(Anti)reconstruction Effects in Free Relatives: A New Argument against the Comp Account

1 Introduction

Free relative clauses, exemplified in (1a-b), have been analyzed in two very different ways.

(1)

1. a. Hansel reads whatever Gretel recommends.

2. b. Hansel reads whatever/whichever books Gretel recommends.

On one view, referred to in the literature as the comp(lementizer) account, exemplified in (2), the wh-phrase is in Spec, CP, and the head of the free relative is occupied by an empty pronominal element (Groos and Van Riemsdijk 1981, Grosu and Landman 1998, among many others). On the opposing view, the head account, exemplified in (3), the wh-phrase is in the head position, and the Spec, CP is either non-projected or empty (Bresnan and Grimshaw 1978, Larson 1987, 1998, Bury and Neeleman 1999, Iatridou, Anagnostopoulou, and Izvorski 2001).

(2) Comp account

[_DPpro [_CPwhatever books [_TP Gretel recommends --]]]

(3) Head account

[_DPwhatever books [_CP/TP Gretel recommends --]]

Arguments adduced in favor of one structure over the other come from the distribution and form of free relatives, the presence of so-called matching effects (Bresnan and Grimshaw 1978, Groos and Van Riems-dijk 1981, Citko 2000), and crosslinguistic evidence involving the behavior of free relatives with respect to extraposition phenomena (Groos and Van Riemsdijk 1981). This squib examines yet another set of facts bearing on the choice between the two accounts, namely, the behavior of free relatives with respect to Principle C effects.

2 Reconstruction on the Comp Account

The comp account and the head account make differing predictions regarding the grammaticality of free relatives involving the configuration given in (4), in which the wh-phrase contains the name coindexed with the pronoun c-commanding the relative-clause-internal gap.

(4) [_{free relative}[_wh-phrase name_i]_j . . . [_TP pronoun_i . . . e_j]]

On the comp account, such free relatives are predicted to pattern with wh-questions, since the wh-phrase occupies the same position in both.

(5) Wh-questions

[_CP[_wh-phrase name_i]_j [_TP pronoun_i . . . t_j]]

(6) Free relatives

[_DPpro [_CP[_wh-phrase name_i]_j [_TP pronoun_i . . . t_j]]]

Wh-questions involving the configuration in (5) constitute a familiar case of a Principle C violation.

(7)

1. a. *Which pictures of Greteli does shei display prominently?

2. b. *?Which report on Hansel'si division will hei not like?

Their ungrammaticality is attributed to the fact that at LF the restriction of the wh-phrase is interpreted in a reconstructed position, in which it is c-commanded by the coindexed pronoun.

(8) *[_CP[_wh-phrase … ]_j [_TP pronoun_i … name_i]]

Interestingly, free relatives involving an analogous configuration, schematized in (6), are grammatical on the coindexed reading.

(9)

1. a. We will comment on whichever pictures of Hanseli hei displays prominently.

2. b. We will ignore whichever reports on Hansel's_i division he_i won't like.

This contrast between free relatives and wh-questions is a puzzle for the comp account. One way to maintain the comp account would be to assume that reconstruction is obligatory in wh-questions but optional in free relatives. However, since the wh-pronoun occupies the same position in both, it is not clear why it should undergo obligatory reconstruction only in the case of wh-questions.

3 Reconstruction on the Head Account

A more promising alternative is to derive the contrast between free relatives and wh-questions from a difference in the structural position of the wh-phrase. And this is precisely the option that the alternative view of free relatives, the head account, allows for. On the head account, the wh-phrase in a free relative occupies the same position as the nominal head in a headed relative.

(10)Headed relatives

[_DP[_head name_i]_j     [_CP that [_TP pronoun_i … t_j]]]

(11) Free relatives

[_DP[_wh-phrase name_i]_j [_CP [_TP pronoun_i … tj]]]

This parallelism predicts that free relatives should pattern with headed relatives rather than wh-questions with respect to reconstruction effects. This prediction is indeed borne out. Munn (1994), Sauerland (1998), and Safir (1999) show that headed relatives do not exhibit Principle...