- Remarks on Beck's Effects:Linearity in Syntax
Investigating similar sets of data, exemplified in (1) and (2), Beck and Kim (1997; hereafter B&K) and Tanaka (1997; hereafter T) arrive at different conclusions.1 (1a), which has a subject negative polarity item (NPI) and an object wh-phrase in situ, is ungrammatical, but its scrambled counterpart (1b) is grammatical. The same contrast exists in Korean, as shown in (2). The judgments on the Korean examples in (2) are B&K's. [End Page 314]
a. ?*Dare-mo nani-o kawa-nakatta-no?
anybody what-ACC buy-NEG.PAST-Q
b. Nani-oi dare-mo ti kawa-nakatta-no?
what-ACCi anybody ti buy-NEG.PAST-Q
'Q nobody buys what?'
a. *Amuto mues-ul sachi-anh-ass-ni?
anybody what-ACC buy-NEG-PAST-Q
b. Mues-uli amuto ti sachi-anh-ass-ni?
what-ACCi anybody ti buy-NEG-PAST-Q
'Q nobody bought what?'
After summarizing the two accounts of the contrast, I demonstrate that T's analysis is empirically superior to B&K's. I further show that T's S-Structure condition can be reduced to a condition on semantic representation Sem, if it is assumed, contra Saito (1992), that scrambling cannot be undone. This conforms to the elimination of traditional levels of representation in Chomsky 2001a.
1 Beck and Kim's (1997) Negation-Induced Barrier
Since B&K assume that local scrambling cannot be undone at LF, their analysis can capture the contrast in question at LF. B&K's analysis consists of the two statements in (3).
a. Negation-induced barrier (NIB)
The first node that dominates a negative quantifier, its restriction, and its nuclear scope is an NIB.
b. Minimal Negative Structure Constraint (MNSC)
If an LF trace β is dominated by an NIB α, then the binder of β must also be dominated by α.
(4) is a schematic representation of (1a) and (2a). The NIB is italicized. (4) assumes that the wh-phrase undergoes movement at LF. Here and in (5), the trace created by LF movement is superscripted with LF.
(4) *[NIB anybody [tiLF buy]-NEG]-PAST-Q-[what-ACC]i
The trace of the wh-phrase, but not the wh-phrase itself, is dominated by the NIB in violation of the MNSC. For B&K, (1b) and (2b) have the structure in (5) at LF.
(5) [tiLF [NIB anybody ti buy-NEG]-PAST]-Q-[what-ACC]i
Given the added assumption that the NIB is not a barrier in overt syntax, ti in (5) is free from the MNSC. Since tiLF is not dominated by the NIB, the MNSC is trivially satisfied in (5). [End Page 315]
2 Tanaka's (1997) Linear Crossing Constraint
T's analysis of the contrast in (1)-(2) appeals to the Linear Crossing Constraint (LCC) on Ā-dependencies: nesting Ā-dependencies (6a) are permissible, but crossing Ā-dependencies (6b) are not.
Assuming that NPIs and wh-phrases in situ undergo invisible operator movement (see Watanabe 1992) to Spec,NegP and Spec,CP, respectively, (1a-b) have the structures in (7a-b), respectively.
While (7a) is ruled out by the LCC, (7b) is correctly predicted to be grammatical.
3 Scrambled Quantificational Expressions
Other things being equal, B&K's analysis is more attractive than T's, since B&K do not appeal to linearity. Perhaps for this reason, B&K's analysis is adopted in modified forms in the recent literature (e.g., Hagstrom 1998, Pesetsky 2000),2 and the contrasts in (1)-(2) (and some others reported in Beck 1996) have come to be known as Beck's (intervention) effects.3 However, to the extent that the two analyses make different predictions, the choice between them must be made on empirical grounds.4 In the remainder of this squib, I show that T's approach is empirically superior to B&K's and hence should be maintained. I will use Japanese data, but the same point can be made using Korean data (for which, see Sohn 1994). Let us consider (8...