Contextual restrictions on ellipsis sites.

SAAB, ANDRÉS; LUIS VICENTE; TROY MESSICK

Campinas

Workshop; Grasping Ellipsis. Its syntax, semantics, acquisition and processing; 2015

Elbourne 2008 proposes that the lexical content of VP and NP ellipsis sites is necessarily embedded under an AND operator as part of the nuclear scope of a THE operator, whose restrictor is a contextual variable R (von Fintel 1994 et seq). We propose that extending this structure to sluicing, and especially assuming the presence of a contextual restriction, affords new insights into various difficult cases.The lexical entries of AND, THE, and R are those of Elbourne, modified only so that they make reference to TP meanings (see Elbourne 2008 for details; we also follow him in assuming that AND, THE, and R are a specific implementation of Merchant´s [E] feature, so as to ensure that they don´t appear in unelided TPs). The goal of AND is to help analyze sluices with split antecedents: it takes n TP meanings as arguments and maps them into the characteristic set of functions that has these meanings as atoms. As its contribution is trivial in sluices with a single antecedent, we ignore it in cases 1 and 2 below. The real work is done by the definiteness operator THE and especially its contextual restrictor R; specifically, R picks up a salient situation s in the preceding discourse, and THE ensures that s is the situation relevant to the interpretation of the sluice. Thus, Jack kissed someone, but I don´t know who can be paraphrased as "there is a situation s of Jack kissing someone, and I don´t know the identity of the individual kissed in s".Case 1: "Else"-modification. Barros 2012 argues that (1a), where the sluice has an "else"-modified antecedent, is an instance of pseudosluicing. While a cleft source (1b) is congruent, an isomorphic source (1c) is not; on the assumption that sluicing can´t repair the incongruence of (1c), Barros concludes that (1a) must stem from (1b). However, this analysis fails to cover examples analogous to (1a), but where the remnant is not a licit cleft pivot (2a)/(2b). We focus here on Spanish objects bearing the differential case marker "a" (glossed DCM), but the problem is fully generalizable to other languages and environments (Lipták 2013).(1) a. Al kissed Beth, and he also kissed someone else, but I don´t know who. b .Al kissed Beth, and he also kissed someone else, but I don´t know who it was. c. # Al kissed Beth, and he also kissed someone else, but I don´t know who Al kissed.(2) Juan besó a María, y también besó a alguien más, a. pero no sé a quién. b. *pero no sé a quién es. c. # pero no sé a quién besó. Contra Barros, we assume that (2a) does stem from (2c), using the contextual restriction introduced by [THE R] to repair the incongruence of (2c). Note that when such restrictions are made explicit with time/place modifiers (2c) becomes congruent (this also holds for English).(3) Juan besó a María el lunes, y también besó a alguien más el martes, pero no sé a quién besó el martes. Similarly, [THE R] allows (2a) to be paraphrased as "there is a situation s of Jack kissing Sally, and there is a situation s´ of Jack kissing someone that is not Sally, and I don´t know the identity of the individual kissed in s´". This is the observed reading of (2a); additionally, the presence of differential case marking on the remnant follows directly from the fact that the sluicing site stems from a regular interrogative (rather than a cleft).Case 2: Opaque contexts. Our analysis also predicts that a sluice will take the "wrong" antecedent if the "correct" one is not salient enough for R to pick up. We illustrate this prediction with the complements of weak assertion predicates (e.g., say or think), which can´t be sluiced (4a) even when a licit sluicing source appears to be available (4b).(4) a. # Jack says/thinks that he kissed Sally, but I don´t know who. b. Jack says/thinks that he kissed Sally, but I don´t know who he (actually) kissed.Because the complements of say and think are referentially opaque, R in (4a) can´t refer back to the situation of Jack kissing Sally. This blocks (4b) as a possible sluicing source. The only salient situation that R can pick up is that of Jack saying/thinking that he kissed Sally; however, the resolution of the sluice produces an incongruent meaning, as the infelicity of (5) shows.(5) # Jack says/thinks that he kissed Sally, but I don´t know who he says/thinks that he (actually) kissed.Case 3: Split antecedents. Elbourne uses the AND operator to account for VP and NP ellipses with split antecedents (6). Sluicing allows analogous examples (7).(6) Whenever Max uses the Xerox or Oscar uses the fax, I can´t. [= if Max uses the Xerox, I can´t use it, and if Oscar uses the fax, I can´t use it either](7) Whenever Jack wants to interview an athlete or Sally wants to profile an actress, the magazine editor asks which. [= if Jack wants to interview an athlete, the editor asks which one he wants to interview, and if Sally wants to profile an actress, the editor asks which one she wants to profile](7) can´t be modeled as a simple disjunction (plus ATB extraction of which) without the mediation of AND, THE, and R (i.e., the editor asks which (athlete/actress) Jack wants to interview or Sally wants to profile). Such an analysis would incorrectly predict a reading where, if Jack wants to interview an athlete, the editor asks which actress María wants to profile (and vice versa). Instead, we propose a direct extension of the analysis that Elbourne sketches for (6). We begin by treating whenever as a quantifier over sets of conditional antecedents, which are then composed point wise with the consequent (Rawlins 2008). In the sluice, AND maps the two antecedent TPs (Jack wants to interview an athlete and Sally wants to profile an actress) into the pertinent characteristic set of TP meanings. The function of [THE R] is then to restrict the interpretation each atom of this set to the situation(s) defined in the corresponding set of conditional antecedents. This results in the (correct) paraphrase of (7) indicated above.Outlook. Narrowly, our analysis shows that Elbourne´s analysis of VP and NP ellipsis with split antecedents can be extended to sluicing (a result that Elbourne 2008:fn. 18 hints at only very tangentially). More broadly, we show that Elbourne´s machinery, and especially the presence of contextual restriction, is necessary even in the absence of split antecedents (a result that Elbourne assumes as correct but doesn´t provide direct evidence for). This reinforces Elbourne´s conjecture that ellipsis sites at large ought to be treated as contextually restricted definite descriptions, with the associated implications for the resolution of ellipsis.Barros 12 Else-modification as a diagnosis for pseudosluicing, NELS 43 Elbourne 08 Ellipsis sites as definite descriptions, LI 39 von Fintel 94 Quantifier domain restriction, PhD, Umass Hankamer & Sag 76 Deep and surface anaphora, LI 7 Lipták 13 A note on overt case marking as evidence for pseudosluicing, Ms., Leiden Rawlins 08 (Un)conditionals, PhD UCSC.