Semantics

[Alex Alsina]

Chris Pinon, in his latest message to this list, says:

> (I have a
> number of doubts about Alex's point (2), but I can put these aside as
> well.)
> 

Why keep your doubts to yourself?  Although I don't have much of an axe 
to grind myself about semantics and how it interfaces with syntax, I 
think this is a sufficiently important issue to deserve having a lively 
discussion about.  Maybe other people will feel they have something to 
say about it, if properly pricked.

I claimed in my previous message (my point (2)) that an area of difference
between LFG and GB is that GB assumes the hypothesis of the Non-Autonomy
of Semantics, whereas LFG is strongly predisposed towards the opposite
hypothesis, namely, the Autonomy of Semantics.  I realized later that it
is incorrect to claim that semantics in GB is completely non-autonomous. 
After all, there are operations (movements) that take place in LF and not
in the (redundantly called "overt") syntax.  (LF is, of course, the
syntactically relevant representation of semantics in GB, although it is
formally represented as syntax.  Therefore, LF is not a syntactic level of
representation, even if it looks like it.)  These movements at LF are
necessary to capture mismatches between semantic scope and syntactic
constituency.  Then, semantics in GB is autonomous only to the extent that
some syntax/semantics mismatches are allowed, those that are amenable to
movement of syntactic constituents.  Given that semantics has some degree
of autonomy in GB, I will change the names of the relevant hypotheses: the
"Reduced Autonomy of Semantics" or RAS, in GB, and the "Full Autonomy of
Semantics" of FAS, typical of LFG.

What are the potential areas of difference between the RAS and the FAS?  
I think there are two.  (1) The RAS assumes a unidirectional interaction 
between syntax and semantics: phrasal semantics is derived from phrasal 
syntax.  This means that a semantic distinction with a syntactic 
correlate can only exist if it is syntactically represented.  In other 
words, the syntactic reflexes of a semantic distinction are not analyzed 
as being due to the semantic distinction, but as being an outcome of a 
syntactic distinction, which is subsequently interpreted as a semantic 
distinction.  The FAS, in contrast, assumes a bidirectional interaction 
between syntax and semantics.  This allows a semantic distinction not to 
have a uniform syntactic representation, but to have syntactic correlates 
thanks to correspondence principles relating syntax and semantics.  (2) 
The RAS assumes that semantic (LF) representations must respect lexical 
(or, at least, morphemic) integrity: the semantic information that 
corresponds to part of a morpheme cannot be treated or accessed 
separately from the remainder of the semantic information that makes up 
that morpheme.  The FAS, in contrast, allows this.

The RAS is a more restrictive hypothesis than the FAS about the 
syntax/semantics correspondence.  But it is made empirically vacuous with 
respect to the first area of difference with the FAS, point (1) above, by 
syntacticizing whatever semantic information appears to be syntactically 
relevant, that is, by representing semantic distinctions as syntactic 
distinctions.  And it can be shown to be empirically inadequate with 
respect to point (2), because there are instances of violation of lexical 
or morphemic integrity in the sense intended here.  Maybe one day someone 
will give a demonstration of these points.  Meanwhile, one should be 
aware of the potential differences between these two hypotheses and 
should realize that, if one adopts the FAS, which is the natural 
hypothesis to adopt within LFG, it has certain consequences, one of them 
being that semantic distinctions should be represented as such (in the 
semantic structure/s) and should not be duplicated at other levels of 
representation.

Alex Alsina