Query: coordination in Categorial Grammar

[John Beavers]

Hi, I've not read any CCG literature that assumes anything more than
binary coordination, but Steedman's general approach has always been to
assume a syntactic meta-rule schematizing over binary coordination of
categories with different valences, so I imagine a natural extension of
this approach would be to further schematize over an arbitrary number of
conjuncts as well:

COORDINATION (<PHI_n>)

X_1 : f_1 ... X_i : f_i CONJ : b X_(i+1) : f_(i+1) ==>
    X : lambda...b(f_1...)...(f_(i+1)...)

Where the "n" schematizes over different valences of X (thus affecting the
lambda expression in the semantics mostly) for i+1 conjuncts (cf. Steedman
1996, 2000).  Pretty hideous looking but that seems like a natural
extension of the syntactic approach.

The alternative for binary coordination is assume "and" has a category
kinda like (X\X)/X : and', so a natural extension there would be to
schematize over left conjuncts like Roger suggested, maybe ((X\$)\X)/X :
and', where $=X, using the $-convention (i.e. X\$ is the set of categories
including X and X\X and X\X\X etc.).

Anyway, that's just guesswork, I haven't read the literature on non-binary
coordination (if there is any).  But I don't think you can do it without
some kind of schematization of this sort or else some form of
underspecified argument structure.

John

On Thu, 10 Oct 2002, Ash Asudeh wrote:

>
> Hello All,
>
> Yes, I know this isn't a Categorial Grammar list, but I think several
> people here are interested enough and know enough to answer the following
> question.
>
> How does CG (either Combinatory CG or Type-Logical Grammar) deal with
> non-binary coordination like "X, Y and Z"?
>
> >From reading up on coordination in Steedman's work and Carpenter's book,
> it seems that the way it is done is to first combine Y and Z and then
> combine this with X (or start with X, this isn't important).
>
> Doesn't this assume that there has to be a coordinator in "X, Y and z"
> other than the overt "and" (e.g., either a null coordiantor, or something
> corresponding to the comma)?
>
> Thanks,
> Ash
>
>