up structure building process Merge of Chomsky (1995), intonation and parallelism can be taken into account, and the problem posed by (1) simply disappears.

References

Bouton, L. 1970. Antecedent Contained pro-forms. Papers from the 6th Regional Meeting of the Chicago Linguistic Society, 154-167.

Chomsky, N. 1977. On Wh-Movement. In Formal Syntax, P. Culicover, T. Wasow, A. Akmajian, eds. Academic Press, New York.

Chomsky, N. 1981. Lectures on Government and Binding. Foris, Dordrecht.

Chomsky, N. 1993. A Minimalist Program for Linguistic Theory. In The Minimalist Program, by N. Chomsky. MIT Press, Cambridge, 167-217.

Chomsky, N. 1995. Categories and Transformations. In The Minimalist Program, by N. Chomsky. MIT Press, Cambridge, 219-394.

Chomsky, N. and H. Lasnik. 1995. The Theory of Principles and Parameters. In The Minimalist Program, by N. Chomsky. MIT Press, Cambridge, 13-127.

De Vries, M. 1997. Een derivationale bindingstheorie. Ms., University of Amsterdam.

Epstein, S.D. 1995. Un-Principled Syntax and the Derivation of Grammatical Relations. Ms., Harvard University.

Fiengo, R. and R. May. 1994. Indices and Identity. MIT Press, Cambridge.

Fox, D. 1995. Condition C Effects in ACD. MIT Working Papers in Linguistics 27, 105-119.

Grozeva, L. 1998. Minimalist Binding and Bulgarian. Ms., University of Groningen.

Lasnik, H. 1976. Remarks on Coreference. Linguistic Analysis 2, 1-22.

May, R. 1985. Logical Form. MIT Press, Cambridge.

Merchant, J. 1998. Antecedent Contained Deletion in Negative Polarity Items. Paper presented at the TIN-dag, Utrecht, January 17.

Postal, P. 1971. Cross-Over Phenomena. Holt, Reinhart, and Winston, New York.

Potsdam, E. 1997. English Verbal Morphology and VP Ellipsis. Proceedings of NELS 27, 353-386.

Tancredi, Chr. 1992. Deletion, Deaccenting, and Presupposition. Dissertation, MIT.

Vanden Wyngaerd, G. and C.J.W. Zwart. 1991. Reconstruction and Vehicle Change. In Linguistics in the Netherlands 1991, F. Drijkoningen and A. van Kemenade, eds. John Benjamins, Amsterdam, 151-160.

Vanden Wyngaerd, G. and C.J.W. Zwart. 1998. Merging Antecedent Contained Deletion. Ms., University of Brussels/University of Groningen.

Wilder, Chr. 1997. Phrasal Movement in LF: de re readings, VP-ellipsis and binding. Proceedings of NELS 27, 425-439.

Zwart, C.J.W. 1998. A Dynamic Theory of Binding. Paper presented at WECOL, Tempe AZ, October 9. To appear in the proceedings.