Modal complement ellipsis and deletion in narrow syntax Keywords: (VP) ellipsis, narrow syntax deletion, Dutch vs English 1. INTRODUCTION This paper starts out from the novel observation that Dutch displays a limited kind of VP ellipsis in the complement of deontic modal verbs (modal complement ellipsis, or MCE), as in (1). I claim (a) that these data involve ellipsis of VoiceP, (b) that this ellipsis process takes place in narrow syntax and (c) that ellipsis licensing is subject to syntactic locality, not adjacency. As a result, the projections inbetween the licensing head and the elided constituent play a crucial role in determining what can be extracted out of the ellipsis site. 2. PUZZLE: PROFORM OR DELETION? There are – at least – two possible analyses for the phenomenon in (1): deletion of syntactic structure or a null proform. A central argument to decide between the two concerns the possibility of extraction out of the ellipsis site. If such extraction is allowed, there must be enough syntactic structure present to host the trace; if extraction is impossible, this can be attributed to the lack of internal syntactic structure, i.e. the presence of a proform. Dutch MCE disrupts this simple picture, however: while objects cannot be extracted out of the ellipsis site (cf. (2)), subjects can (cf. (3)). Since modals are raising verbs (see Wurmbrand 2003 for argumentation), the subject in (3) has indeed been extracted out of the ellipsis site. Hence, the data in (2) and (3) taken together suggest that neither deletion nor a proform approach gives us the right result. I show, however, that an analysis involving deletion in narrow syntax can account for these conflicting extraction data. 3. ANALYSIS Deletion in narrow syntax – Deletion in narrow syntax comes about as follows: the head H of the constituent that will be elided bears an interpretable ellipsis feature [iE] (comparable to Merchant’s 2001, 2004 [E]-feature). The ellipsis-licensing head L bears a matching uninterpretable ellipsis feature [uE]. When L is merged, its ellipsis feature is checked via Agree and HP is elided in narrow syntax (Baltin 2007). As a result, the ellipsis site is no longer accessible for any syntactic operations after merger of L. An important corollary of this analysis, then, is that only constituents that move out of the ellipsis site prior to the merger of the licensing head can survive the deletion process. Dutch MCE – For Dutch MCE the modal is the licensing head bearing [uE] and Voice° bears an interpretable ellipsis-feature. I argue that modals select a non-finite TP complement, giving us the structure in (4). Observe that TP is the only projection inbetween the ellipsis site and the licensing head. Consequently, only subjects, which move to [Spec,TP], can extract out of the ellipsis site and survive the ellipsis. As there is no landing site for the object outside of VoiceP prior to the merger of the ellipsis licensing modal, the object cannot be extracted. Hence, the contrast between subject and object extraction is derived in a straightforward way. 4. EXTENSION OF THE ANALYSIS: ENGLISH VP ELLIPSIS (VPE) In English VPE, both objects (cf. (5)) and subjects (cf. (6)) can be extracted out of the ellipsis site (cf. Schuyler 2002, Merchant to appear a). The difference between Dutch MCE and English VPE can now be linked to the size of the deleted constituent and the position of the licensing head: the head licensing English VPE is the modal or auxiliary in T°, and v° is the head bearing the [iE]-feature (see Merchant 2007, to appear for argumentation that English VPE involves deletion of vP). In this case the projection inbetween the ellipsis site and the licensing head is VoiceP (cf. (6)). Given that Voice° is a phase head (Baltin 2007), its specifier provides an escape hatch for both subjects and objects. Time permitting, I further extend this analysis to other elliptical constructions such as sluicing, stripping, pseudogapping, British English do and NP-ellipsis.VP V’ V° TP modal [uE] T’  deletion in narrow syntax T° VoiceP Voice’ Voice° … [iE] Agree TP T’ T° VoiceP phase head  escape hatch [uE] Voice’  ellipsis site Voice° vP v’ v° [iE] Agree Examples (1) A: Wie wast er vanavond af? – B: Ik kan niet [ afwassen vanavond]. who washes there tonight off I can not off.wash tonight ‘Who is doing the dishes tonight?’ – ‘I can’t.’ (Dutch) (2) A: Wat gaat Tim aan Lara geven? – B: * Wat moet hij [aan Lara twat geven]? what goes Tim to Lara give what must he to Lara give INTENDED READING: ‘What’s Tim going to give Lara?’–‘What should he?’ (Dutch) (3) Deze broek moet niet gewassen worden, maar die rok moet wel [ tdie rok gewassen worden]. this pants must not washed become but that skirt must PRT washed become ‘These pants don’t need to be washed, but that skirt does.’ (Dutch) (4) (5) What is Tom going to buy? – I don’t know. What should he [buy twhat]? (6) Mina wasn’t arrested, but she should be [arrested tPeter]. (7) References BALTIN, M. (2007). Deletion versus Pro-forms: a false dichotomy?. MS New York University. MERCHANT, J. (2001). The syntax of silence. Sluicing, islands and the theory of ellipsis. OUP. MERCHANT, Jason (2004). “Fragments and Ellipsis”. In: Linguistics and Philosophy 27, pp. 661- 738. MERCHANT, J. (2007). Voice and ellipsis. Ms. University of Chicago. MERCHANT, J. (to appear). “Variable Island Repair under Ellipsis”. In: K. Johnson (ed.), Topics in Ellipsis. CUP. SCHUYLER, T. (2002). Wh-movement out of the site of VP Ellipsis. MA Thesis UCSC. WURMBRAND, S. (2003). Infinitives: Restructuring and Clause Structure. Mouton de Gruyter: Berlin/New