more semantics

[Alex Alsina]

I think some interesting points have come up in this discussion on
semantics and some light has been shed on shady areas (at least, as
far as I'm concerned).  I'll make some minor comments about points
that have been made before talking again about alternative conceptions
of semantics and the syntax/semantics interface.

Gert Webelhuth says:

> It is not necessary for GB to make the strong claim that every
> sentence that is n times ambiguous have n distinct syntactic
> representations. Bob May, for instance, in his book on LF (was it
> May (1985) ?) allows for one LF to have more than one distinct
> semantic interpretations.

If this is true (and by "n distinct syntactic representations" Gert
means "n distinct LF representations, which are semantic in substance
but syntactic in appearance"), there is something I'm missing.  I take
the following assumptions to be valid:

	1. Ambiguity (as opposed to vagueness) is a linguistic
phenomenon and therefore should be captured by the grammar and
represented in some linguistic level of representation.

	2. LF in GB is the level of representation that deals with
linguistic meaning and therefore any distinction involving linguistic
meaning must be represented at LF.

>>From these assumptions it follows that, in GB, ambiguity, which has to
do with meaning, must be captured at LF; in other words, in GB, an
ambiguous string of words must have alternative grammatical
representations, distinct at least at LF, one for each of the
alternative meanings.  If this is not correct, it must be because one
of the two assumptions in 1 and 2 is not valid, but I wouldn't know
which one or in what way it doesn't hold.  I took a look at May's book
but I didn't see any reference to the idea that a single LF
representation could have more than one semantic representation, but
maybe I didn't look in the right places.

I totally agree with Gert on the following point:

> 	What I personally find most worrisome about the relationship
> between syntax and semantics in GB is the claim that there are
> universal syntactic configurations that must exist in sentences
> with certain meanings (e.g. the Kratzer/Diesing claims about 
> VP-internal vs. external subjects of different types of predicates).

This is a consequence of the view generally taken in GB that
linguistic semantics (i.e., LF) is largely isomorphic with syntax and
can only differ from the latter in very restricted ways, which are
modelled by movements of syntactic constituents (and maybe in a few
other ways such as reconstruction, which is the undoing of movement
and deletion operations).  Since LF is derived from syntax by these
very limited means, it follows that a semantic distinction cannot
appear at LF unless it is somehow represented in the syntax (i.e.,
s-structure).  If a semantic distinction has a syntactic correlate, it
is by definition linguistically significant and should therefore be
represented at LF.  But, given the derivative or interpretive nature
of LF, it cannot appear at LF out of the blue but must be represented
as a syntactic distinction at s-structure.  The standard approach to
this within GB is to syntacticize the syntactically relevant semantic
distinction, which allows one to maintain the unidirectional
derivation from s-structure to LF, which prevents LF from imposing
constraints on the syntax.

There is an alternative, which is inconsistent with the standard GB
view of the syntax/semantics interface.  According to this
alternative, there is a level of representation in which all (and only
the) linguistically significant semantic information is expressed,
which we can call Grammatical Semantics, using Tara and K.P.
Mohanan's term.  In this respect, Grammatical Semantics is like LF,
but it differs from the latter in that it is not derived from syntax,
but is generated independently of syntax and is related to syntax by
mutually constraining principles of correspondence.  Information
represented at syntax may constrain the associated information
represented at grammatical semantics, and vice versa: the information
represented at grammatical semantics may constrain the associated
information represented at syntax.  This bidirectional interaction
between syntax and semantics is not allowed in the GB model.

According to the Grammatical Semantics view of things, would the claim
that I cited by Higginbotham be true, namely, the claim that, in order
to know the meanings of sentences, ALL you need to know is the syntax
and the meanings of the words that make up the sentences?  (I
emphasize the ALL, because Dick Hudson's latest message suggests a
possible misunderstanding: there is no doubt that the syntax of
sentences and the meanings of words influence the meanings of
sentences.  That is consistent both with the LF approach and with the
Grammatical Semantics approach outlined here.  What distinguishes the
two approaches is whether that is ALL the information you need.)
According to the Grammatical Semantics approach, that is not ALL you
need to know.  Jackendoff's work, for example, provides many instances
of this: semantic information that is syntactically significant, but
is not derivable only from the syntax and the meanings of words.  See
chapter 10 of Jackendoff 1990 for some clear examples of this.

One of the assumption that the GB approach to semantics makes is the
separation of lexical semantics and phrasal semantics.  Phrasal
semantics, that is, the semantics of phrases, is not built out of
lexical semantics, that is, the semantics of words or morphemes,
composed according to the syntactic constituency.  Lexical semantics
is acknowledged to be relevant for syntax, so the meanings of words
(or morphemes) are assumed to constrain the syntactic properties of
words (or morphemes).  Once we enter the domain of syntax, words (or
X{o}s) are assumed to be syntactically and semantically atomic.
Whatever the lexical semantic representation of morphemes and words
may be, it is assumed to be irrelevant once we are in the phrasal
mode.  LF, which is derived from syntax, cannot see the semantic
representations internal to morphemes.  (The Hale and Keyser approach
I believe makes this observation vacuous by representing lexical
semantics in syntactic terms.)  So, the following scheme seems to be
characteristic of GB:

lex. semantics --> lex. syntax --> phrasal syntax --> phrasal sem.

This scheme should be read as follows: lexical semantics determines
the syntactic properties of words, which are the building blocks of
phrasal syntax; the semantics of phrases and sentences is derived from
their syntactic structure without accessing the internal semantic
structure of words.

An alternative to this, which is essentially what Jackendoff assumes,
would propose that lexical semantic representations not only constrain
the syntactic properties of the words associated with them, but are
the building blocks of phrasal semantics.  In other words, the
semantics of phrases and sentences is made up directly of the semantic
representations of lexical items (whether they be morphemes or words)
composed according to principles internal to the semantics,
constrained by the syntactic representations that they correspond to.
The model that describes this would look as follows:

	lexical semantics <--> lexical syntax
		^		     ^
		|		     |
		v		     v
	phrasal semantics <--> phrasal syntax

It should be clear that either of the two approaches described here is
compatible with the overall framework of LFG. I will not try to draw
any conclusions about which of the two approaches is adopted in any
particular theory developed within the LFG framework, or any other
framework for that matter.  I have tried to outline the options
available.  Maybe this will help a critical reader decide which type
of approach a particular theory adopts.  For example, Joan Bresnan's
comments about some current semantic theories developed within LFG
have helped me see whether these theories adopt what I would call the
LF approach to semantics or the Grammatical Semantics approach.  (But
I will say no further about these theories until I have studied them
carefully.) 

Alex Alsina