Grammar with a "G"

[Rob Freeman]

I've been kind of distantly following this discussion, but Greg's message came
close enough to my beliefs to make me feel it might be worth posting. Has
anything come up yet regarding 'emergent structure' from 'analogy based'
processing? This has become a vigorous little sub-field of natural language
processing research in recent years: 'example-based', 'memory-based',
'case-based' reasoning etc. Basic idea is that grammar is just the observed
regularities of collections of examples, and analogies to them, which are what
really control our perceptions. You still get grammar, but because the basic
mechanism is analogy it has soft edges.

Rob Freeman
rjfreeman at usa.net

Greg Thomson wrote:

> I liked (sort of) Gerry Altmann's comparision of linguists' descriptions of
> grammars with the periodic table in Chemistry. Regarding the latter, it is
> interesting to see that substances combine in certain proportions,
> resulting in certain "regular" properties. But what is behind all of that?
> That is far more interesting. So also, descriptions of formal regularities
> in languages can be fun and fascinating, but then comes the "so what?" All
> those patterns in the spoken or written production would be there even if
> no linguist looked at them, and they aren't there just too look nice.
> "Grammar" is _doing_ something, and that's what's ultimately interesting.
> Leaving linguistics aside for a minute, we can consider a familiar
> illusion: the illusion of directly perceiving a speaker's thoughts. What I
> mean is, language users listening to speech are often (perhaps, typically)
> rather unaware of linguistic form as they subjectively "grasp" something
> else. How does linguistic (ultimately acoustic) form cause that to happen
> so well and so rapidly? And in what does that "happening" consist? Recall
> Givon's suggestion that we move on to "reinterpret grammar as mental
> processing instructions". How are concepts (let's say, those involved in a
> particular flow of narrative understanding) constructed and managed? Well,
> "grammars", including those of yet-to-be-documented languages, are an
> enormous source of evidence bearing on that more interesting question, to
> the extent that we use them for that purpose. Meanwhile, we still come back
> to the unlikeliness that a  model of the external form of a language will
> turn out to be a useful model of anything inside the language user.