What Ellipsis tells us about (Direct) Compositionality - and Vice Versa

Pauline Jacobson, Brown University
Friday, October 19, 11am

The hypothesis of direct compositionality is that the syntax is a system of rules "building" expressions (that is, proving them well-formed), while the semantics works in tandem to supply a model-theoretic interpretation to each expression as it is "built" in the syntax. There are two interesting consequences to this: it makes no use of an intermediate level of representation (e.g., "Logical Form") mediating between the syntax and the interpretation, and - when coupled with a reasonably strong view of the syntax - entails that the grammar makes no use of statements that consult representations. (Thus trees are a representation for convenience for the linguist - and represent how the syntax proved something well formed and (rather roughly) represent how the compositional semantics interpreted the expression, but they play no role in the statement of any grammatical processes.)

Ellipsis phenomena provide a number of apparent challenges to this view. In this talk, I will examine these challenges and attempt to answer several of these: arguing not only that the relevant phenomena can be handled under direct compositionality but that in fact the direct compositional approach does better. In particular I will focus on one case concerning "Antecedent Contained Ellipsis" and a group of facts which have been taken to necessitate quite complex conditions on Logical Forms: I argue that if we think (direct) compositionally and take the meanings seriously, these facts fall into place. Finally, I will use ellipsis as a way to comment on the relationship between compositionality within the grammar, and facts about processing. There are two key points to keep in mind here. First, we know from much of the psycholinguistics literature that listeners compute meanings on-line as incoming material is heard. It is difficult to imagine any theory of processing which can capture this if the grammar itself is notÊ rooted in a direct compositional architecture. On the other hand, the interpretation of ellipsis (and also of related facts concerning contrastive stress) looks non-compositional in that we are "supplying" bits of meaning (both for ellipsis and/or for the purposes of interpreting contrastive stress) for subexpressions. But once we make a distinction between listeners' memories and the formal system (which has no memory, and cannot "look down" or "go back" in a strictly compositional approach) we can make sense of this apparent violation of compositionality.