> From: Steve_Finch_at_TTSGMD@thomtech.com
> Date: Tue, 12 Mar 96 10:47:08 EST
>
> Surely the competence/performance distinction cannot be justified by the
> "infinite number of sentences" argument? Any more than an "infinite
> number of facial expressions" (most of which will never be expressed)
> justifies a competence/performance distinction for inter-muscular
> dependencies in the muscular movements within the face, or the
What I was trying to say is this:
A ``competence'' theory which merely classifies objects (sequences of
words, combinations of facial expressions or what have you) into two
classes: viz. ``possible'' and ``impossible'', says nothing about the
distribution of these objects. If you want to account for the data
you actually observe you will need to supplement your binary
classification somehow with a ``performance model''; e.g., define a
probability distribution on the class of ``possible'' (and
``impossible''?) objects.
Thus the categorical nature of the competence grammars (i.e., the way
that sentences are classified into a small number of categories)
requires an additional ``performance'' component to any language model.
But if you abandon the binary classification of standard competence
models, e.g., by using a probabalistic grammar, then the things that
need to be explained by the performance model change.
For example, it is conceivable that a probabalistic grammar would
assign vanishingly small probabilities to deeply centre-embedded
structures. If so, the non-occurence of such structures in natural
text would not have to be explained by a performance model.
But as Philip Resnik points out, there is much more to performance
than centre-embedding and the probability distribution of sentences in
a corpus.
Mark
---------------------------------------------------------------------
Rank Xerox Research Centre Tel: (33) 76 61 50 37
6, chemin de Maupertuis (33) 76 61 50 50
F38240 Meylan
FRANCE Fax: (33) 76 61 50 99
---------------------------------------------------------------------