2015/07/01 by Andrés Saab
Arts and Humanities · Computer Science · Psychology · #Natural Language Processing Techniques #Phonetics and Phonology Research #Syntax, Semantics, Linguistic Variation
paper · pdf · doi:10.1162/ling_a_00192
crossref issued 2015/07/01 · crossref published 2015/07/01 · crossref published-print 2015/07/01 · openalex publication_date 2015/07/01 · crossref created 2015/07/09 · crossref deposited 2024/06/09 · openalex created_date 2025/10/10 · crossref indexed 2026/07/29 · openalex updated_date 2026/07/31
In this squib, I discuss an ellipsis puzzle raised by Barros (2012), involving else-modification in sluicing contexts. On the basis of empirical evidence from Spanish, I conclude that an island repair approach to this puzzle is superior to at least two alternative analyses.Barros (2012) shows that a nonisomorphic strategy should be available to resolve some elliptical sluicing examples involving else-modification in English.If the elliptical gap in (1) had an underlying structure containing something similar to (1a), then we would expect a semantic clash, given that Jack likes Sally in the antecedent counts as a partial answer to the question I do not know who he likes. Assuming that being in a not-knowing situation with respect to a question Q implies not having any partial answer to Q, (1a) is derived as a kind of semantic inconsistency (Romero 1998). Such a semantic clash vanishes in the short copulative strategy in (1b). (See Barros 2012 for an explicit formulation.) In any case, our intuition as speakers is that only (1b) is a suitable nonelliptical counterpart for (1).This experiment can be replicated in Spanish with interesting additional consequences. Thus, in (2a) we should assume that the elliptical site includes an underlying form containing a short copulative structure that is nonisomorphic with respect to its antecedent (see (2b)), because an isomorphic resolution would introduce the semantic clash found in (1a) for English (see (2c)).It should be noted that a full cleft is also odd here for the semantic reasons discussed above.Therefore, like English, Spanish resolves some elliptical sites as short copulatives (i.e., pseudosluicing is attested in the language). Consider (4), however, which contains a differentially marked object.Here, as in (2b), an isomorphic resolution is semantically inconsistent (see (4b)). The problem, now, is that a nonisomorphic truncated cleft sentence is simply ungrammatical, given that the verb ser ‘to be’ is incompatible with a differential-object-marking (DOM) object (see (5a)).1 Once again, a full cleft strategy is not available here because, even if the result were syntactically well-formed, it would reintroduce the semantic clash already mentioned (see (5b)).2A semantically consistent and syntactically isomorphic strategy would be (a) to take the full coordinate structure in (4) as the antecedent for the sluicing sentence in (4a) and (b) to extract the DOM object from the second conjunct, violating the second part of the Coordinate Structure Constraint (CSC) as originally formulated by Ross (1967).Implementing the strategy just sketched, the underlying structure for an example like (4a) would be the one illustrated in (7) (paceMerchant 2001). (E = elided)That this is a semantically consistent analysis follows from the fact that the antecedent is not a partial answer to the question expressed in the complement of the verb saber ‘know’. The reason for this is the presence of the indefinite in the second conjunct: bare indefinites in general cannot be partial answers to questions.The oddness of (8A′) follows if such an answer is not part of the set of partial answers denoted by Q. The same argument can be extended to (7), but, of course, in this case the question cannot be expressed in its full form because doing so would violate the CSC. A semantically similar question could be the following:(9A′) shows that the antecedent in (7) cannot be in the set of partial answers to the particular question expressed in the elliptical constituent. Then, we can safely conclude that this antecedent is the only one that is syntactically isomorphic and semantically consistent with the elliptical site in (7). If this is correct, then this example should be considered a case of island repair (paceBarros, Elliott, and Thoms’s (2014) conclusion against island repair in general), unless other semantic mechanisms are allowed.For instance, a mutual-entailment approach (see Merchant 2001 and much subsequent work) would claim that a sluicing example like (4a) could have the following underlying structure in the elliptical site:As noted by a reviewer, it seems easy to obtain the right equivalences in this case under a mutual-entailment analysis, according to which ellipsis may apply whenever the antecedent and the elided phrase are mutually entailed, regardless of syntactic isomorphism. As shown by Romero (1998), else-modification allows for a ‘besides’ reading under which the individual argument modified by else must be distinct from some salient individual provided by the (linguistic) context. Consider the following example from Romero 1998:51:The semantics for the else than–expression is as follows (Romero 1998:52):According to the denotation in (12) and the semantics of (11), the only felicitous reading for this sentence is that the argument introduced by else is the entity or individual (we call) Harry. This way, the question denied in (11) (‘‘to which individuals—besides Harry—she talked’’) does not introduce any contradiction with knowing that she talked to Harry. If the entity returned by such an argument were distinct from Harry, then the result would be clearly infelicitous.Returning to (10), we can see that the antecedent [A kissing someone else (than x)], where x = María, entails the elided constituent [E kissing t besides María], where t is the trace of the wh-remnant interpreted, under standard assumptions, as an indefinite (Chung, Ladusaw, and McCloskey 1995, among many others), and vice versa.There are, however, strong reasons to reject the semantic alternative just mentioned. The evidence comes from a hitherto unnoticed asymmetry in Spanish involving pre- and postverbal coordinated subjects and the first part of the CSC in (6) (see below for additional discussion of the same asymmetry involving the second part of the CSC). Consider the following examples:The sentences without más ‘else’ in the sluice remnants are strongly infelicitous, whereas adding más ‘else’, depending on the main focus of the sentence, is felicitous under two readings: that is, in the case of (13), that I do not know who else—besides Juan—caused a scandal, or that I do not know who else caused a scandal on top of the two persons that I know who did it.I would like to claim that this asymmetry follows from the fact that extraction of a conjunct from a preverbal position is not allowed. Thus, (13) with más ‘else’ present in the remnant is semantically consistent under the two relevant readings because no extraction from a coordinated structure is needed for the relevant readings to obtain.By contrast, for the relevant reading to obtain under the absence of más ‘else’ in the remnant, the only strategy would be to extract a conjunct from the preverbal subject, namely, the wh-remnant, which is impossible.It follows, then, that the sentences in (13)–(15) produce the semantic anomaly already discussed above. Interestingly, the sentences without más ‘else’ in the wh-remnant in (13)–(15) improve considerably when conjunct extraction takes places from a postverbal position, with a preference, for some speakers, for extractions in unaccusative environments.Crucially, this contrast cannot be attributed to some utterancefinal effect (see Barros, Elliott, and Thoms 2014), maybe linked to the focus property of final constituents in languages like Spanish. The sentences are perfect if one splits the coordinated subjects by distributing them in preverbal position across different clauses. I will illustrate this point with (14)/(19), although the same effects are obtained with (13)/(18) and (15)/(20).Therefore, this new set of contrasts must be attributed to the ban on conjunct extraction from a preverbal position. This prohibition could result from a freezing effect associated with some preverbal Ā-constituents. If the relevant freezing effect is determined at the conceptual-intentional interface (Gallego 2009, paceRizzi 2006), then we can conclude that island repair, conceived of as a PF phenomenon, cannot be at play here.3It is worth noting that Barros’s (2012) pseudosluicing strategy cannot be used here, as DOM objects show exactly the same effect as pre- and postverbal subjects.The sentence in (22a) contains a topical DOM object, whereas in (22b) the same coordinated object remains in situ. Again, we see that Barros’s effects do not obtain with the preverbal object. Notice that this set of data seems to be enough to reject the radical semantic analysis discussed earlier: a nonelliptical continuation with además ‘besides’ in the last clause is perfectly grammatical in the same contexts. For the sake of brevity, I only illustrate this with (14).Importantly, this analysis does not necessarily extend to extraction from subjects in EPP languages like English, where, paceRizzi (2006), a different analysis can be assumed, one in which the preverbal subject is PF-derived (see Merchant 2001 for a first suggestion and Van Craenenbroeck and Den Dikken 2006 for more evidence). Spanish preverbal subjects, instead, are not EPP-driven: they seem to be more amenable to an Ā analysis as proposed in Ordóñez 1997 and much subsequent work. Such a derived Ā-position would then be triggered by grammatical and discourse factors not connected to the EPP property in any relevant sense.Interestingly, if Merchant’s (2001) original analysis for derived islands in English, according to which EPP features triggering such derived positions are canceled under ellipsis at PF, is on the right track, then we would predict a contrast between preverbal coordinated subjects in English and Spanish. An anonymous reviewer notes that this is indeed borne out: (24), where a preverbal coordinated subject contains an else-correlate, admits the absence of else in the sluice remnant with the relevant reading (i.e., that I do not know who was the other person John was singing with).The reviewer reports the same judgment for Brazilian Portuguese, another language in which the EPP seems to be at work for the T node (see Barbosa, Duarte, and Kato 2005 for extensive discussion).Thus, in languages in which preverbal subjects are EPP-driven, a solution of the type proposed by Merchant (2001) is available. Such a solution is not tenable in languages in which pre- and postverbal subject asymmetries are triggered by information structure considerations, such as Spanish.This contrast between coordinated DPs in pre- and postverbal positions also seems to affect the second part of the CSC involving extraction from coordinated CPs (see (6)), even though judgments are more subtle here, maybe because of the heaviness of coordinated CPs. Yet there is a clear difference between (26) and (27).Whenever the coordinated CPs are in preverbal position, as in (26), the absence of más ‘else’ in the remnant gives rise to a judgment of semantic inconsistency, which does not arise when the CPs are in postverbal position, as in (27). Notice again that the problem is the coordinated structure in preverbal position and not preverbal CPs per se, which allow for a resolution of the ellipsis site that does not involve extraction from a preverbal CP.Like the strategy in (16), this type of evasion strategy constitutes an instance of what Barros, Elliott, and Thoms (2014) call short strategies for other cases of putative island repair across languages. This conclusion is forced by the fact that subject CPs in topic position are islands for extraction, as witnessed by the contrast between (29) and (30).The short strategy is impossible for (26) and (27) without más ‘else’ because in these cases the use of such a strategy would introduce a semantic inconsistency. The island repair solution, according to which extraction from a conjunct is allowed by ellipsis, is semantically consistent but inapplicable to topic preverbal positions for the reasons adduced above. Thus, (27) arises as the only available option also when two CPs are coordinated. Finally, notice that these facts would be incompatible with the mutual-entailment approach already discussed here, because, again, a nonelliptical continuation of (26) containing además ‘besides’ is felicitous.So far, the island repair solution seems to be the only option for resolving the elliptical sites in cases similar to (4a). But of course things are more complex and, before any conclusion is reached, we should discuss a potential counterexample to the island repair analysis raised by two anonymous reviewers and Gary Thoms ( pers. comm.).4As we will see, the solution to this problem connects with syntactic isomorphism in interesting ways. In my opinion, the putative counterexample involves split antecedents.As noted by a reviewer, a full isomorphic continuation here, containing just the verb and the trace of the wh-element, would be as infelicitous as (4b) is. And a pseudosluice source is unlikely for the same reasons as in (5a). I would like to claim that cases like (32) and similar ones are instances of split antecedents that should be resolved by whatever theory resolves well-known cases of split antecedents like (33) (Elbourne 2001, 2008:214).Returning to (32), both reviewers suggest that my analysis must depart from isomorphism in order to account for it, given that a coordinate structure must be postulated in the elliptical site that is absent in the antecedent(s).Although obviously I am not forced to assume such an underlying structure, I would like to suggest in a rather preliminary way that this is, indeed, the underlying structure in the elliptical site of (32). This is not a radical departure from syntactic isomorphism if structural accommodation of antecedents is allowed (e.g., Fox 2000, Van Craenenbroeck 2013, Thoms 2013). In particular, I suggest that a slight modification of Van Craenenbroeck’s (2013:19) assumption in (35) would be enough to account for split antecedents like this.By elements . . . freely available in any discourse, Van Craenenbroeck, following Merchant (2004, 2010), refers to expletives and copulas like be that are easily accommodated in pseudosluicing cases or in discourse-initial fragments, for instance. I conjecture that a conjunction like and forms a natural class with the set of elements that are freely available for accommodation. Indeed, and and be, as linker elements, seem to share some basic properties. For instance, they are neither θ-role nor Case assigners. If this is correct, then the elliptical site in (34) would be semantically consistent and syntactically isomorphic to an accommodated antecedent built up from the two nonelliptical sentences in (34) and the conjunction y/and. This, of course, constitutes a departure from syntactic isomorphism, but it is independently needed for other well-known cases of split antecedents, as already shown for (33).It is important to emphasize that I am not claiming to have a theory of split antecedents. There are indeed other options to explore in this respect, Elbourne’s (2008) analysis of split antecedents being a serious candidate. Elbourne proposes that VP- and NP-ellipsis sites are definite descriptions that can take an operator ANDn, which in turn allows resolving the split antecedent cases by taking n TP meanings as arguments and mapping them into the characteristic set of functions that has these meanings as atoms. Interestingly, this analysis could be extended to deal with the cases being explored here without resort to island repair. However, there are two basic problems that such an analysis should resolve before we can take it as a serious alternative. The first involves the pre- and postverbal asymmetries noted in connection with the radical semantic approach discussed above (see (13)–(15) and (18)–(20)). The second is related to the basic assumptions of Elbourne’s theory. Indeed, if VP-ellipsis is analyzed as [vP v [THEP THE RP . . . ANDP . . . ]] (see Elbourne 2008 for details), then every extraction from a VP-ellipsis site must be seen as a case of island repair, given that definite DPs are islands for extraction. Therefore, Elbourne’s approach to ellipsis not only fails to resolve the problem of island repair under ellipsis, but in fact magnifies it. As far as I can tell, some modifications of the syntax of VP-ellipsis proposed by Elbourne must be reconsidered in order to resolve these issues, a task I leave for future research (see Saab and Vicente in preparation, however, for an adaptation of Elbourne’s theory to sluicing).5Interestingly, the need to accommodate sentences like (32) follows the same governing rule as other cases of split antecedents. As is well-known, a split antecedent reading is triggered in cases like (36) but not in cases like (37) (Fiengo and May 1994apudElbourne 2008: 208).Now, consider (38), which is similar to (32), with the exception that (38) has a different subject for each nonelliptical clause.As in (37), the natural reading here is that I do not know who Pedro kissed; that is, there is no need to postulate an underlying coordination in the ellipsis site. This would follow from Elbourne’s rule: ‘‘Split antecedent readings are available only if the context gives the audience some reason to entertain them’’ (Elbourne 2008:208). Translating this rule to island repair phenomena in general, I suggest that island repair effects obtain only when there are strong reasons to entertain them. For the cases explored in this squib, it seems that the reason is semantic consistency.Summing up, in this squib I have discussed three solutions to Barros’s (2012) example in (1): (a) Barros’s own solution in (1b) in terms of a copulative strategy; (b) a mutual-entailment approach of the type proposed by Merchant (2001) (see (10)); and (c) an island repair solution (see (7)). Solution (a) is strongly disconfirmed by empirical evidence from Hungarian (Lipták 2013) and, as I have shown here, from Spanish. I have also rejected solution (b) on the basis of new empirical considerations involving pre- and postverbal coordinated subjects in Spanish. To conclude, then, the patterns analyzed here seem to show that some elliptical sites contain islands in their underlying form—in other words, that solution (c) is preferable to its competitors. This, of course, does not force us to accept any of the existing theories of island repair; it forces us only to accept the very existence of the phenomenon that distinguishes long-distance dependencies in elliptical and nonelliptical structures.