semantics

[Miriam Butt]

I'm sorry if I sounded like I was saying that I believed constituent
structure determined semantic projections. That is certainly not what
I meant. I believe I said (or meant to have said), that as far as I
could tell in the work I've seen (which excludes that of van Benthem
1987) that the semantic (sigma) structure is strongly dependent on
syntactic representations. By this I meant primarily f-structure,
though the need for an occasional reference to c-structure has been
argued for in works like that of Andrews and Manning (1993).

In fact, Joan's clarifying remark 

> The claim is that the mapping from f-structure to
> compositional semantic structure (sometimes called the "sigma
> projection") can be as simple as the mapping from syntactic structure
> to LF.

states exactly where I was coming from.  That is, if you are doing
compositional semantics, you have to have some pieces that you are
composing. These pieces tend to be lexical pieces that have been
arranged according to some syntactic analysis, theory-particular
persuasions aside.  

I am also very interested in the relationship between lexical and
"big" compositional semantics and I think a lot remains to be
figured out in this area.  I also agree with Joan that Dalrymple et al.
would want to keep lexical semantics separate from the compositional semantic
structures.  However, there is a difficulty in the approach described
there in that they assume a theory of linking, which presumably
involves some lexical semantics, but are not very clear on exactly how
the linking bit fits into the overall architecture.  That is, if you
look at how the mapping principles are actually implemented within
semantic composition via linear logic, it turns out that basically the
following schema is followed (Mary -- correct me if I'm getting it
wrong!). 

1) Have f-structure
2) Use mapping principles to get thematic roles (or some equivalent
   thereof) 
3) Look for a logical deduction of the meaning of the sentence based
   on the information available.  

Now, while this characterization is a bit more derivational than is
actually fair, my problem with this approach has been that I always
envisioned the mapping principles as constraining the possible
f-structures that could correspond to a given argument array. In
contrast, here the mapping rules are being used to access the
predicate-argument structure via the f-structure to then do semantic
interpretation.  But, perhaps my confusion as to the role of the
mapping principles in the Dalrymple et. al approach just stems from a
particularly narrow view that I have of linking.  


Miriam