When Minimalism Isn't Enough: An Argument for Argument Structure

1 Introduction

A central hypothesis in Chomsky's (1993, 1995) Minimalist Program is that only two interface levels of representation are available on which to state linguistic generalizations: PF and LF. This conceptually appealing simplification of previous versions of transformational grammar (e.g., Chomsky 1981) claims that D-Structure and S-Structure are not needed and thus should not be part of the syntactic architecture. Chomsky focuses on LF, showing not only that constraints thought to hold of D-Structure and/or S-Structure can be stated on LF but also that doing so provides the best analysis of the data. In this squib I will examine the one empirical argument of this type that Chomsky makes, arguing that although conceptually appealing it fails on empirical grounds.

Chomsky's argument involves an analysis of the interpretations of (1) and (2). Both sentences are ambiguous: himself can refer to either Bill or John. What is interesting, however, is that on the idiomatic interpretation of (2), where take picture means 'photograph', himself can only refer to Bill.

(1) John wondered [which picture of himself] Bill saw.

(2) John wondered [which picture of himself] Bill took.

Chomsky's claim is that whatever level of representation idiom interpretation takes place on is the level at which reflexive binding is determined. Specifically, he claims that at LF the idiom take picture must form a unit for interpretation and that LF is therefore the correct level for the determination of reflexive binding. If reflexive binding were free to apply at other levels of representation (e.g., D-Structure and/or S-Structure), the correlation between binding and interpretation in (2) would not be explained. This suggests strongly that LF, not D-Structure or S-Structure, is the only level for determining binding relations, supporting the basic Minimalist Program architecture.

In this squib I will argue that it is correct that the level at which take picture is a unit is the level at which reflexive binding is determined. However, I will argue that the relevant level is not LF, but a level of Argument Structure. My analysis is couched in terms of the sort of Argument Structure proposed in recent work in Head-Driven Phrase Structure Grammar (HPSG) but it is hoped that the argument I make here is general enough to extend to other approaches as well. The argument has two parts. First, I show that the idiom take picture cannot always be a unit at LF; that is, LF is not the right level of representation to characterize the interpretation of the idiom. Second, I show that the sentences in (1) and (2) are wrongly grouped together: (1) involves a binding theory-"exempt" anaphor (Pollard and Sag 1992, 1994), also called a logophor (Sells 1987, Reinhart and Reuland 1993), sensitive to pragmatic conditions, which accounts for the ambiguous binding; the idiomatic interpretation of (2), however, involves a true anaphor, sensitive to syntactic Argument Structure, which accounts for its obligatory binding by its subject Bill. Rather than providing an empirical argument in favor of the Minimalist Program, the contrast in (1) and (2) constitutes an empirical argument against its basic architecture, suggesting it must be supplemented with at least a syntactic Argument Structure representation.

2 The Idiom at LF

Chomsky's claim is that at LF the idiom take picture must be a unit for interpretation. In this section I will argue that take and picture can be separated at LF even while maintaining the idiomatic unit interpretation, casting doubt that LF is the level where idiom interpretation takes place.

Let us suppose that the relevant unit is some sort of VP or V' containing the verb take and the direct object headed by picture.

(3) [_VP take [_DP picture]]

The following examples are based on observations by Diesing (1992), who argues that a DP's position at LF correlates with certain semantic properties. If the structure in (3) is correct, then the idiom's object should have semantic properties associated with being VP-internal only. However, the idiomatic object can pass every test for being...