Local Dependencies in Comparative Deletion

This paper investigates the principles that allow an adjectival constituent to be "missing" in the surface form of comparative constructions like , a phenomenon that I will refer to descriptively as comparative deletion (CD). Traditionally, examples like (2) have been analyzed in terms of Logical Forms like (2), in which an operator in the comparative clause (the c-clause) binds a degree variable inside an adjectival constituent that can be "deleted" in the surface form in accord with general principles of ellipsis (see e.g. Bresnan 1973, Chomsky 1977, Napoli 1983, Izvorski 1995).

1. Jupiter is more massive than Saturn is.
2. Jupiter is more massive than [Opx Saturn is [ex massive]]

An empirical problem for this analysis, which has not been previously discussed, is that CD differs from other ellipsis operations such as VP deletion in an important and unexpected way. Whereas the interpretation of a null VP may be recovered from any semantically appropriate VP within recent discourse, the interpretation of the missing adjectival constituent in CD must be recovered locally. This fact is illustrated by (3) and (5) below. The second conjunct of (3) is ambiguous between the read-ing in (4)a, in which the elided VP receives its interpretation from the VP in the first conjunct and the reading in (4)b, in which the elided VP receives its interpretation from the VP in the second conjunct.

3. Marcus read every book I did, and I bought every book Charles did.
4a. ...and I bought every book Charles read
4b. ...and I bought every book Charles bought

The parallel CD example in (5), however, is not ambiguous: this sentence has only the interpretation in (6)b, not the one in (6)a (although this would be a perfectly acceptable sentence).

5. The table is wider than the rug is, but it's not longer than the rug is.
6a. *...but it's not longer than the rug is wide
6b. ...but it's not longer than the rug is long

A more theoretical problem for an ellipsis approach to CD arises when we adopt the independently motivated analysis of the adjectival projection defended in Abney 1987 and Corver 1990, 1997, in which adjectives project extended functional structure headed by degree morphology (a "Degree Phrase" or "DegP"). Under this analysis, the expression that provides the source for the missing material in a comparative like must be a constituent of the form shown in (7) (where the c-clause has been raised to eliminate antecedent containment; cf. May's (1985) discussion of antecedent-contained VP deletion). The result is that the LF of the c-clause in (1), on an ellipsis analysis, should be (8), not (2).

7. [DegP more [AP massive] than e]
8. [Opx Saturn is [more massive than ex]]

Not only does (8) not accurately represent the meaning of (1), if the c-clause is interpreted as a definite description of a degree (see e.g. von Stechow 1984), then actually predicts that should be uninterpretable, since there is no unique degree d such that Saturn is more massive than d.

The goal of this paper is to show that the solution to the problem for the DegP analysis also provides an explanation for the local dependency observed in examples like (5). Specifically, I argue that CD constructions like (1) do not involve ellipsis at all; instead, the operator in the c-clause directly binds a variable corresponding to a Degree Phrase, not a degree variable inside DegP. That is, the LF of the c-clause in (1) is neither as in (2) nor (8), but rather as in (9). Crucially, since no ellipsis is involved in the derivation, the problem presented by a DegP analysis of the adjectival projection disappears.

9. Jupiter is more massive than [Opx Saturn is ex]

The question that remains is how the c-clause gets interpreted as a definite description of a degree. First, I show that if the operator in e.g. (9) binds a DegP, then the basic interpretation of the c-clause should actually a function from adjective meanings to degrees. Then, building on proposals in Klein 1980, Larson 1988, and Lerner & Pinkal 1995, I provide a compositional semantics for an Abney-style syntactic analysis of DegP in which the adjective that heads the comparative provides the argument to the c-clause, thus establishing the local dependency between the "missing" adjective meaning in the c clause and the adjective that heads the comparative. I conclude by discussing the implications of the proposal for the broader analysis comparatives in English, and for the analysis of comparatives in languages like Russian and Spanish that do not display the same range of ellipsis phenomena as English, yet do have comparative deletion constructions.