LINGUIST List 5.118

Wed 02 Feb 1994

Disc: The Last Phonological Rule: reply to Goldsmith

Editor for this issue: <>

Directory

1. Alex Monaghan, The Last Phonological Rule: reply to Goldsmith

It was great to read John Goldsmith's reply to my review, and I agree
entirely with the list administrators that this is exactly the sort of
interaction to which LINGUIST is ideally suited.

I have a few comments/clarifications to make. firstly, in referring to
the various chapters in the book as "working papers" I was using scare
quotes rather than citation quotes: much of the book had a "working paper"
feel to it for me. This is reinforced by phraseology such as "I ... can do
no more than invite the reader to reconsider some of these questions with
me" (p.21), "The phonological conclusion that the present paper aims towards"
(p.22), "Virtually all of these final remarks are speculative at this point"
(p.56), "This paper is an attempt to confront [problems of universality and
naturalness in phonology]" (p.61), "This chapter reports on initial results"
(p.146), and other similar usages.

I do not mean to dispute the validity of publishing the material in the book,
or to suggest that it would be unacceptable as a series of journal articles.
in fact, I applaud the authors for their honesty: this is, after all, a
largely unploughed field and it is thus difficult to be certain about one's
results. moreover, I think that in many cases these chapters are far above
the standard of the majority of journal articles, since the latter often
present dubious or unfinished work as though it had been received direct
from the relevant cultural deity.

As far as the connectionism or otherwise of harmonic phonology is concerned,
I certainly got the impression that HP assumes this n-dimensional space (very
similar to state-space) and that the symbolic rules had effects in this space.
I find that counter-intuitive, since not all the possible points are
observable in language. I also find the fact that a small number of such
rules can combine to produce effectively unpredictable results unintuitive.
neither of these feelings has anything to do with the Goldsmith/larson
implementation (i assume this isn't THE gary larson!): my misgivings about
that are quite different.

One further brief point. in his reply, Goldsmith mentions "a
finite-dimensional space, the space of connection weights", but as far
as I can tell this space is actually infinite: there is an infinite
range of possible values for each connection weight, and this is an
unavoidable mathematical consequence of gradient systems such as
connectionism. this is the source of one of my persistent doubts about
connectionist approaches to phonology.
 alex.