LINGUIST List 2.507

Fri 13 Sep 1991

Disc: Compositional Semantics

Editor for this issue: <>

Directory

1. "Larry G. Hutchinson", Re: 2.496 Queries
2. John Coleman, RE: 2.496 Queries
3. "Michael Kac", Re: 2.496 Queries
4. "John Nerbonne", Ad: Question: "compositionality" of semantics
5. "Bruce E. Nevin", compositionality of semantics

Message 1: Re: 2.496 Queries

wrt Stephen Spackman's question about "compositionality":
at minimum, this means that the meaning of a sentence is a function of the
meanings of its parts. It's the imprecision of the terms "function" and
"parts" (in this context) that tends to derail discussion. The strictest kind
of compositionality requires that there be a procedure for computing the
meaning of any SUBPART of a sentence from that subpart's subparts. For
example, if your grammar claims a certain sentence consists of NP + VP, then
the string constituting the VP must itself have a computable meaning,
derived via a procedure which applies to the string's subparts. The strictness
of "strict compositionality" is illusory, of course, if there is any debate
about syntactic subparts!!
Idioms, in the traditional sense, are often cited as proof that semantics
can't be compositional in natural languages. And many argue that they prove
nothing. Nominal compounding can also be claimed to prove noncompositionality.
My favorite example, which has been kicking around for years, is "Susan dated
an occasional sailor." Strict compositionality would seem to require that
there be a procedure which determines what "an occasional sailor" means. Or
"occasional sailor", if that string is itself a syntactic unit in your grammar.

Message 2: RE: 2.496 Queries

Stephen P Spackman asks "what would a NON-compositional semantics be
like?"
Two possible examples spring to mind. One is idioms (phrases whose meaning
isn't a function of the meanings of their constituents and the way in which
they're put together --- unless it allows very specific ways of putting those
particular constituents together to define a single idiomatic interpretation).
The other hinges upon the necessity for the combination of meanings of the
constituents to be a FUNCTION. So, if the meaning of a phrase was computed
by putting the meanings of the words together in a non-functional fashion
(by some kind of nondeterministic procedure, say), the meaning of the
whole would not be compositional. It might be different in random ways
at different times, for example. (Wierd, huh?)
--- John Coleman

Message 3: Re: 2.496 Queries

Stephen Spackman inquires about what would and would not constitute a com-
positional semantics.
The essence of compositionality as I understand it (others may understand it
differently) is that the semantic value of every composite expression is a
function of the parts of that expression and their manner of combination,
a principle commonly attributed to Frege (though I'd be surprised if it didn't
have antecedents farther back). It's in some ways a very strong requirement
to put on a semantics but in other ways a rather weak one.
For some ideas about non-compositional semantics (and arguments to the effect
that natural language semantics isn't in fact compositional), I would refer
anyone interested to some recent work by Alexis Manaster-Ramer and Wlodek
Zadrozny on exploiting the notion 'construction' in analysis and in parsing.
They may weigh into this discussion themselves and I won't presume to speak
for them. Wlodek and I have had some animated discussions on the issue.
Michael Kac

Ad: Question: "compositionality" of semantics
>> at the extremes it seems that "having a
>> compositional semantics" is used for both "ignoring pragmatic issues
>> completely" and "having some FORMAL theory of interpretation, with or
>> without reference to 'meaning'"
The question of semantic compositionality is generally only raised
within the the camp of semanticists who view natural language
semantics as a kind of model theory---as the kind of effort one
undertakes to interpret a logical language.
Within this camp "compositionality" has a precise meaning---a semantic
interpretation function is compositional iff the interpretation of a
syntactically complex constituent depends functionally on the
interpretation of its constituents. This is discussed at length in
the standard introduction to semantics, Dowty, Peters & Wall's {\it
Introduction to Montague Grammar}. It derives from Montague's work,
and seems to be an articulation of one of Frege's principles.
Compositional semantics needn't ignore pragmatics, and there are
examples by Montague showing how indexicals may be treated
compositionally. Karttunen & Peters did an early treatment of
implicature, etc. (I've also heard the term used among computational
linguistics to designate the level of semantic representation derived
directly from syntax. Used in this sense, "compositional" embodies
few or no hypotheses about the nature of semantics. Maybe it's
the source of your "ignoring pragmatics" sense.)
>> [...] what could a NON-compositional semantics be like?
A semantics might eshew compositionality in favor of operating in a
constraint-based fashion. For example, a semantics for NP VP
combination might require that the the NP semantics bind the subject
position in the relation denoted by the VP without saying anything
about scope of the subject NP---which could be determined by other
factors. Then the semantics is less than compositional.
--John Nerbonne
 nerbonnedfki.uni-sb.de

My understanding of compositionality is that the semantic attributes
of a complex expression should follow from those of its elements.
Easier computation, for obvious reasons. Cf. IMO Aristotle's
fallacy of composition for one sort of problem.
	Bruce Nevin
	bnbbn.com