Hi David,
On Wednesday 19 October 2005 06:33, Dr Hatch wrote:
> ...as I flunked my phonology course as a fresher, perhaps someone explain
> what the hell all this has to do with that subdiscipline in NORMAL
> language.
:-)
The issue runs like this. Chomsky's first argument for innatism, as I
understand, was not "poverty of stimulus", it was phonological.
The argument was not so much that (phonological) representations could not be
learned (i.e. "poverty of stimulus"), but that learning led to certain
irreconcilable problems. To borrow from John Goldsmith's earlier posts:
'The Hallean argument against the phoneme took the form: if you apply American
structuralist principles, then you are faced with the conclusion that the
same phonological rule will sometimes be labeled a "rule of
allophony" (regulating the allophones by which a phoneme is realized) and
sometimes a rule at a higher level, a rule of morphophonology.'
and later...
'a phonemicist analysis of Russian requires two rules that are very similar to
bear different labels ("rule of allophony" and "rule of morphophonology").'
That seems a bit technical, but what I understand of it is this: if you learn
purely from patterns in the raw data, the rules you find cause you to "chase
your tail". You can't define phonemes without taking into account how they
combine, but you also can't take into account how they combine until you
define the phonemes.
Or perhaps I am getting my problems mixed up... Anyway, the point was you got
problems in the rules which could be learned directly from the data, and that
led Chomsky to propose something innate must guide the learning process, to
sort the wheat from all the chaff.
This is the debate I'm looking for evidence of in machine learning circles,
not "poverty of stimulus".
I agree there are lots of good refutals of "poverty of stimulus" around. I
want to see discussion of the idea (distributional) learning led to problems,
inconsistencies, "mixing of levels" and such, in the rules.
-Rob
This archive was generated by hypermail 2b29 : Wed Oct 19 2005 - 03:39:23 MET DST