LINGUIST List 2.540

Sat 21 Sep 1991

Disc: Compositional Semantics

Editor for this issue: <>

Directory

1. Paul Saka, RE: compositionality
2. "Michael Kac", Re: 2.523 Compositional Semantics
3. stan kulikowski ii, re: compositional meanings
4. , 2.523 Compositional Semantics
5. Alexis Manaster Ramer, Compositionality

	There is a dissertation on non-compositional semantics
by Jean Kazez, U of Arizona, 1990. While I do not have Jean's
current e-address, I know that she's in the philosophy dept at
Southern Methodist U.
	As others have observed, idioms do not threaten compo-
sitional semantics. However, contrary to a recent posting,
ambiguity very well might. The recent proposal, recall, says
the following:
	Let us suppose that expression E consists
	of constituents C1...Cn arranged in structure X.
	If E is ambiguous between meanings M1 and M2, then
	we cannot say that f(C1...Cn, X) = M1 AND that
	f(C1...Cn, X) = M2, for then f would not be a
	function and the semantics would not be compositional.
	Therefore let us say that f(C1...Cn, X) = {M1, M2}.
But this presents a weird view of ambiguity. It says that an
ambiguous expression has ONE meaning, M1-and-M2. In contrast,
what we want is a theory that says an ambiguous expression
means M1 OR M2.
	This does not mean that compositional semantics
is necessarily wrong; it all depends on what kinds of
ambiguity exist. If lexical and structural ambiguity are
the only kinds of ambiguity, then semantic composition
(the function f) could always correctly yield the unique
output that is appropriate to the input arguments (the
constituents and the mode of combination). However, if
additional kinds of ambiguity exist, then compositional
semantics would seem to be a doomed enterprise.
	In view of the evidence for additional kinds
of ambiguity, it is ironic that so many consider
compositionality to be a virtue in their theories.
I am thinking not only of deictics, which are generally
conceded to be something special, but also polysemy,
metaphor, and other kinds of pragmatic ambiguity, all
of which I discuss in a forthcoming paper.

Message 2: Re: 2.523 Compositional Semantics

Dave Chalmers talks about compositionality as a constraint and Kai von
Fintel characterizes it as a methodological principle. My own inclination
is to think that the latter position is the defensible one though I am
saying this partly to open up discussion of the question.
Alexis Manaster-Ramer alludes in his posting to context-sensitivity of a kind
that semanticists don't seem to like (if I'm remembering it correctly). I'd
be interested in hearing more about exactly what he's thinking of.
Michael Kac

 now wait a minute. some of you seem to be saying that there is always a
compositional meaning and sometimes a noncompositional meaning. i use a number
of expressions which i know the meaning of the whole and do not know the
meaning of some or all of the parts:
 ad hoc
 post hoc
 vis a vi
 bric a brac
 et cetera
 flotsom and jetsom
 willy nilly
 pell mell
 helter skelter
 some are obviously other-language inclusions into common parlance and most of
the rest seem to have a rhyme structure. perhaps these are entered as single
lexical items with vaccuous word boundaries? certainly there are very limited
syntactic operations permitted on structures with noncompositional meanings.
about the only one that i can think of is the common-masculine gerundive use of
"the f-word", which seems to have a universal insertion ability between
otherwise bound items yet does not destroy noncompositional constructs...
 well, that about exhausts my noncompositional two f***ing cents worth. :)
 stan
 . stankuliUWF.bitnet

In response to some recent contributions,
I beg to submit the following:
-- The issue I raised with 'dog' and 'dogs' was not about how to
analyze plurals, but rather that it seems to me (and I may be wrong)
that linguists and other NL mavens who talk about compositionality
seem to assume that some kind of nondestructiveness is intended,
although it is admittedly difficult to formalize this notion. Yet
it is precisely that nondestructiveness that is supposed to prevent
e.g. a language in which "trees", "are", and "green" mean what they do
in English, but in which "trees are green" means "the quarterback fainted
last Saturday" (to use Dave Chalmers' nice example).
-- Margaret Fleck is right that, for all practical purposes, a relation
mapping a single syntactic structure to several semantic ones can
be replaced by a function giving us a set of semantic structures,
but we have the right to insist that syntax-to-semantics mappings
yield semantic structures and not sets of the same. Indeed, it
seems to me that, while hard to test, this constraint is the one
part of compositionality that has some teeth. One can certainly
advance linguistic plausibility arguments as to whether there exist
syntactic constructions which are semantically ambiguous even though
there is no purely syntactic evidence that they are in fact more thab
one construction.
-- Again, Margaret is right that computability is not much of a constraint,
but it is a slight beginning. As I tried to argue, it seems to me that
people usually assume something stronger, and it would be interesting
to see if people share my intuitions on this.
-- Finally, what is really required, I believe, is a hierarchy (or
several hierarchies) of compositionality. Just as Turing-computability
(Chomsky-generability) can be constrained to get, e.g., context-sensitive
or context-free languages, so it seems to me that compositionality
in the broadest sense should mean just what I originally said
(any computable function will do), but there should be a variety
of more restricted types (one way, suggested by van Benthem's work,
would be to use the familiar automata hierarchies here, so that
"context-free" functions would be those computable by a pushdown
transducer, for example). There are enormous difficulties with making
this work, but I think it can and should be done.

On the subject of idioms, it is again kind of trivial but perhaps
relevant to note that one always assumes that there is a finite
number of them, whereas the compositional rules generate an
infinite set of expressions. This is very much in the spirit
of the kind of language-theoretic or automata-theoretic treatment
I was arguing for. In the same way that a finite number of special
cases is always tolerated in language and automata theory, so, too,
it would seem to be tolerated in a compositional theory of semantics.
As such, it seems to me that the issue of idioms IS relevant to this
topic, pace what some of the contributors to the discussion have said.