Some Problems with a Lowering Account of Scrambling

Bošković and Takahashi (1998) (B&T) propose that scrambled arguments are base-generated in their surface positions (adjoined to IP), but may undergo covert lowering to their canonical positions in order to be assigned a θ-role at LF, in accordance with Last Resort.¹ They argue that since no independent principle of grammar requires it, such movement need not leave a trace, and therefore the Proper Binding Condition does not come into play.

They cite examples, principally from Japanese, showing that the appropriate LF configuration to account for various binding and scope phenomena is that in which the scrambled element appears in its canonical position (and nowhere else). For instance, in (1) (B&T 1998:354) the scrambled dative QP daremo-ni 'everyone' cannot take scope over the nominative QP dareka-ga 'someone' despite its surface position.

(1) Daremo_i-ni dareka-ga [Mary-ga e_i atta to]

everyone-DAT someone-NOM Mary-NOM met that omotteiru.

thinks

= for some x, x a person, x thinks that for every y, y a person, Mary met y

≠ for every y, y a person, there is some x, x a person, such that x thinks that Mary met y

Similarly, in (2) (B&T 1998:355) the anaphor otagai-no 'each other' cannot be bound by the scrambled argument Mary to Pam-ni 'Mary and Pam'.²

(2) *[Mary to Pam]_i-ni [otagai_i-no     hahaoya]-ga

Mary and Pam-DAT each.other-GEN mother-NOM

[John-ga     e_i atta to] omotteiru.

John-NOM     met that think

'Mary and Pam, each other's mothers think that John met.'

However, a number of Korean examples, where the facts differ from those of the corresponding Japanese sentences, raise problems for B&T's account. These examples all suggest that in certain cases the scrambled element must be present in its scrambled or some other noncanonical position at LF.

Consider (3), an approximate Korean counterpart to (1).

(3) [Motun saram]_i-ul nwukwunka-ga [Mary-ka     e_i

every person-ACC someone-NOM     Mary-NOM mannassta-ko] sayngkakhanta.

met-COMP     thinks

= for some x, x a person, x thinks that for every y, y a person, Mary met y

= for every y, y a person, there is some x, x a person, such that x thinks that Mary met y

According to B&T, in (3) the QP motun saram-ul 'every person' must covertly move to the canonical position e in order to receive a θ-role from mannassta 'met'. But then it is difficult to account for the fact that in this example, as opposed to its Japanese counterpart, a wide scope interpretation is indeed available for the scrambled quantifier.³ On the other hand, the wide scope interpretation follows naturally if motun saram-ul is allowed to remain in the IP-adjoined position at LF.

Now consider (4) (Kim 1995:54), which demonstrates that wh- in-situ in an embedded clause takes scope only over that clause.

(4) Ne-nun [Mary-ka   nwukwu-lul salanghanun-ci]

you-TOP Mary-NOM who-ACC love-Q

al-ko       sip-ni?

know-COMP want-Q

1. a. 'Do you want to know who Mary loves?'

2. b. *'Who do you want to know whether Mary loves?'

Given B&T's analysis, we would expect that the scrambled version of(4), with the wh-phrase in front, should have exactly the same possible interpretations, since the two would be indistinguishable at LF. But as (5) (Kang and Müller 1996:274) shows, this expectation is not borne out. Rather, the scrambled but not the in-situ wh-phrase can take wide scope.⁴

(5) Nwukwu_i-lul Shin-Sook-un [Suna-ka   e_i

who-ACC       Shin-Sook-TOP Suna-NOM

sungbaehanun-ci] a-ni?

admire-Q         know-Q

1. a. 'Does Shin-Sook know who Suna admires?'

2. b. 'Who does Shin-Sook know whether Suna admires?'

The preceding examples demonstrate that, in Korean at least, there are cases where scrambling can provide interpretive options not available to the unscrambled variant, and further that these options...