LINGUIST List 2.514

Sat 14 Sep 1991

Disc: Compositional Semantics

Editor for this issue: <>


  1. , 2.507 Compositional Semantics
  2. Richard Coates, Re: 2.507 Compositional Semantics

Message 1: 2.507 Compositional Semantics

Date: Fri, 13 Sep 91 14:15:16 EDT
From: <>
Subject: 2.507 Compositional Semantics
I think it deserves to be pointed out that technically compositionality
has the effect of requiring that every syntactic structure be semantically
unambiguous. Otherwise, the word 'function' in the usual definition of
compositionality would have to be replaced by the word 'relation'. This
is not something that lends itself to factual testing, of course, but
it IS a constraint of sorts. As Michael Kac points out, some of us have
been rather unhappy with this constraint, incidentally, because it prejudges
the question of how syntax relates to semantics. It seems to me that
semantic and syntactic criteria should never be confused, as they necessarily
are in this kind of approach.
I believe, incidentally, that recent PP (or GB, as it used to be called)
literature has some instances of arguments about exactly this point, with
somebody (I forget who) arguing that in certain cases a single syntactic
structure should be assigned two or more distinct meanings (as I would
also argue, but I don't work within that framework).
More significant perhaps is the fact that people usually seem to have
in mind some kind of constraints on what kind of function is involved
in the syntax-to-semantics mapping (as well as constraints on the
form of the syntax). To my knowledge, these have never been specified
formally, but here are some examples that I suspect people would agree
 (1) Assume that dog means 'dog sg.' and that -s means 'pl'. Then
 to get dogs to mean what it does, we would need to somehow suppress
 the singular part of the meaning of dog when combining it with the
 pl. suffix. This is indeed, as I believe, the traditional kind
 of analysis (from ancient Greek times till the beginning of structuralism),
 and perhaps the kind that "naive speakers" tend to come up with. I also
 tend to think that it might be the right analysis. In any case, it
 is not compositional in the informal sense I am trying to clarify.
 That is, it seems to me that the informal notion of compositionality
 embodies some kind of stricture against losing or destroying information.
 (2) Take idioms, e.g., kick the bucket. Now, obviously there IS
 a function that takes the normal meanings of the three words
 and puts them together in such a way that we end up with 'die'.
 However, one might argue that this function offends against the
 constraint against information-loss. Another way of looking at it
 (perhaps applicable to (1) as well) is that we want the syntax-semantics
 functions to be context-free (again, I know of no formal definition)
 in the sense that the same semantics effect should hold in all
 (syntactically indistinguishable) environments. Thus, one might
 try to say that one or more of the words in the idiom has a special
 meaning just in this one collocation, but that would make the semantics
 context-sensitive in some sense that people seem not to like.
The last point is that people seem to assume that the functions must
be computable if the semantics is to be compositional. One could
imagine arguments that NL semantics is not computable, of course, and
that would, I think, count as an argument against compositionality
in the usual sense.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue

Message 2: Re: 2.507 Compositional Semantics

Date: Fri, 13 Sep 91 20:58:14 +0100
From: Richard Coates <>
Subject: Re: 2.507 Compositional Semantics
I don't think it's helpful to drag idioms into the debate (or at least not
all idioms). For a canonical idiom like _spill the beans_, we ought to say
that it is analysable lexically, and that the meanings of its constituent
parts are accessible to an etymologically-minded user. That shouldn't
commit us to saying that *AS AN IDIOM* it consists of units having their
literal senses. It means in that case at the level of its lowest
dominating node, here VP, and not at the level of the terminal and
intervening nodes. (Special arrangements for tense/aspect, of course!)
The meaning of _she spilt the beans_ is a compositional function of the
meanings of {she}, {past}, {spill the beans}.
Richard Coates
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue