Computational Linguistics Colloquium

Thursday, 4 July, 16:15, Seminar Room, Building 17

Adjunct valents, cumulative scopings and impossible descriptions

Bob Levine
Department of Linguistics
The Ohio State University

Recent traceless accounts of extraction, such as Bouma, Malouf and Sag's important recent paper in NLLT v.19, `Satisfying constraints on extraction and adjunction' (BMS), rely on a mechanism roughly characterizable as valence reduction as the principle means for terminating extraction pathways. In place of empty categories occupying positions in phrase structure, this approach assumes that gaps correspond to cases where a lexical head has failed to combine with a category it normally selects. The connectivity mechanism which links fillers to their gap sites via SLASH propagation is, on this approach, revised to mark relevant heads so that their valence is reduced by exactly the filler element(s) to which they are linked; thus, in a sentence such as `On which table did you put the book', the sign corresponding to `put' which appears in the HPSG representation of this sentence simply does not take a PP valent, and hence the latter never appears as a sister to the head, though a description of suppressed PP complement does appear on the verb's ARG(ument)-ST(ructure) and DEP(endent)S list. But on such an account, the extraction of adjuncts is immediately ruled out on the assumption long held in syntactic theory that adjuncts are unselected sisters of phrasal projections.

BMS argue that this assumption has no strong motivation. They defend the position that it is not only conceptually preferable (assuming the correctness of their overall approach to extraction) but empirically motivated to take adjuncts to be selected elements appearing on COMPS lists, and to be suppressed as valents in cases of adjunct extraction by exactly the same interaction of relational constraints as extracted complements. They offer, as independent corroboration for this line of analysis, evidence from a number of quarters which strongly suggest that adjuncts are indeed dependents of heads, and emphasize in particular certain facts about the possibility of adjuncts scoping within complex lexical items in Japanese and Dutch to support their treatment of adjuncts as valents of lexical heads.

In this presentation I show that a rather common kind of interaction between adjunct modification and coordinate structure constitutes a severe contraindication to the BMS treatment of adjuncts, hence of adjunct extraction and of extraction in general. The data in question are sentences displaying cumulative scoping over coordinations, such as

Robin came in, grabbed a chair, sat down and whipped off her logging boots in fifteen seconds flat.
In how many seconds flat did Robin find a chair, sit down and whip off her logging boots?

I show that under the BMS analysis, there is no way these sentences can be licensed under the normal (and for many speaker the exclusive) reading in which the adjunct applies to the whole complex event rather than to any particular conjunct. The requirement that adjuncts be selected by lexical heads, in combination with the fact that each adjuncts so selected refers in its own description to specific properties of the head that selects, turns out to make it impossible to provide a consistent description of the adjuncts---which I show must be treated as fillers in both (1) and (2)---which satisfies the connectivity properties BMS assume as part of the filler/gap relation. Treatment of adjuncts as unselected adjoined sisters of phrasal heads, on the other hand, immediately yields an empirically satisfactory, almost trivial account of case such as (1) and (2), and so emerges as clearly preferable to the BMS approach. If time permits, I will also sketch a novel encoding of CONTENT specifications in which psoa-types include both lambda abstraction and structured meanings, whereby the treatment of adjuncts as actually adjoined yields a completely satisfactory, elegant account of the Japanese and Dutch adjunct scoping facts that BMS believe gives their postion crucial support.

If you would like to meet with the speaker, please contact Frederik Fouvry.