semantics

[Alex Alsina]

Good.  I'm glad someone else thinks the topic of semantics is worth
discussing on this list.  Miriam, in her message, expresses some
disagreements with my claim that GB and LFG take a different position
with respect to the way semantics interfaces with syntax (or at least
that GB takes a very specific view about this, which is different from
the approach that seems natural for LFG).  To repeat myself, these two
views are:

> the "Reduced Autonomy of Semantics" or RAS, in GB, and the "Full
> Autonomy of Semantics" of FAS, typical of LFG.

Miriam confirms what I suspected, that most computational work within
LFG, while possibly paying some lip service to the FAS, in practice
adopts the RAS.  In Miriam's words:

> the FAS
> might be there in theory, but is definitely not reflected in practice.
> The s(emantic)-structure is very clearly a derivative of the syntactic
> (f-structure).  S-structure is projected "off of" f-structure.  

So, LFG semanticists are closet RASists.  I'm not saying that being a
FASist is better, but let's not say one thing and do another.  If
Miriam is right about this (and it seems so to me), we should ask
ourselves what it is that prevents work within LFG from treating
semantics as a fully autonomous level of structure that interfaces
with other levels of grammar by means of correspondence principles.
It seems to me that nothing in the conceptual foundation of LFG should
prevent this.  If anything prevents treating semantics as fully
autonomous, it must be something about the technical apparatus
currently used in most computational work in LFG for linking the
various levels of structure.  Maybe the formalisms used for this
purpose do not allow things to come out the way they are supposed to
in a framework that claims that different types of grammatical
information are represented as formally independent levels of
structure that constrain each other through principles of
correspondence.  Hmm, something to look into if anyone feels so
inclined. 

The other point that Miriam raises is about the status of LF in GB.
Contrary to the view that I expressed, that LF is the linguistic
representation of meaning (or the representation of linguistic
meaning), that is, semantics as far as the language module is
concerned, Miriam claims that LF must be considered a syntactic level
of representation.  In order to settle this point, we should
distinguish the form and the function of a level of representation.
LF is FORMALLY represented as a syntactic structure, in
the sense that it uses the same symbols and formalisms as the truly
syntactic S-structure: it uses the tree structure to represent
part-whole relations and syntactic categories (N, NP, etc.) to label
the nodes in the tree.  There is no doubt about that, but does this
make it a syntactic level of structure FUNCTIONALLY, in the sense that
it expresses some syntactic information not expressed elsewhere (e.g.,
at S-structure)?  I think the answer is clearly 'no'.  The information
that LF uniquely represents is not syntactic.  To the extent that LF
is different from S-structure, it is expressing nonsyntactic
information, information that is only relevant for the semantic
interpretation.  What aspect of the syntax of the sentence "John saw
everyone" does the LF representation '[ [everyone] John saw e ]' try
to capture?  Clearly none.  When Steve Wechsler pointed out in this
discussion that an ambiguous string such as "some girl kicked every
boy" is assigned different syntactic (LF) representations in GB to
capture the ambiguity, he noted that there is no convincing syntactic
motivation for these alternative syntactic representations.  If we
take the position that LF is a semantic level of representation, then
it is natural that there should be no syntactic motivation for LF
representations and none should be sought, because these
representations are not syntactic in function, although they are
syntactic in form. 

Hey, what I'm saying does not disagree with what illustrious GB
semanticists say about this.  I used to be under the impression that
LF was officially supposed to be syntactic (although its function is
clearly to express semantic information), but Mohanan was insistent
about its official semantic status.  So, I decided to investigate the
matter in order to settle it and hopefully prove him wrong.  The end
result is that (although there is some ambivalence and obscurity of
expression concerning this point) there are some clear statements
about the role of LF (such as in Higginbotham's paper "On Semantics")
that define it precisely as the grammatical representation of meaning,
or semantics as far as linguistics is concerned.  So, it turned out I
was wrong: the official GB position does acknowledge that LF is a
semantic level of representation.  Now, you may think there are
empirical reasons for disagreeing with the GB position.  That's
another matter, but let's be clear about what things are and what they
are claimed to be.

Further on, Miriam says:

> 
> That is, both GB and LFG ultimately believe in an independent level of
> semantic representation

I have to disagree here.  It's difficult for me to say what GB and LFG
believe in (especially if you haven't met them personally).  So, all I
can talk about is what people working under these labels do or say.
We agree that an independent level of semantic representation is the
accepted position in LFG (at least, in theory).  Higginbotham, in his
paper "Elucidations of meaning", is very clear about what I would
consider to be the standard position regarding semantics in GB.  He
says that all you have to learn about the meanings of sentences are
the meanings of the words that make up the sentences: if you know the
syntax and you know the meanings of the words, then you know the
meanings of the phrases and sentences that these words make up.  This
is a very clear statement of the NONindependence of semantics in GB.
You can't take a more extreme position than this.  It is not a
hypothesis that can easily be proved to be empirically incorrect or
theoretically undesirable.  So, until this is done, it will continue
to be assumed in GB, because it appears to be a very restrictive
hypothesis about the syntax/semantics interface.

Alex Alsina