Bošković (2005, 2008) outlines a phase-based analysis of adjunct extraction in determinerless languages like Serbo-Croatian (SC). The analysis is modeled on recent phase-based analyses of P(reposition)-stranding (see Abels 2003, 2012) wherein the richness of the functional structure of the PP determines whether extraction is possible. This squib identifies a problem for a unified analysis of these two phenomena. Bans on adjunct extraction are obviated under sluicing while bans on P-stranding are not.
1 The Paradigm in Question
First, Bošković (2008) (as well as Uriagereka (1988) and Corver (1992)) notes a distinction between languages with regard to adjunct extraction from nominals. Languages that lack overt determiners, like SC (1), allow extraction of nominal modifiers. Languages with determiners, like English (2) and Spanish (3), do not.
[Iz kojeg grada]i je Petar sreo [djevojke ti]?
from which city is Petar met girls
‘From which city did Petar meet girls?’
(2) *[From which city]i did Peter meet [girls ti]?
*[¿Con qué]i comiste [una hamburguesa ti]?
with what ate.2nd a hamburger
‘What [topping] did you eat a hamburger with?’
Bošković develops a phase-based, derivational account for this distinction, as follows. English is a language with DPs; and assuming that DPs are phases, no movement can cross them without first landing in the specifier position Spec,DP. However, the PP in (2) is adjoined to its host NP and cannot move to that specifier position because of conditions that block derivational steps that are too ‘‘small.’’ These antilocality restrictions require that a moving element minimally cross an entire phrasal boundary, segments being insufficient (Ishii 1997, 1999, Abels 2003, Grohmann 2003). As illustrated in (4), there is thus no licit derivational avenue for extracting the adjunct.
This problem does not arise for SC, which, according to Bošković, lacks a DP layer. As NPs contrast with DPs in not being phases,1 there is no need for the adjunct to first move to the NP edge before exiting the nominal, and so antilocality can be honored. Consequently, the adjunct can move unimpeded out of the NP.
(5) Whoi did you talk to ti?
*Wemi hast du mit ti gesprochen?
who have you with spoken
‘Who have you spoken with?’
*¿Quiéni contaron todos con ti?
who counted everyone with
‘Who did everyone count on?’
(from Campos 1991:741)
The same logic applies to P-stranding. As illustrated in (8), the derivational constraints that work together to rule out (2) also rule out (6). In German, the preposition is a phase head that a moving element cannot cross without first landing in its specifier position. However, the complement of this phase-head preposition will not cross a full phrasal boundary in moving to the specifier position and is thus blocked.
In English, preposition heads are not phasal, and movement can proceed across them freely (Abels 2003).
2 The Problem: Ellipsis Facts
The phase-based analysis of P-stranding is understood derivationally so as to accommodate an important fact: ellipsis does not improve P-stranding violations.2 The tight parallelism between this proposal and Bošković’s analysis of extraction from nominals would suggest a parallel interpretation of his proposal. This predicts that subsequent ellipsis should leave the nominal extraction data unaffected. For example, just as German and Spanish do not allow extraction of the object of a preposition even if followed by sluicing, so English and Spanish should not allow extraction of nominal modifiers even if followed by sluicing. However, as (9)/(10) and (11)/(12) indicate, this prediction is incorrect.3
Johannes hat mit jemandem gesprochen, aber ich weiss nicht *(mit) wem.
Johannes has with someone spoken but I know not with who
‘Johannes spoke with someone, but I don’t know who.’
¿Con qué chica ha salido Juan? *¿(Con) Elena?
with what girl has gone.out Juan with Elena
‘What girl did Juan go out with? Elena?’
(11) Ivan wanted to eat a cheeseburger with a particular topping. I just can’t remember with which topping.
Ayer comí una hamburguesa con algo pero no me acuerdo con que.
yesterday ate.1st a hamburger with something but no CL recall with what
‘Yesterday I ate a hamburger with something, but I don’t remember with what.’
(13) *[With which topping] did Ivan want to eat a cheeseburger?
Both Spanish and English are DP languages in Bošković’s sense and are correctly predicted to forbid extraction from nominals. However, as these ellipsis data indicate, the effects of ellipsis on acceptability do not correlate with its effects on P-stranding. Bans on P-stranding hold under ellipsis, whereas bans on adjunct extraction from nominals do not hold under sluicing.
3 Short Sources
Some explanations of island effect amelioration under sluicing have posited that there is sometimes no island in the elided material (see discussion in, e.g., Merchant 2001, Van Craenenbroeck 2010). For example, English disallows left-branch extraction (14a) generally, but allows it under sluicing (14b). An explanation of this could be that there is a clefting ‘‘short source’’ of (14b) that does not actually involve leftbranch extraction (14c).
*How large did Ivy sell a car?
Ivy sold a large car, but I can’t remember how large.
Ivy sold a large car, but I can’t remember [how large]i
the car was ti
It is logically possible that the apparent island obviation in (11) actually stems from a short source and that the derivational ban on adjunct extraction from nominals can still hold. However, the copular short source of (11) is unacceptable.
(15) *With which topping was the cheeseburger (to be)?
The with-phrase above is used in a modifying sense and cannot be used as a predicate, ruling out the cleft source. Further, Merchant and Van Craenenbroeck note that else-modification cannot cooccur with clefting, but can with sluicing. As a reviewer notes, this holds with the relevant example (16), further militating against the cleft short source.
(16) Joe wanted to buy a burger with ketchup but I don’t know with what else (*it was).
Thus, the with-phrase cannot be base-generated as a copular predicate. Nor can it be base-generated as a VP-adjunct. This would only permit a comitative reading (‘to eat with Dana’), an instrumental reading (‘to eat with a fork’), or a manner reading (‘to eat with relish’). In sum, the particular modifier reading of the phase can only arise via NP-adjunction and thus effect an island violation in lieu of sluicing. Given the lack of a short source for this case of island amelioration, example (11) may be a unique case with the palliative effect of sluicing necessarily attributed to the ellipsis itself.
The differential effects of ellipsis on derivations involving P-stranding and adjunct extraction from nominals (even within a single language, as in Spanish) suggest that different conditions regulate the two processes. A unified phase-based account of the kind proposed by Bošković (2005, 2008) would thus seem to be inapposite.
1 Bošković (2005, 2008) gives two different yet related theories about the phasehood of nominals, and we have concentrated on the simpler one. In other work (Bošković 2010), he argues that some NPs are phases in languages without determiners. Adjunct extraction is allowed here because the adjunct is generated on the edge of a phasal NP. The problem presented in this squib holds for both theories.
2 There is disagreement in the literature over how much P-stranding violations are or are not ameliorated by ellipsis. If it is the case that they are often salvageable (as suggested in Nykiel and Sag 2009), then there is an apparent problem for the derivational prohibition against P-stranding. We assume that the prohibition holds under sluicing, but our argument does not crucially rely on this.
3 It should be noted that Rodrigues, Nevins, and Vicente (2009) show that under most circumstances P-stranding under sluicing in Spanish (and Brazilian Portuguese) is superficially possible. This is because the sluice has an underlying clefting source in which there is no P-stranding. The example in (10) is an instance where the clefting source is ruled out (though see Almeida and Yoshida 2007 for arguments against a clefting source). A reviewer notes that the fact of a licit clefting source might be correlated with the fact that DPextraction under sluicing is generally free.