Sag&Wasow coordination question

Tom Wasow wasow at csli.Stanford.EDU
Wed Oct 27 15:34:44 UTC 1999


As Stefan Mueller points out, NP coordination requires a separate rule.
Our book has a problem (Chapter 5, Problem 1) in which students are
supposed to discover this.

Regarding the issue of the two interpretations you offer, only the first
one is consistent with the way we use the tag notation.

Tom

On Wed, 27 Oct 1999, Lars G. Johnsen wrote:

> Hi all
>
> I am using Sag&Wasow's book in a syntax course and have a question
> concerning the coordination schema and its relation to unification. The
> schema, in chapter 4, is formulated as:
>
> [1] -> [1]+ CONJ [1]
>
> There seems to be a problem with unification using this schema, or at
> least it has caused me to scratch my head, but then i may have missed a
> paragraph in the book. Here are two interpretations:
>
> (1 - bottom up) The tag [1] means that the constituents in the
> coordination must be unified with each other.
> ================
>
> Problem: Given that constituents typically have a HEAD feature, the
> schema (as it stands) will block any coordination of phrases with
> conflicting agreement features (assuming these in the HEAD), something
> we dont want in e.g. Norwegian, where NPs have gender agreement (a)
> below, or in English with plurals (b):
>
>  a) NP[+neut] CONJ NP[+masc].
>  b) a man and two dogs entered
>
>
> (2 - top down) The schema is interpreted as having instances like
> follows:
> ================
>
> [phrase HEAD noun] -> [phrase HEAD noun]+ CONJ [phrase HEAD noun]
>
> These underspecified structures are then unified in turn with
> constituents that make up the coordination, but constituents are not
> unified with each other, so that examples (a-b) will be well formed
> coordinations. This is almost the same as saying that coordination
> extracts a common denominator from its constituents.
>
> Problem: Under this interpretation, nothing prevents us (or so it seems)
> from instantiating the schema as:
>
> phrase -> phrase+ CONJ phrase
>
> or
>
> word -> word+ CONJ word
>
> thus allowing *any* sequence of phrases or words to be coordinated,
> which again is something we don't want.
>
> Comments anyone?
>
> Best, Lars G. Johnsen
>



More information about the HPSG-L mailing list