LINGUIST List 7.1742

Mon Dec 9 1996

Disc: Analogy

Editor for this issue: Anthony Rodrigues Aristar <aristarlinguistlist.org>

Directory

1. Michael Israel, Re: 7.1692, Disc: Analogy
2. James L. Fidelholtz, Re: 7.1692, Disc: Analogy
3. Daniel L. Everett, Re: 7.1692, Disc: Analogy

Message 1: Re: 7.1692, Disc: Analogy

This is in response to Stephen Straight's response to Peggy Speas'
reply to Esa Itkonnen's posting on analogy. Peggy offers a standard
objection to analogy as an explanation of linguistic creativity.
The basic problem is how to constrain analogy so as to account for the
all of the analogies people do not make. As Speas puts it,

>>To say that language works 'by
>> analogy' simply begs the question - which analogy? Of course
>> English speakers draw analogies like the following:
>>
>> play : plays :: glark : glarks
>>
>> But Chomsky's point is that there are lots of reasonable
>> analogies that no English speaker ever draws. Like:
>>
>> John is easy to please : To please John is easy ::
>>
>> John is eager to please : To please John is eager.
>>
>> So the question is WHY speakers make some analogies and not
>> others. The claim is not that language cannot involve analogical
>> reasoning; it's just that you have to investigate WHICH analogies
>> are made and which ones aren't in order to get at the root of
>> knowledge of grammar.
>

I agree with Stephen that this is not a very compelling objection;
however, I think the reason is rather simpler than what he suggests.
The reason is just that the proposed analogy is not in fact reasonable at all.

Linguistic analogy (at least at the level of syntax, and probably morphology
as well) does not operate on formal structures alone; rather
it operates on form-meaning pairs. Thus while sentences with 'easy'
and sentences with 'eager' may appear formally similar, they in fact
encode, as is well-known, very different sorts of semantic relations
between their predicates and arguments. Analogies between such
sentences are thus not made for the simple reason that such sentences
are not in fact analogous: they do not encode similar relations between
form and content.

Stephen suggests that the fact that researchers can come up with these
sorts of analogies shows that they are in fact possible. He then suggests
that the reason we find no evidence for such analogies is that they
present insurmountable problems to processing and/or comprehension
and are therefore suppressed.

The claim then is that a general mechanism of analogy plus general
constraints on processing can be relied on to explain which sorts of
analogical change do occur and which don't, all preferably without any
need for anything like an abstract (innate, encapsulated and domain-
specific) Universal Grammar. As Stephen puts it:

>the ongoing interaction of the acquirer's developing comprehension
>and production processing -- which we need anyway -- forces
>various outcomes as the acquirer endeavors to make sense of input
>and to produce intelligible output -- without any need for
>innate, overarching principles of language structure.

To my mind, this is fine as far as it goes. But surely, in order for analogy
to operate at all, speakers must rely on linguistic structures of some sort
which they can feed into their analogical equations. And so in order to
have a theory of analogy, we will in fact require principles of language
structure of some sort.

My suggestion is just that since syntactic and morphological analogy
apparently require correspondences between both form and meaning
in linguistic structures, we might prefer a theory of linguistic structures
which allows us to capture such correspondences in a fairly direct way.

There is currently no shortage of such theories available on the theoretical
market. Basically, any sign-based theory, one in which the basic linguistic
units are taken as pairings of form and meaning, will do. Cognitive Grammar,
Construction Grammar and HPSG spring to my mind. I know there are others
that will spring to others' minds.

This of course does not solve the problem of deciding what sorts of 
similarities speakers will draw between analogous structures. But at least 
it reduces the problem, or at least relates it, to the more general cognitive 
problem of how people manage to make correspondences between distinct 
conceptual structures in general, as between the structure of an atom and 
that of the solar system, or between a complicated love affair and a 
difficult journey through strange and dangerous lands.

The point is that analogy as an explanatory mechanism has not been vitiated,
even if it does remain in need of refinement. The difficult cases should help
us make the needed refinements, both to the theory of analogy in general and
to our theories of the linguistic structures that might feed linguistic
analogies.

Michael Israel

Message 2: Re: 7.1692, Disc: Analogy

On Sat, 30 Nov 1996, H Stephen Straight wrote:

> First, the anti-analogists cite alleged impossible analogies, but
> in so doing they provide incontrovertible evidence that such
> analogies _can_ be drawn. (This is reminiscent of the efforts to
> prove to a speaker of language X that something that can be said
> in language Y cannot be said in language X by paraphrasing this
> alleged impossible Y thing in X, as if paraphrases didn't count!)
> 
	Love it! My all-time favorite example is, of course, Dorothy 
Lee's justly famous/reprinted article on the Trobriands' sweet potato. 
After doing her brilliant job of explaining to us perfectly the economic, 
cultural and, yes, linguistic implications of the yam for these guys, she 
finishes up telling us, in effect, 'but of course WE can never understand 
this, 'cause we're not native speakers'. Ken Hale is of course right 
about the relative advantage of native speakers, but, as you point out, 
even us linguists can paraphrase, or even possibly notice something that 
flew right by some native speaker.
	Jim

Message 3: Re: 7.1692, Disc: Analogy

On 30 Nov 1996, H Stephen Straight responded to Peggy Speas' posting on
analogy. The basic form of his rebuttal is that the analogies people fail
to use are ruled out by independent principles, rather than any failure of
analogy per se. He calls these 'principles of comprehension and
production'. But Speas also claims that such analogies are ruled out by
independent principles, which she calls grammar. Straight claims that his
independent principles lead to a more parsimonious account because we need
comprehension and production principles independently. But surely this
argument only goes through if one believes that there is no independent
justification for grammar. This would be ironic since there is a lot more
research on explicit principles of grammar than there is on comprehension
and production. And even if you think that this quantitative judgment is
false, there is no denying that the kinds of sentences Speas discusses
have a long tradition of analysis behind them. If one thinks that the
grammatical analysis of such constructions is inferior to a
production/comprehension account, as Straight seems to, then one must show
how. Straight's rebuttal falls short of its goals. Since the issue is
important, it is worth pointing out where the problems lie.


Quoting Speas, Straight's case begins:

> >... But Chomsky's point is that there are lots of reasonable
> > analogies that no English speaker ever draws. Like:
> >
> > John is easy to please : To please John is easy ::
> >
> > John is eager to please : To please John is eager.
> >
> > So the question is WHY speakers make some analogies and not
> > others. The claim is not that language cannot involve analogical
> > reasoning; it's just that you have to investigate WHICH analogies
> > are made and which ones aren't in order to get at the root of
> > knowledge of grammar.

Straight replies:
> This line of argument has always struck me as self-defeating, in
> at least three ways:
> First, the anti-analogists cite alleged impossible analogies, but
> in so doing they provide incontrovertible evidence that such
> analogies _can_ be drawn.

Straight misses the point here. The point is not that analogies can or
cannot be *drawn* but whether they are causally implicated in
grammatical/language development. As Speas points out, the kind mentioned
above never are, although some types may be. There are good grammatical
reasons for this and those reasons are independently motivated, so the
issue of parsimony does not arise with respect to the grammatical
treatment.

> Second, the anti-analogists claim that language acquirers never
> make the cited analogies, even though the examples they give
> typically reveal that such analogies result in output that would
> be extremely hard to interpret using the comprehension and other
> processing strategies that many other completely non-analogical
> cases force upon the acquirer (as in the "To please John is
> eager" example). The possibility remains, therefore, that
> evidence for such false analogies fails to emerge because it is
> immediately suppressed by the acquirers themselves because of its
> uninterpretability. (Shades of colorless green ideas rise up to
> haunt us.)

This also misses the point. Speas (and a generation of syntactic studies)
has already shown that these analogies are not drawn for grammatical
reasons. Again, unless one is trying to say that grammar is unnecessary
(and the best theories of processing that I am aware of do not replace
grammar, they rely on it), appeal to vague notions of processing by no
means enjoys any obvious epistemological priority over appeal to clear
grammatical principles.


> Third, the anti-analogists explain the (alleged) non-existence of
> such false analogies by appealing to supposed universal
> principles of "grammar" known to the language acquirer and yet
> somehow separate from the acquirer's developing processes of
> language comprehension and production.

We must read different books. Grammatical theory, at least in the
Chomskyan tradition, interacts in many fruitful ways with processing
theory (which includes comprehension, although I am not altogether sure
what Straight means by 'production'). Some extremely promising processing
theories (conceptually and empirically) are the ones that interact most
closely with grammar, rather than try to supplant it.

- Dan Everett