LINGUIST List 5.83

Sat 22 Jan 1994

Disc: Goldsmith, The Last Phonological Rule

Editor for this issue: <>


  1. , Book discussion: Goldsmith, The Last Phonological Rule

Message 1: Book discussion: Goldsmith, The Last Phonological Rule

Date: Mon, 17 Jan 94 11:16 CST
From: <>
Subject: Book discussion: Goldsmith, The Last Phonological Rule

[Below is another contribution to our Book Discussion Forum. We
hope these contributions will be explicit enough that even those
who haven't yet read the book will feel free to comment on the
ideas involved. And the author(s) are cordially invited to join in.]

John Goldsmith (ed.), "The Last Phonological Rule".
University of Chicago Press: Chicago 1993. 240pp.
Reviewed for LINGUIST by Alex Monaghan, Dublin City University

The book consists of a lengthy introduction and 6 chapters with
different authors: Goldsmith, Mohanan, Lakoff, Wheeler & Touretzky,
Karttunen, Hyman. I'll deal with the various bits in rough order of
appearance below.

A few general things first. I was looking forward to reading this book,
as I had been interested by Goldsmith's connectionist-style articles and
expected this to be a properly worked-out and nicely-presented account
of a connectionist theory of phonology. I was disappointed on both
counts: the book is a collection of "working papers" on non-SPE-based
accounts of phonological processes, and the actual editing and general
presentation leaves much to be desired. Nevertheless, I learnt a lot
from the various papers and I would recommend it to anyone interested in
declarative or connectionist approaches to phonology. I am still not
convinced by the claims of Goldsmith and friends, but my interest was
certainly increased by reading "The Last Phonological Rule".

In the introductory chapter, Goldsmith clearly states the tenets of his
approach to phonology: SPE-style phonological rules are long overdue for
the sort of revision which phonological representations have undergone
in the last two decades (roughly, since Goldsmith's thesis); the basic
concepts of rule-ordering and unattested derivations in phonology are
unnecessary and cognitively implausible; and the obvious remedy is to
take a connectionist view of phonology as, roughly, a multi-dimensional
space ("state-space" in connectionist terms) where the progression from
underlying forms to surface phenomena is a traversal of this space.

In Chapter 2, Goldsmith fleshes out this approach by giving the fullest
account to date of his theory of Harmonic Phonology. He proposes three
levels of "phonological" representation: the morphonemic level (M), the
phonemic level (W) and the phonetic level (P), together with intra-level
and cross-level rules. These levels are crucially unordered, so that M
is as close to W as it is to P, allowing M-P rules. The central thesis
is stated thus (p.22): "all rules which apply at a particular level have
the explicit function of moving a representation as far as possible
toward meeting the phonotactics of that level".

This approach seems to me to suffer from the same problems as beset
connectionist approaches to syntax, i.e. that the mechanisms proposed do
not bear much relation to our intuitions about how language works, and
the combination of these simple mechanisms rapidly makes it difficult to
tell how they are achieving their results, thus obscuring the nature of
the very phenomena which they are intended to explain. The connectionist
response to this might be expressed by a comparison between two
analogies: SPE sees phonological rules as an abacus, where the various
beads symbolise different segments/features/processes which can clearly
be seen to move along their respective rods, and the linguist interprets
the resulting configurations; harmonic phonology is more like a digital
calculator, where the actual processes employed are much simpler and
more limited, but the linguist doesn't see how the process is carried
out --- we just push the buttons and wait for the answer to appear.

Whether the calculator is applicable to phonology remains to be proven.
There are, however, definite problems with Goldsmith's button-punching.
Firstly, there are various stages at which a number must be specified
which may vary infinitely (even within the conventional 0-1 range). The
choice of this number is frequently arbitrary, amd there is a
potentially infinite number of other choices which would produce the
same behaviour. Developing rules, or deciding which is the best rule,
therefore has serious practical problems. Secondly, it appears that only
a very small range of numbers is actually required to model natural
languages, but the approach allows for a much larger (infinite) range:
surely this is a major theoretical problem --- why do languages not make
use of the full range of possibilities? Thirdly, although Goldsmith
states that his rules have the effect of pushing a representation
towards an attested surface form it is not clear how they guarantee to
push hard enough or far enough: moreover, in the computer implementation
discussed at the end of the chapter they appear to fall back on devices
such as intrinsic ordering and the Elsewhere condition.

Mohanan's chapter is in many ways similar to Goldsmith's, although with
different concerns and without an implementation. It addresses the
problem of cross-linguistic regularities or tendencies in natural
languages, and proposes that these can be represented as poles of
attraction in state-space so that grammars tend to cluster around
particular points/configurations (e.g. lots of coronal consonants, few
voiceless nasals). The basic idea is that these attractors replace rules
and constraints in phonology, and that some attractors are in UG whereas
others are language-specific.

This all seems very sensible until we come to the question of
naturalness: Mohanan suggests that languages which come closest to these
attractors are more natural than those with, say, a couple of voiced
nasals. I find it bizarre to attribute such status to statistical
tendencies amongst phonological systems: what if the world's languages
happened to have more clicks than nasals, would this be evidence that
the natural characteristics of human languages had been perverted by a
prehistoric invasion of extraterrestrial insects, or simply that
gunpowder was first discovered in the Kalahari?! Mohanan attempts to
push much of this off onto phonetics at the end of the chapter, but I
don't like it. The second half of chapter 3 gets vague and weird, and I
don't have space to go into why. Nonetheless, the basic idea of
attractors replacing SPE rules is an interesting one.

Lakoff (chapter 4) is another overt connectionist, and proposes similar
ideas ro Goldsmith. The thing which struck me here was the lack of any
proof that the mechanisms proposed were actually substantively different
from more conventional approaches, and in fact the implementation of
Lakoff's approach discussed by Wheeler & Touretzky (chapter 5) appears
to fall back on more conventional mechanisms.

Karttunen (chapter 6) and Hyman (chapter 7) concentrate on the power of
conventional phonological rules, but from rather different perspectives.
Karttunen points out that SPE-style rewrite rules are actually only as
powerful as FSTs, suggests that if phonology were done by
Koskeniemi-style machines this would actually be more elegant and could
be seen as avoiding rule ordering and intermediate derivations. Hyman
looks at two problems of Bantu tonology to see whether there is any
strong argument for or against rule ordering, and concludes that it
depends on your definition of elegance: he also points out,
interestingly, that nobody has argued against extrinsic rule ordering on
grounds of lack of insights or generalisations, and that there is no
reason to believe that phonology can't involve both sequential and
simultaneous rule-application (although it would be nice if it didn't).

In sum, there are a lot of interesting new ideas in this book, and some
of them will doubtless influence phonology significantly. However, I
wouldn't like to say which ones! It does seem to be the case that
phonological rules are currently more powerful than they need to be, and
that a simplification of phonological formalisms and devices would be a
GOOD THING, but whether the PDP/connectionist approach constitutes such
a simplification or whether it just adds to the mess is unclear.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue