A New Argument for Verb-Stranding Verb Phrase Ellipsis

Abstract

In Gribanova, to appear, I develop an analysis of Russian constructions like (1), in which the verb moves to an Asp head just below T, in conjunction with the ellipsis of a vP-sized constituent (verb-stranding verb phrase ellipsis, or VVPE); a general schema is provided in (2).This analysis provides a natural testing ground for several difficult issues in Russian syntax, one of which is clausal structure. Using VVPE to probe this issue is especially promising for Russian, which makes use of complex verbal forms that are also potentially syntactically complex, but inseparable. In Gribanova, to appear, I follow explorations of Irish (McCloskey 2011) and Hebrew (Goldberg 2005a,b) in leveraging data from VVPE to shed light on this complexity: identity conditions on the stranded verb in VVPE can be understood to indicate which parts of the verbal complex originate inside the ellipsis site, and which parts originate outside the ellipsis site (i.e., above vP).This proposal is controversial largely because Russian, like Hebrew, is also an object drop language, which means that examples like (1) may also be successfully analyzed as instances of object drop (and perhaps PP drop). In Gribanova, to appear, I maintain that both operations—VVPE and object drop—are active in Russian and that they can be distinguished by probing the syntactic environments in which they appear: object drop is restricted inside islands, as demonstrated by the degraded nature of object drop inside islands when there is no linguistic antecedent; VVPE, like most forms of constituent ellipsis, is permitted inside islands, as long as a linguistic antecedent is available.Opponents of this approach (see Bailyn 2011, Erteschik-Shir, Ibn- Bari, and Taube 2011, 2012) correctly point out that judgments of such contrasts are gradient, subject to pragmatic effects, and difficult to obtain. In part, this is because the syntactic conditions under which the claimed contrast between object drop and VVPE is supposed to emerge are already quite complex. Bailyn’s (2011) and Erteschik-Shir, Ibn-Bari, and Taube’s (2011, 2012) proposals differ: the former takes (1) to be an instance of NP/PP-ellipsis—surface anaphora, requiring a linguistic antecedent—and the latter claims that argument drop (deep anaphora) is involved. But both proposals take the position that VVPE is not possible in Russian.In this squib, I present novel evidence from negation and disjunction— involving much less gradient judgments—in favor of the hypothesis that VVPE is involved in Russian (section 1). The hope is that this diagnostic will be effective for other languages in which there is debate about how to analyze such constructions (i.e., Hebrew, as well as the East Asian languages: see Saito 1985, Otani and Whitman 1991, Hoji 1998, Kim 1999). One of the interesting points made by Bailyn (2011) is that if these constructions are instances of VVPE, then they behave differently from other types of ellipsis in Russian. In section 2, I build on this discussion and contextualize it in light of the different kinds of ellipsis found in Russian and the new evidence from section 1.The reason such a complex diagnostic is used to distinguish VVPE from argument drop in Gribanova, to appear, is that many of the tests typically used to distinguish the two are either ineffective for Russian or have been shown to be faulty in other languages. For example, the availability of strict and sloppy readings is insufficient evidence for VVPE: Hoji (1998) demonstrates that sloppy readings are not definitively characteristic of VP-ellipsis (VPE) in other languages. Goldberg (2005b) makes productive use of the fact that VVPE should be able to elide more than one constituent, that is, VP-internal PPs and DPs. This seems to work well for Hebrew, because Hebrew PPs do not undergo drop or ellipsis independently and their absence must therefore be attributed to VVPE. For Russian, however, the judgments about whether PPs can drop independently of DPs are not reliable enough for this test to be useful.Here I pursue a novel and simpler approach, which involves an antecedent with coordination or disjunction of two VP-sized constituents, each of which may contain numerous constituents. We can start with the relatively simple case in (3).1The grammaticality of such examples points to the ability of VVPE to elide fairly large portions of the sentence, composed of multiple constituents. VVPE can also elide the coordinator, and the meaning of the elided portion in (3) can be reconstructed only as indicated.Proponents of an NP-ellipsis (NPE)/argument drop analysis might be able to account for the missing pieces of (3), if they were willing to posit that PPs can undergo ellipsis or drop along with DPs, or alternatively that what is elided is a conjunction of nonconstituents (pen on table and book on chair). I know of no evidence that the latter may occur in Russian. On the former view, the difficulty would be that there is no way to account for the missing coordinator in the response; not surprisingly, omitting just the coordinated elements and leaving the coordinator leads to strong ungrammaticality (for which there is likely more than one explanation).Proponents of an NPE/argument drop account might argue that the coordinator is simply null in (3). But this is less likely to be the case for disjunction, which tends not to have null realizations crosslinguistically (Payne 1985, Winter 1995). This makes examples like (5) even more convincing evidence in favor of a VVPE analysis.The negated verb in (5) takes scope outside the (elided) disjunction in the partially elided response, yielding only one possible reading, corresponding to the structure in (6).This, by De Morgan’s laws, can be converted (abstracting away from verb movement) to (7), which is in fact the only obtainable interpretation: that the person doing the responding put neither a pen on the table nor a book on the chair.(5) may not be interpreted as in (8a–c), which suggests that those structures are not possible models of (5).The same ellipsis process can also apply to the negative concord items, ‘(n)either/(n)or’.Proponents of an NPE/argument drop analysis will have difficulty accounting for such examples. Since there is no known operation that could independently elide or make silent the coordination, the disjunction, or the negations corresponding to ‘(n)either/(n)or’ in (3), (5), and (9), respectively, the only available analytical alternative would be to claim that what is argument-dropped/elided is a conjunction of nonconstituents.2 VVPE is a much more natural analysis of (3), (5), and (9): the antecedent for the ellipsis is one large VP, composed of two coordinated/disjoined VPs out of which the verb has escaped.An important point made by Bailyn (2011) is that constructions like (1) do not behave like other instances of Russian clause-level constituent ellipsis with respect to a few key properties, one of which—the behavior of subjects—I will discuss in this section. Bailyn takes this set of observations as a sign that constructions like (1) should not be analyzed as VVPE; but the evidence presented in section 1 seems to contradict such a conclusion. Given this, I pursue an alternative approach here, taking VVPE to be a legitimate operation and attempting to view VVPE in the context of other types of Russian constituent ellipsis. Russian makes prolific use of clause-level constituent ellipsis; a comparison among the different types, including VVPE, sheds light on what might otherwise look like a mystifying or unusual pattern.The empirical point at issue, originally observed by McShane (2005), is that uttering the subject in constructions like (1) is degraded (10), unless the subject is in some way contrastive (11).This pattern is troubling, since it diverges from what is found in Auxstranding VPE (AVPE), which is taken to be ellipsis of the complement of T (Kazenin 2006).Under the uncontroversial assumption that subjects raise to the specifier of TP (or higher) in Russian, this behavior seems strange: why may a noncontrastive subject appear in AVPE, but not in VVPE? Both operations should be able to leave behind a subject, since both operations elide a constituent—either the sister of T or the sister of Asp—that is smaller than the whole TP.It is worth noting that neither an NPE nor an argument drop approach to these data provides an obvious solution to the puzzle, in the sense that something special must be said if a dropped or elided object is to condition the appearance of a subject. Thus, the question of what explains the pattern in (10)–(12) is unanswered for any of the available accounts of these phenomena. I do not attempt to answer this question here; instead, I attempt to contextualize the contrast between VVPE and AVPE by considering how they compare with other types of clause-level constituent ellipsis in Russian.If the argument from section 1 holds up, then Russian makes use of at least three different types of clause-level constituent ellipsis processes: polarity ellipsis (13) (ellipsis of the complement of a polarity head (Σ), which itself is merged above T); ‘‘traditional’’ VPE, with a stranded auxiliary (14); and VVPE (15).3Kazenin (2006) demonstrates that AVPE and polarity ellipsis exhibit all the true characteristics of surface anaphora. He also makes a number of crucial empirical observations about the discourse structure of these types of ellipsis; when we bring these observations together with evidence from VVPE, a previously unnoticed pattern emerges.Kazenin notes that polarity ellipsis is inherently contrastive, in that it requires the phrasal remnant to be interpreted as a contrastive topic, whether or not it is the subject (16B–B′). Presentational or backgrounded readings of the phrasal remnants in (16B–B′) are not available, and neither is a contrastive focus reading, when forced by tol’ko ‘only’ (17).By contrast, AVPE is more permissive, in that the phrasal remnant may be interpreted as a contrastive topic (18), as a contrastive focus (19), or as a backgrounded (20) or presentational (21) focus (though see Kazenin 2006:28–29 for some important qualifications).Overall, then, polarity ellipsis is much more restrictive in the potential interpretations available for the phrasal remnant (contrastive topic only) than is AVPE.Kazenin (2006) does not consider VVPE in his discussion, but it is worth asking how VVPE patterns with respect to the discourse conditions that clearly are involved in other types of clausal constituent ellipsis processes. As it turns out, VVPE seems to pattern exactly like polarity ellipsis: if there is a remnant, it must be interpreted as a contrastive topic (22B–B′). Backgrounded or presentational focus readings are not licensed, and a focus reading forced by tol’ko ‘only’ is marked, just as in the case of polarity ellipsis (23).Importantly, Maša ne poslala, with no overt continuation in (22B), is acceptable only on a contrastive reading for the phrasal remnant that invokes an implicit continuation. So a backgrounded reading of Maša is unacceptable even without the continuation a Vasja poslal. Similarly, a backgrounded pronominal subject in (22B′) is also unacceptable.A reviewer brings up the interesting question of whether NPE/ object drop, when we can isolate it, may impose different discourse conditions on the subject than VVPE. Much more serious investigation is needed to address this question properly, but preliminarily, I would note that the discourse conditions on subjects in NPE/object drop versus VVPE seem to be partially overlapping, but not identical. They are partially overlapping in that backgrounded subjects are degraded even in NPE/object drop, if there is a linguistic antecedent.Because (24) involves dropping or elision of only the object in the VP of the response, leaving behind a PP argument, we can be reasonably sure that it is not a case of VVPE (which would have had to elide the PP argument as well as the direct object). Yet, uttering a pronominal subject in the response to (24A) is still degraded for many speakers. Despite this similarity, the discourse licensing conditions for subjects in VVPE and NPE/object drop cannot be said to be entirely identical: in argument drop with a situational antecedent (25)—which cannot be VVPE for lack of a linguistic antecedent—subjects are permitted, with either a backgrounded or perhaps a presentational focus interpretation.Finally, recall that in VVPE, focused phrasal remnants are unacceptable (23). In NPE object drop, however, they appear fully acceptable.As with (24), the presence of an argument PP in the response to (26A) is meant to guarantee that the response is an instance of NPE/object drop, rather than VVPE. Thus, the preliminary conclusion is that NPE/ object drop has fewer or different licensing conditions on subjects than does VVPE. VVPE permits only contrastive topics as phrasal remnants, whether they are subjects or not. NPE/object drop seems to permit focused subjects, and at least in the case of object drop, subjects with a backgrounded or presentational focus interpretation. This is not surprising if both VVPE and NPE/object drop are attested in Russian, as originally maintained in Gribanova, to appear.Returning now to VVPE, we can observe that the similar patterning of polarity ellipsis and VVPE is not entirely surprising, since the primary informational content of both a stranded verb (with or without negation) and a stranded yes/no particle is much the same: in both types of ellipsis, some previously made statement is being confirmed or denied. That the discourse structures of these types of ellipsis should be similar is, then, expected.In light of these patterns, what is unexpected is that AVPE behaves differently from polarity ellipsis and VVPE, allowing a broader array of interpretations for phrasal remnants. Kazenin (2006) provides an elegant structural account that accommodates the differences between polarity ellipsis and AVPE; it is possible that such an account may be extended to VVPE, though this is too big an endeavor to pursue here. The broader picture that emerges from this comparative discussion is that the discourse structure of ellipsis operations is an important component of any account that seeks to fully explain the structural possibilities involved in any sort of ellipsis.This squib has presented two pieces of evidence that an argument drop or NPE analysis of constructions like (1) would have difficulty accounting for. The first comes from conjunction and disjunction of multiconstituent elements under negation, which fits nicely with a VVPE analysis but poorly with an argument drop or NPE analysis. The second comes from the discourse structure of VVPE, which, like polarity ellipsis, does not permit subjects to be uttered unless they are contrastive. That VVPE and polarity ellipsis pattern together can be seen as a reaffirmation of the idea that both operations are the same in type: that is, that both involve ellipsis at the clausal level.From a broader perspective, once we include VVPE as a potential kind of ellipsis operation in Russian, the result is that Russian makes use of at least three distinct ellipsis operations at the clausal level, each with specific syntactic and discourse properties. As demonstrated—briefly and incompletely—in this squib, this prolific use of ellipsis within Russian provides an ideal opportunity to gain more insight into the comparative differences, discourse-related and structural, among these various kinds of ellipsis.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 131 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.