Towards a Theory of Radical Island Evasion Under Ellipsis Patrick D Elliott1 LEL Postgraduate Conference 22nd May 2013

1.

Introduction • Assumption: the ‘silent structure’ analysis of ellipsis, see e.g. Ross (1969), Merchant (2001) and Lasnik (2001) amongst others. (1)

Mary kissed someone last night, but i’m not sure who {she kissed}2

• Question? Is there anything special about silent structure? • Null hypothesis: No. • The claim (see, e.g. Lasnik (2001) and Merchant (2008)): Yes. Silent structure is special in that it can ‘repair’ island violating movement. – Has sweeping consequences for the notion of islandhood. • Structure of the argument: – Focusing on sluicing, where repair seems to be fairly unconstrained. – We must allow structural mismatches between the ellipsis site and its antecedent. – Allowing structural mismatches provides an alternative explanation for island ‘repair’: The island violation was never there to begin with. – No need for an e-site specific repair mechanism. More elegant. – The evasion account (suitably constrained) is empirically superior. – Conclusion: Repair is epiphenomenal. Move along, nothing to see here. 2.

Sluicing and Silent Structure

2.1 Some Useful Terminology • An example of a sluicing3 construction: (2)

[Jane married someone]A , but we’re not sure who {...}E

• who = remnant 1 [email protected]

- comments and criticism are welcome! are used to indicate elided material throughout. 3 Coinage due to Ross (1969). 2 Braces

2

Radical Island Evasion • someone = correlate • {...}E = e(llipsis)-site • [...]A = antecedent

2.2 Assumptions: • Broadly speaking, I adopt Merchant (2001)’s analysis of sluicing as movement + clausal ellipsis, conditioned by semantic identity. (3)

[Jane married someone]IPA , but we’re not sure who x { she married t }IPE

• Precise formulation of semantic identity: e-GIVENness. (4)

e-GIVENness (def.) An expression E counts as e-GIVEN iff E has a salient antecedent A, and modula ∃-type shifting4 ... a. b.

(5)

..A entails the f-clo(E) ..E entails the f-clo(A)

f-clo(sure) (def.) f-clo(α) is the result of replacing f-marked parts of α with ∃-type variables.

• e-GIVENness works applied to (3). – E = Jane married t – A = Jane married someone – f-clo(E) = ∃x s.t. Jane married x – f-clo(A) = ∃x s.t. Jane married x – f-clo(E) ⇐⇒ f-clo(A) (trivially) – E is e-GIVEN • Rought and ready version: E is e-GIVEN iff E entails A and vice-versa. • There are plenty of cases where it is obvious that we need semantic identity to account for structural mismatches: (6)

Decorating for the holidays is easy if you know how... a. *...{decorating for the holidays} b. ...{to decorate for the holidays} (Merchant (2001) p28)

• Despite to decorate for the holidays not having a syntactically isomorphic antecedent, it still counts as e-GIVEN. 4 ∃-type

shifting raises expressions to type and existentially binds unfilled arguments.

Radical Island Evasion

3

• There are some well-known problems with e-GIVENness. Consider the following example of VP ellipsis: (7)

*John will [beat someone at chess]A , and then Mary will {lose to someone at chess}E

• However, ∃x s.t. x will beat someone at chess entails that ∃x s.t. x will lose to someone at chess. • Plenty of competitors to e-GIVENness for the correct formulation of semantic identity, see e.g. Thoms (to appear) for one concrete proposal. • I am not so much concerned here with the precise formulation of the identity condition. What is important is that the identity condition is semantic, and therefore that there is some slack in the extent to which E can deviate structurally from A. • Despite its short-comings, e-GIVENness should be good enough for our purposes here. 2.3 Arguments for Movement + Deletion • Movement + deletion is not uncontroversial. • Recent proliferation of What You See Is What You Get (WYSIWYG) accounts. The implementations differ, but the primary claim is that there is no silent structure - the remnant is assigned its interpretation directly via some semantic mechanisms. See e.g. Culicover and Jackendoff (2005), Barker (2012) and Jacobson (2013). I’ll have a couple of things to say about this kind of approach later on. • Classical arguments for movement + deletion approach to sluicing: – Case-connectivity – Preposition-Stranding Generalisation (PSG) – ...and many more (see Merchant (2001) for extensive argumentation.) • Focus here will be on the PSG. • Basic argument: Availability of a P-less sluice tracks the availability of P-stranding movement in a given language. Receives a straightforward explanation under movement + deletion account. Explanation far from clear under WYSIWYG accounts. • Data from German: (8)

a.

[Mit x wem] hat Lena tPP geredet With who.DAT has L. t talked “With whom has Lena talked?”

b. *Wem x hat Lena mit tDP geredet Who.DAT has L. with t talked “Whom has Lena talked with?”

4

Radical Island Evasion

(9)

a.

Lena hat mit jemandem geredet, aber ich weiß nicht mit wem L. has with someone.DAT talked, but i know not with who.DAT “Lena has talked with someone, but i don’t know with whom”

b. *Lena hat mit jemandem geredet, aber ich weiß nicht wem L. has with someone.DAT talked, but i know not who.DAT “Lena has talked with someone, but i don’t know who”

• Question: Why would restrictions on sluicing pattern with restrictions on movement? • Obvious answer: Sluicing involves movement. • Also, no a-priori reason to reject movement + deletion. Evidence of phenomena involving phonological manipulation conditioned by semantic identity elsewhere: Case in point - Phonological reduction. • Phonological reduction subject to a similar (if somewhat looser) semantic identity condition. (10)

First John told M ARY that I insulted her, and then J ILL heard that i insulted her.

(11)

*First John told M ARY that I insulted her, and then J ILL had to go to the bathroom.

• ∃x s.t. John told x that I insulted x ≈> ∃x s.t. x heard that I insulted x • We can think of ellipsis as being a particularly radical form of phonological reduction, subject to a stronger identity condition. 3.

Island Repair

3.1 The Phenomenon • Examples such as the following have been used to motivate the claim that sluicing can repair island violations (see e.g. Lasnik (2001)). (12)

They want to hire someone who speaks a Balkan language, but i’m not sure which Balkan language... ...{they want to hire someone who speaks t } (Lasnik’s analysis)

• Lasnik presupposes that the ellipsis site and its antecedent must be structurally isomorphic - No significant deviation from the antecedent is allowed. • Spelling out the putative ellipsis site results in ungrammaticality. This is because a relative clause is an island for movement. (13)

*[Which Balkan x language] do they want to hire someone who speaks t.

Radical Island Evasion

5

• Note that there is nothing semantically incoherent about this question. The extracted element can be questioned in-situ; This really is a constraint on movement. (14)

They want to hire someone who speaks WHICH Balkan language?

• The claim: When the island violation is embedded in an ellipsis site, the violation is ‘repaired’ and the sentence is saved. • One consequence: A re-analysis of some (Merchant (2001)) or even all (Lasnik (2001)) islands as ‘PF-islands’. • Roughly speaking, island-violating movement proceeds successive-cyclically, leaving PF-uninterpretable traces which lead the derivation to crash if not deleted via ellipsis. The specific implementation of this (which has its own theoretical problems) is unimportant for our purposes, but see Merchant (2008) for a recent proposal. • In layman’s terms: It is only the fact that island violating movement is pronounced that leads to ungrammaticality. 3.2 Island Evasion • Consider again (12) repeated below as (15): (15)

[They want to hire someone who speaks a Balkan language]A , but i’m not sure which Balkan language {...}E

• The faulty premise here is that A dictates the form of E. • Rather, i will assume that A constrains what E can mean. A and E may deviate in form. This allows us to posit a non-isomorphic ‘short source’. No repair, but no island violation present to begin with: (16)

...i’m not sure which Balkan language {theye−type should speak t}E

• In lieu of a detailed semantics for e-type anaphora and modals, i will not attempt to show here that the short source in (16) satisfies e-GIVENness. See Merchant (2001) p297-299 for a detailed proof. • So far, i have not deviated much from the analysis of sluicing presented in Merchant (2001). Rather than claiming that all islands can be repaired, as in Lasnik (2001) and Merchant (2008), Merchant (2001) clusters islands into ‘PF-Islands’ and ‘Propositional Islands’. PF-islands are amenable to repair, whereas propositional islands must be evaded. I argue here Merchant (2001) doesn’t go far enough. Namely, i will argue for radical island evasion - no island violations are amenable to repair, but rather a set of non-uniform evasion strategies do the heavy-lifting. • We have already seen one evasion strategy - A short source may be used to evade a Relative Clause Island violation. In the next section i shall concentrate on another island constraint: The Left Branch Condition (LBC), analysed by both Merchant and Lasnik, amongst others, as a PFisland. I will argue that semantic identity allows us to evade LBC violations via a ‘predicational source’, and provide strong evidence that this is indeed the case.

6

Radical Island Evasion

4.

The Left Branch Condition and Island Evasion • The LBC (originally proposed in Ross (1967)) captures the fact that in many languages, the DP is an island for ‘left-branch’ modifiers, such as possessors (17) and attributive APs (18) (see esp. Corver (1990)). (17)

*Whose x did Jane kiss [t husband]?

(18)

*[Howxtall] did Mary marry [a t man]?

• Focus on adjectival Left Branch Extraction (LBE); Possessors independently licensed nominal ellipsis (19) - introducing a confounding factor - whereas attributive APs don’t (20). (19)

Jane kissed Mary’s husband, and Susan kissed Helen’s {husband}

(20)

*Mary married a tall man, and Jane married a short {man}

• Merchant (2001) argues based on data like (21) that LBEs are repaired under ellipsis. (21)

Mary married a tall man, but i’m not sure how tall {she married a t man} (Merchant’s analysis)

• Again, note that the analysis presupposes that the e-site and its antecedent must be isomorphic. Merchant does have his reasons for doing this (nothing in his analysis forces him into this position), but i won’t be exploring those reasons here. 4.1 Evidence from Morphosyntax • Given that in (21) the adjectival remnant how tall has been extracted from a modifier position, a repair account predicts that it should pattern in its morphosyntax with modifiers generally. • We shall see evidence from West Germanic and Hungarian suggesting that this is not in fact the case. Rather, the adjectival remnant patterns with a predicate. • Data from German (Dutch patterns in same way): (22)

Der Mann ist groß(*en). The man is tall(. ACC). “The man is tall”

(23)

Lena hat einen großen Mann geheiratet, aber ich weiß nicht wie groß(*.en) tall. ACC man married, but i know not how tall(*. ACC). L. has a “ Lena married a tall man, but i don’t know how tall.”

• Correlate/remnant mismatch: – Correlate = großen

Radical Island Evasion

7

– Remnant = groß • Hungarian data (Zoltan Galsi p.c.): (24)

a.

b.

(25)

John ismer n´eh´any magas(*ak) l´anyt J. knows some tall(*.PL) girls “John knows some tall girls.” A l´anyok magasak The girls tall.PL “The girls are tall.”

a. *John ismer n´eh´any magas l´anyt, de nem tudom milyen magas. J. knows some tall girls, but not know.I how tall “John knows some tall girls, but i don’t know how tall.” b. John ismer n´eh´any magas l´anyt, de nem tudom milyen magasak. J. knows some tall girls, but not know.I how tall.PL “John knows some tall girls, but i don’t know how tall.”

• Correlate/remnant mismatch: – Correlate = magas – Remnant = magasak • A cross-linguistic investigation into correlate/remnant mismatches in the adjectival domain is ongoing, but initial results seem quite suggestive. • Note that a similiar pattern can be found in the domain of fragment answers, which, following Merchant (2004) and Griffiths and Lipt´ak (2012), I also assume involve clausal ellipsis. (26)

a. b.

Lena hat einen WIE großen Mann geheiratet? L. has a HOW tall. ACC man married? Sechs Fuß groß(*en) Six foot tall(. ACC)

• Note that WYSIWYG accounts of clausal ellipsis would seem to make the wrong predictions here. Both Barker (2012) and Jacobson (2013) incorporate some manner of ‘brute-force’ mechanism for ensuring featural matching between correlate and remnant in order to derive case-matching effects. It is not at all clear why the very same mechanisms should not equally apply to adjectival remnant/correlate pairs, predicting similarly robust matching effects which are simply not attested. 4.2 Evasion Strategy: A Predicational Source • Morphosyntactically, an adjectival remnant patterns with a predicate rather than a modifier, even when its correlate is in a modifier position. This provides us with a clue as to the form that the ellipsis site takes5 . 5 It’s

quite common in the lit. to invoke a cleft source to account for otherwise mysterious phenomena (see e.g. Vicente (2008) on apparent P-stranding violations in Spanish.) but that won’t work here since predicates generally are degraded as pivots, e.g. ?/??It is TALL that John is.

8

Radical Island Evasion

• My concrete proposal is that a predicational copular clause (following Mikkelsen (2005)’s typology) underlies apparent repair of LBC-violations. C LAUSE T YPE Predicational Specificational Equative (27)

S UBJECT hei he,ti hei

C OMPLEMENT he,ti hei hei

Mary married a tall man, but I don’t know... ...[howxtall] {[DP the [NP man that Mary married]] was t}

• I assume the theory of e-type anaphora outlined in Elbourne (2001) (not crucial). Elbourne proposes that e-type anaphora spell-out a determiner with an elided NP complement. For all intents and purposes then, we have a DP in the e-site. • The predicational source is e-GIVEN (given certain assumptions about definite descriptions): – f-clo(A) = ∃d s.t. Mary married a d-tall man – f-clo(E) = ∃d s.t. The man Mary married was d-tall – f-clo(A) ⇔ f-clo(E) • Given that a predicational source is allowed by the grammar (assuming semantic identity) we might na¨ıvely expect to see mismatches in the adjectival domain across the board. This isn’t the case however. • Merchant (2001) shows that in Greek, an adjectival remnant must pattern with a modifier rather than a predicate. • This does not constitute evidence against my analysis however. Crucially, Greek differs from German in allowing post-adjectival NP ellipsis. This allows us to posit an isomorphic yet nonisland-violating source of the following kind (i shall use English words for the sake of exposition): (28)

Mary married a tall.AGR man, but i don’t know [how tall.AGR {a man}]DP {she married tDP }

• Greek therefore makes use of a different evasion strategy from German/Hungarian, involving island pied-piping (of the entire DP), followed by clausal ellipsis, together with an independent operation of NP ellipsis. • The fact that the adjectival remnant must pattern with a modifier is very revealing. It suggests that the availability of a non-isomorphic source is blocked by the possibility of an isomorphic source. There are various different ways of deriving this; One option would be to build it directly into the grammar. I will tentatively propose, however, that this blocking effect should follow from constraints on processing. • Arregui et al. (2006) argue that processing ellipsis involves two distinct parsing mechanisms: The syntactic parser, and the discourse parser. When the ellipsis-site is encountered, the syntactic parser takes a first pass, seeking out a suitable linguistic antecedent to fill the gap. If one cannot be found, the discourse parser steps in and generates a non-isomorphic source (conditioned, of course, by semantic identity). This captures the intuition that generating a non-isomorphic ellipsis site is

Radical Island Evasion

9

more costly than ‘slotting in’ a suitable linguistic antecedent. It follows that in a language like Greek, a non-isomorphic ellipsis site is not ruled out by the grammar, but it is deemed unacceptable due to an independent processing constraint. • The blocking effect can also be seen in simple constructions such as the following: (29)

Lena hat jemandem geholfen, aber ich weiß nicht... wem {sie t geholfen hat} who.DAT she t helped has b. *wer {es war t} who.NOM it was t a.

• Given that it has been argued forcefully in Vicente (2008) and elsewhere that semantic identity allows a non-isomorphic cleft source, as in (29)b., the fact that the corresponding sluice is ungrammatical is mysterious. I will suggest that (29)b. is blocked by the availability of an appropriate linguistic antecedent. 4.3 More Evidence: Non-Predicative Adjectives • Difference between predicative and non-predicative adjectives in left-branch sluices provides further evidence against isomorphic e-site and for a predicative source. Usefully, this diagnostic can be applied to languages without rich inflectional morphology (like English). • Predicative adjectives (e.g. diligent) and non-predicative adjectives (e.g. hard as in ‘hard-working’) appear as modifiers (30), but only predicative adjectives appear as predicates (31). Both kinds can be wh-fronted, so long as they’re gradable (32). (30)

Billy hired a diligent/hard worker.

(31)

This worker is diligent/*hard.

(32)

[How diligent/hard] a worker did Billy hire t?

• A non-predicative AP remnant is markedly degraded compared to a predicative counterpart (with a modifier correlate). This suggests that the remnant is merged as a predicate, ruling out an isomorphic e-site (presumably due to illicit Left Branch Extraction)6 . (33)

*?Billy hired a hard worker, but I don’t know how hard.

(34)

Billy hired a diligent worker, but I don’t know how diligent.

• Same can be shown for “old” as in “old friend/colleague”; for related data from fragments see e.g. Merchant (2004, 688). 6 Ad

Neeleman (p.c.) suggests one possible alternative explanation: [hard worker] could constitute an idiom chunk. (33) is degraded because it violates a strict adjacency requirement. Note however that the AP and NP are discontinuous in (32), and nonetheless the idiomatic interpretation is available, suggesting that adjacency is not pertinent.

10 Radical Island Evasion • Simplest explanation: sluicing never repairs LBEs, and ‘repaired’ LBEs have an underlying predicational structure. • Note furthermore that it’s not clear why non-predicative adjectives aren’t possible remnants under a WYSIWYG account, since the explanation relies on the structure of the e-site (and under a WYSIWYG account, there is no e-site). 4.4 More Evidence: Tag Questions • If we follow Griffiths and Lipt´ak (2012) in giving fragments the same analysis as sluicing, they provide indirect evidence for the presence of a predicational source with a left-branch remnant. • Pronoun subjects in tag questions agree with the subject of the previous assertion (Mikkelsen 2005). (35)

Bill is happy, isn’t he?/*isn’t it?

(36)

It was B ILL, wasn’t it?/*isn’t he?

• We can use tag questions to diagnose the structure of the clausal e-site underlying a left-branch fragment (see Barros and van Craenenbroeck (2013)): (37)

a. b.

Mary married a TALL man. Ah yes, EXTREMELY tall, wasn’t he?/*wasn’t it?/*didn’t she?

• The pattern of acceptability indicates that the putative source is predicational (38)

Predicational source He was EXTREMELY tall, wasn’t he?/*wasn’t it?/*didn’t she?

(39)

Cleft source It was an EXTREMELY tall man, wasn’t it?/*wasn’t he?/*didn’t she?

(40)

Isomorphic source She married an EXTREMELY tall man, didn’t she?/*wasn’t he?/*wasn’t it?

• We conclude that the clausal ellipsis underlying fragments does not repair illicit LBE, rather, the island violation is evaded via a non-isomorphic predicational e-site (41)

5.

Ah yes, EXTREMELY tall {he was t} x

Outlook and Conclusion • The evasion account of left-branch sluices fits in with a a larger body of work treating ‘repair’ phenomena as being largely epiphenomenal (see e.g. Fukaya (2007), Abels (2011) and Barros (to appear))

Radical Island Evasion

11

• Other proposed evasion strategies: – Short sources explain ‘repair’ of relative clause and other islands in sluicing and fragment answers (Merchant (2001), Barros, Elliott and Thoms (2013)) – When short sources are ruled out (e.g. when the correlate is an NPI - see Lasnik (2001)), a cleft source can be invoked to explain the ‘repair’ (Barros (to appear)). – Clausal island pied-piping can explain asjunct island ‘repair’ (Barros, Elliott and Thoms (2013)). – Variability of ‘repair’ in contrastive fragments can be explained due to the independent unavailability of cleft sources (Barros, Elliott and Thoms (2013)). – Cleft sources explain apparent exceptions to the P-stranding generalistion (van Craenenbroeck (2012)). • The identity condition on ellipsis must be semantic, to allow for the flexibility necessary to generate non-isomorphic ellipsis sites. • Repair phenomena can be explained on the basis of an inventory of evasion strategies, each following independently from existing resources. No need to posit any ellipsis specific repair mechanisms. • Island repair is epiphenomenal.

References Abels, Klaus. 2011. Don’t fix that island! it ain’t broke, paper presented at the Islands in Contemporary Linguistic Theory conference, Vitoria-Gasteiz. Arregui, A., Clifton Jr, C., Frazier, L., and Moulton, K. 2006. Processing elided verb phrases with flawed antecedents: The recycling hypothesis. Journal of Memory and Language 55(2):232–246. Barker, Chris. 2012. Scopability and Sluicing. In Press. Barros, Matthew. to appear. A non-repair approach to island-sensitivity in contrastive TP ellipsis. In Proceedings of Chicago Linguistics Society 48. Barros, Matthew, Elliott, Patrick D., Thoms, Gary. 2013. More variation in island repair: The clausal/non-clausal distinction. Presented at CLS 49. Barros, Matthew, and van Craenenbroeck, Jeroen. 2013. Tag questions and ellipsis. Presented at DGfS AG4: Parenthesis and ellipsis: Cross-linguistic and theoretical perspectives. Corver, Norbert. 1990. The syntax of left branch extractions. Ph.D. thesis, Tilburg University. Culicover, Peter W., and Jackendoff, Ray. 2005. Simpler Syntax. Oxford: Oxford University Press. Elbourne, Paul. 2001. E-type anaphora as NP-deletion. Natural Language Semantics 9(3):241–288. Fukaya, Teruhiko. 2007. Sluicing and stripping in Japanese and some implications. Ph.D. thesis, University of Southern California. Griffiths, James and Lipt´ak, Anik´o. 2012. Contrast and island-sensitivity in clausal ellipsis. Syntax. Jacobson, Pauline. 2013. The short answer: Implications for Direction Compositionality (and vice-versa). ms. Lasnik, Howard. 2001. When can you save a structure by destroying it?. In PROCEEDINGS-NELS 31(2):301–320. Merchant, Jason. 2001. The syntax of silence: sluicing, islands, and the theory of ellipsis. Oxford: Oxford University Press. Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and philosophy 27:661–738.

12 Radical Island Evasion

Merchant, Jason. 2008. Variable island repair under ellipsis. In Kyle Johnson, ed., Topics in Ellipsis, 132–153, Cambridge, United Kingdom: Cambridge University Press. Mikkelsen, Line. 2005. Copular clauses: specification, predication, and equation. Amsterdam: John Benjamins B.V. Ross, John Robert. 1967. Constraints on variables in syntax. Doctoral Dissertation. MIT. Cambridge MA. Ross, John Robert. 1969. Guess who? In Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan, eds., Chicago Linguistics Society, 252–286, Chicago, Illinois. Thoms, Gary. to appear. Lexical mismatches in ellipsis and the identity condition. In Proceedings of NELS 42. van Craenenbroeck, Jeroen. 2012. Ellipsis, identity and accommodation. Ms. Vicente, Luis. 2008. Syntactic isomorphism and non-isomorphism under ellipsis. Ms., University of California, Santa Cruz.