• No results found

In this section, I discuss Thoms’ (2013) theory of EM. His proposal is based on LF parallelism and builds on the works by Fox and Lasnik (2003) and Griffiths and Lip- ták (2014). I provide a summary of the account in the next section. In section 3.2.2, I show that Thoms’ LF parallelism theory suffers from conceptual and empirical problems. Moreover, I show that it does not meet the requirements on a theory of EM (cf. (40)).

3.2.1 Thoms (2013)

Thoms (2013) argues that EM is not constrained by syntax, but that its application is subject to an LF parallelism constraint, as informally stated in (42). It follows from

the definition in (42) that LF parallelism is a condition that is only active under ellipsis.

(42) LF parallelism (Fox and Lasnik, 2003):

Variables in the antecedent A and the elliptical clause E must be bound from parallel positions.

Thoms shows that LF parallelism captures why, in ellipsis constructions with more than one remnant, the two remnants must be clause mates (the Clause Mate Con- dition, Lasnik 2013). Consider the contrast between the Multiple Sluicing examples in (43) and (44). In both (43) and (44), thewh-phrasewhatin the antecedent takes scope over the matrix clause at LF. As dictated by LF parallelism, the trace/variable of the remnant in the ellipsis site must be bound from the same position. This is the case in (43), wherea booktakes matrix scope. In (44), on the other hand, thewh- phrasewhattakes clause bound scope at LF in the antecedent (cf. Dayal, 2002). The corresponding remnant in the ellipsis site,a book, on the other hand, takes scope over the matrix clause, where it binds its trace. LF parallelism is thus not satisfied in (44) and the sentence is therefore ungrammatical.4

(43) a. A: Who bought what?

B: John a book (and Mary a pencil). b. LFA[[who]x.[what]y.[xi bought yj]]

LFE[[John]x.[a book]y.[xibought yj]]

(44) a. A: Who said you bought what? B:∗John a book (and Mary a pencil).

b. LFA[[who]x.[xisaid[what]j λy.[you boughtyj]]

LFE[[John]x.[a book]y.[xi said you boughtyj]]

Next, we consider elliptical structures with a single remnant, comparing cases in which there is a contrastive correlate in the antecedent with cases where there is a non-contrastive correlate in the antecedent. To begin with the latter, consider (45), which features a non-contrastive correlatea Balkan language. This correlate can be bound in situ from the matrix clause via choice function mechanisms (cf. Rein- hart, 1997). The fact that the correlate takes matrix scope in the antecedent, allows the corresponding remnant,Serbo-Croatianto move to, and take scope from, the corresponding position in the ellipsis clause (even crossing an island boundary).

(45) a. A: I heard they hired someone who speaks a Balkan language fluently. B: Yeah, Serbo-Croatian.

b. LFA∃f[I heard they hired someone who speaks f(a Balkan language)].

LFE[Serbo-Croatian]λxi.[I heard they hired someone who speaks xi]

4I refer the reader to Thoms (2013) and Park and Kang (2007) for the LF parallelism account of Multi-

According to Thoms, contrastively focused phrases take clause bound scope at LF by undergoing Quantifier Raising. This explains the contrast between (45) and (46). In (46), the contrastively focused correlateBulgariantakes clause bound scope at LF, see LFAin (46b). The corresponding remnant in LFE,Serbo-Croatian, takes ma-

trix scope. Since the traces of the correlateBulgarianand the remnantSerbo-Croa- tianare not bound from identical positions, this derivation is ruled out by LF par- allelism.

(46) a. A: I heard they hired someone who speaks BULGARIANfluently. B:∗No, SERBO-CROATIAN.

b. LFA[I heard they hired someone[Bulgarian]x.[who speaksxi]].

LFE[Serbo-Croatian]i λx.[I heard they hired someone who speaksxi]

To summarize Thoms’ (2013) theory, the idea is that EM is not constrained in the syntax, but its application is subject to the independent requirement of LF paral- lelism. Part of the elegance of the LF parallelism theory is that it makes use of a condition for which there is independent support outside the domain of EM.5Un- fortunately, this theory suffers from several conceptual and empirical problems.

3.2.2 Problems for the LF parallelism account of EM

In this section, I discuss several problems for the hypothesis that EM is constrained by LF parallelism. To begin with, the LF parallelism theory fails to meet all the re- quirements that a theory of EM should meet. That is, it fails to address all the ques- tions in (40), repeated here as (47).

(47) Requirements on a theory of EM:

• Why is EM parasitic on ellipsis? (Ellipsis question) • What locality conditions is EM subject to? (Locality question)

• What triggers EM? (Trigger question)

One question that the LF parallelism theory does answer is the locality question. The answer to this question is that EM is not constrained by any locality condi- tions (cf. the island violation in (45)). One of the problems with the claim that EM is not constrained by syntactic locality conditions, is that it leaves no room for cross- linguistic variation. That is, if EM is unconstrained under ellipsis, it should be so in any language. In section 5, I show that there is, contrary to what the LF parallelism theory predicts, cross-linguistic variation with regard to movement under ellipsis.

Since LF parallelism is a condition on ellipsis, it may appear as if the LF paral- lelism theory also answers the ellipsis question. This is not the case, though. Under the LF parallelism theory, the question remains what constrains EM when no ellip- sis applies. Because LF parallelism does not come into play when no ellipsis takes place, the expectation is that nothing constrains EM when no ellipsis applies. This

is clearly incorrect, as EM only occurs under ellipsis, but in the absence of ellipsis, EM is impossible, see the contrast between (48a) and (48b).

(48) a. * John has travelled to Spain and Bill has[to India]i travelled ti

excep t i onal movement

b. John has travelled to Spain and Bill has[to India]i [VPtravelled ti]

The answer to the trigger question under the LF parallelism theory is that it is not triggered at all. Application of EM is free, as long as LF parallelism is obeyed. Again, this is problematic when non-elliptical structures are considered. If nothing trig- gers EM, why can’t EM be triggered in the absence of ellipsis?

These considerations lead to the conclusion that EM is constrained by some fac- tor and that this factor is absent when ellipsis applies and not the other way around. EM is not more constrained (e.g. by LF parallelism) when ellipsis applies, rather it is less constrained: a movement is possible in elliptical, but not in non-elliptical structures. Likewise, there must be something that triggers EM under ellipsis, and this trigger must be absent when no ellipsis applies (see section 6).

Another theoretical problem for the LF parallelism theory is the following. Un- der the LF parallelism theory, the LF position of the correlate determines the LF position of the remnant. However, it does not follow from this that the LF position of the remnant should be the same as thePF positionof the remnant. Since May (1977), it is well-known that PF spell-out position and LF scope position do not al- ways coincide. To make the problem more concrete, consider the example in (49). Nothing in the LF parallelism theory rules out (49a) with the derivation in (49b). In this derivation, LFAand LFE satisfy LF parallelism: the remnanta bookand its

corresponding correlatewhattake scope over the embedded clause, binding their trace from an identical position. At PF,a bookis spelled out in a left-peripheral po- sition in the matrix clause. In other words,a bookhas moved into the matrix clause in the syntax, but at LF reconstructs to (i.e. is interpreted in) a position where it scopes only over the embedded clause, thereby satisfying LF parallelism. The prob- lem is even more severe if the de dicto reading in (49), in which John and Mary do not have a specific book and pencil in mind, involves narrow scope of the indefinite with respect to the intensional verb (Russell, 1905; Fodor, 1970; Montague, 1973; Partee, 1974; Cresswell and Stechow, 1982; Keshet, 2008, a.o.). In that case, (49b) is in fact the only possible LF/PF representation for (49aB).

(49) a. A: Who believed you bought what? B:∗John a book, and Mary a pencil.

b. LFA[[who]x.[xibelieved[what]j λy.[you bought yj]]

LFE[John]x.[xibelieved[a book]j λy.[you bought yj]]

PFE [[John]i [a book]j[tibelieved you bought tj]]

Next to these conceptual problems, there are empirical problems for the LF par- allelism theory, as well. One problem is that, if contrastively focused correlates take

clause bound scope by Quantifier Raising, (50) should be ungrammatical, since no LF parallelism obtains. The focused correlateGreektakes scope over the embedded clause at LF, but the corresponding remnantAlbaniantakes matrix scope (50b).6

(50) a. A: Did Abby claim she speaks GREEK fluently? B: No, ALBANIAN.

b. LFA[did Abby claim[Greek]i λx.[she speaksxifluently]]

LFE[[Albanian]λxi[Abby claimed she speaksxi fluently]]

A final drawback of the LF parallelism theory, is that it leaves unexplained what I have called the landing site question (i.e. why EM must land next to the ellipsis site), illustrated by the contrast between (51a) and (51b).

(51) a. John has travelled to Spain and Bill has[to India]i [VPtravelled ti ].

b. * John has travelled to Spain and Bill[to India]ihas[VPtravelled ti ].

Under the LF parallelism theory, the landing site of EM is constrained by the LF po- sition of the correlate. The answer to the question why the remnant lands next to the ellipsis site, is thus that it is a coincidence: the correlate happens to be in an identical position in the non-elliptical antecedent. It seems the LF parallelism the- ory overlooks a generalization here. I provide an answer to the landing site question in section 4.2.

To sum up, Thoms’ (2013) theory of EM in terms of LF parallelism suffers from conceptual as well as empirical problems. Moreover, I showed that it fails to meet the requirements on a theory of EM (cf. (47)). In the next section, I explore an alter- native view on EM, retaining the idea that EM involves leftward movement.

4 EM and ordering statements

In this section, I answer the ellipsis question and the landing site question. I will implement my proposal in Fox and Pesetsky’s theory of Spell-out Domains.