LINGUIST List 10.1952
Fri Dec 17 1999
Review: Coleman: Phonological Representations
Editor for this issue: Andrew Carnie <carnielinguistlist.org>
What follows is another discussion note contributed to our Book Discussion
Forum. We expect these discussions to be informal and interactive; and
the author of the book discussed is cordially invited to join in.
If you are interested in leading a book discussion, look for books
announced on LINGUIST as "available for discussion." (This means that
the publisher has sent us a review copy.) Then contact Andrew Carnie at
carnie
linguistlist.org
Directory
Stefan, Re: review of Coleman's book
Message 1: Re: review of Coleman's book
Date: Thu, 16 Dec 1999 20:38:19 -0000
From: Stefan <stefanploch.freeserve.co.uk>
Subject: Re: review of Coleman's book
A) John Coleman, 1998, Phonological Representations. Their Names,
Forms and Powers, Cambridge University Press, Cambridge et al., xii
+ 345 pages.
B) Stefan Ploch, School of Oriental and African Studies (University
of London)
C) Synopsis
Coleman's monograph (`CLM') looks at phonological representations
from the perspective of declarative phonology (DCP). In other
words, the purpose of Coleman's investigation is to provide further
support for the claim (already made in Coleman 1995, Scobbie 1993,
and Scobbie, Coleman and Bird 1996) that underlying
representations, derivations and rewriting rules are unnecessary.
The main advantages of DCP's formal rules over generative
approaches are, in Coleman's view, "explicitness, transparency of
expressive power, logical completeness and consistency" (CLM:9), on
the one hand. For example, Coleman argues that the clear
differentiation between notation and denotation can help
phonologists to be more aware of the restrictiveness of their
respective frameworks (CLM:6). On the other hand, certain general
parsing algorithms for context-free grammars can be used to
determine syllable structure or the ill-formedness of some string
of segments (CLM:9) without having to stipulate a series of derived
levels of representation.
After an introduction and overview in chapter 1, Coleman starts off
his discussion of phonological representations with an
investigation of a series of notation types that are (more or less)
commonly employed by phonological theories (chapter 2). The main
part of this chapter looks at form (notation) and content
(denotation) of the IPA. However, Coleman also provides a
(de)notational comparison between the IPA, Sweet's Organic Alphabet
(Henderson 1971), feature-matrix representation and of a strict, a
relaxed and a superweak segmental interpretation of componential
representation. (These three types of interpretations of
componential representation are frequently employed in acoustic,
articulatory and coarticulatory studies (respectively).) Another
interesting aspect of this chapter (and CLM in general) is that, in
line with the fact that Coleman's approach to representations
distinguishes between (syntactic) notations and (semantic)
denotations, he also makes use of model-theoretic semantics. In
this way, analyses can be made to relate to a model M which
determines a class of possible (speaker- and utterance-independent)
sound type sequences for all IPA expressions and a sequence of
sound tokens for a given phonetic expression, speaker and occasion
of utterance (CLM:31).
In his analysis of the semantics (denotation) of the IPA, a
segmental phonetic notation system, Coleman (CLM:37f.) finds that
each member of the set of IPA symbols denotes the intersection of a
set of physical properties while a diacritic denotes a function
from denotations of basic expressions, i.e. the members of the set
of all elements/symbols of the IPA alphabet (with or without
diacritics) (CLM:26), to denotations of basic expressions.
Furthermore, concatenation denotes temporal adjacency -
cocatenation, temporal overlap. Using a model of coarticulation
that is based on temporal overlap, Coleman accounts for hard
coarticulation and epenthesis as non-phonological phenomena that
find their motivation in the overlap of phonetic components. The
main advantage of this phonetic account, says Coleman, is that it
makes a phonological explanation unnecessary (CLM:45).
In chapter 3, Coleman investigates phonemic and
transformational-generative phonology. This part of the book is
particularly noteworthy because it contains a practical overview
over the historical development of notations and denotations in
phonemic and transformational phonology, of features, types of
oppositions and transformations, and rule types. Coleman is
particularly interested in the major flaws of transformational
approaches: in the invertibility problem (i.e. the difficulty in
deciding whether, in a given transformational grammar with deletion
rules, underlying forms are derivable from surface forms by inverse
transformations), the rule ordering problem (i.e. the difficulty in
formulating a unique rule order such that all universal phonetic
rules follow language-specific phonological rules) and the
excessive power of transformational grammars (often caused in a
theory by SPE-ancestry). The usage of context-sensitive rules with
deletion results in unrestricted rewriting systems whose symptoms
are excessive power, unlearnability or both. Since these flaws are
intrinsic to such approaches and since even a theoretical
preference for general (rather than specific) rules has in
Coleman's view "not been shown to appropriately restrict the class
of languages that can be defined'' (CLM:89), he concludes that
transformational grammars need not be improved upon.
He then goes on to provide a method for the determination of
declarative statements (as opposed to transformational rules).
Subsequently, Coleman provides evidence for his claim that
unification-based phrase-structure grammars ('UPSGs') are a useful
phonological tool in that they allow cross-categorial
generalisations and exhibit less redundancy in the formulation of
rule schemata than 'classical' phrase-structure grammars ('PSGs').
In chapter 4, Coleman looks at the representational mechanism
employed by non-linear theories, i.e. by autosegmental, metrical
and dependency phonology. He finds that the rule formalisms used by
these frameworks are, like their predecessors, far too
unconstrained. In a discussion of the well-formedness constraints
of autosegmental phonology ('AP'), Coleman argues that a decade of
autosegmental research exhibits little improvement in relation to
the avoidance of language-specific rewriting rules over earlier
frameworks; association conventions and the no crossing constraint
('NCC') are flawed in that proposed autosegmental spreading
mechanisms are unnecessary and the NCC is only a constraint on
diagrams (i.e. it is a notational constraint) , not on AP
representations.
In addition to that, Coleman claims that Dresher & Kaye's (1990)
computational learning model for metrical phonology "benefits from
formalisation using the resources of unification-based grammar"
(CLM:17). Dresher & Kaye set up eleven parameters (in relation to
metrical structure), e.g. 'The word-tree is strong on the
[Left/Right]' or 'Feet are [Binary/Unbounded]'. Using the formalism
of UPSG , Coleman can successfully reformulate ten of these eleven
parameters as context-free rules ('principles' without parameters),
e.g. Dresher & Kaye's P(arameter) 5 'Feet are
quantitative-sensitive [Yes/No]' as '(syllable) --> O R'. (There is
no space here to discuss P9 'A weak foot is defooted in clash
[No/Yes]', the only parameter that Coleman could in his opinion
only formulate in UPSG terms if he was more familiar with this
particular topic in metrical phonology.)
In chapter 5, Coleman provides an introduction to basic concepts
employed in DCP: In opposition to the phonological mainstream,
which attempts to find a formalism that is able to account for a
given set of data and, subsequently, to check whether this
formalism is constrained enough, DCP selects a formally constrained
formalism first and then to adapt it to phonological problems.
Phonological relations are expressed in the representations rather
than in the rules (CLM:165) via context-free immediate dominance
and linear precedence constraints. Rules are not formally distinct
from representations (as in e.g. Chomsky 1957) but are partial
representations. Due to the elimination of rule ordering and the
existence of only one level of representation, DCP is
computationally simpler than transformational models. Furthermore,
DCP (supposedly) maintains a strict distinction between phonetics
and phonology: phonological descriptions have no phonetic content.
DCP 'interfaces' with phonetics by the assumption of two (separate)
levels of description: a phonological and a phonetic one. So there
is no unity between phonological labels (that are terminologically
phonetically motivated, e.g. 'voice' (CLM:171)) and the phonetics
(phonetic voicing); any such unity is imaginary.
Chapter 6 contains a declarative analysis of Japanese words that
makes use of Firthian prosodic theory but with a 'classical'
generative notation for features and certain innovative
interpretations. This is followed by a declarative analysis of
English. (Below, I will come back to a number of details of the
Coleman's analysis of English words but will because of lack of
space not discuss his analysis of Japanese.)
D) Critical evaluation
In Coleman's view, transformational systems have not been shown to
be successful in restricting the class of definable languages (cf.
CLM: chapters 3 and 4). Let me say clearly that, even though I do
mostly not agree with Coleman's claims, I do think that Coleman has
a point: transformational theories with deletion or substitution
rules result in the invertibility problem. For example, in GP, the
most constrained of all transformational frameworks (cf. Williams
1998), a low tone element L lexically assigned to the rightmost
onset position in a domain may be deleted via derivation (i.e. at
the phonological level of representation as opposed to the
underlying lexical level) if it precedes a domain-final silent
empty nucleus, resulting phonetically in final obstruent devoicing
(cf. Brockhaus 1995 who provides this analysis for Northern
Standard German.) If such element deletion could occur freely or
without heavy restrictions, this would spell unrecoverability of
underlying forms from derived ones.
However, such transformations are only grammatical in government
phonology ('GP'), i.e. Brockhaus' framework, in a highly
restrictive set of circumstances. The precise definition of this
set depends on the GPist. For example, Harris (1997) proposes that
only 'licensed' positions are sites of element deletion; Kaye (p.c.
and 1993), on the other hand, prefers element addition to loss and
therefore proposes derivational H-addition to onsets preceding
domain-final silent empty nuclei instead of (Brockhaus') L-loss to
account for final obstruent devoicing. Relevant here is that
according to GP, one of the purposes of phonology is to provide
parsing cues to the hearer, i.e. cues that help the hearer parse
the continuous input string into distinct cognitive units. There is
no reason why this help should have to be 'perfect'. As long as the
phonology provides the hearer with enough information to find a
match in the lexicon for a (derived) form within a certain amount
of processing time, there is no problem. To establish such temporal
processing parameters (defining cognitive 'speed' in natural
language processing) is one of the aims of certain GPists. In other
words, as long as there is no reason to assume that the human
parsing device (i.e. phonology) is (unlike other human properties)
100% perfect and non-redundant, Coleman's rejection of
transformational models, particularly of GP, remains somewhat
unfounded, a problem that Coleman does unfortunately not discuss.
(For further details on the computational properties of GP and a
critique of Coleman 1995, cf. Williams 1998, particularly chapter
7.) Note that Coleman does also not investigate in any details the
issues raised in Barton, Berwick & Ristad ('BBR') (1987) which he
only refers to in passing. Since BBR can show that strong linear
precedence constraints, binary constituent structure and
terminal-string distinguishability result in efficient parsability
- cf. also Williams 1998:181f. - it can be said that Coleman has no
argument for his claim that only a non-transformational approach is
restrictive enough for the cognitive manipulation of natural
language. Unfortunately, Coleman only constructs a very weak
transformational opponent of DCP so that DCP's success is more
limited than Coleman lets on.
In the following, I will provide four more arguments in favour of
GP, a transformational model, and against DCP. In section 1, I will
point to some of the properties of GP which make it a theory that
is far more restrictive than the transformational theories Coleman
argues against (mainly, SPE, AP and metrical phonology).
Furthermore, it is possible to eliminate one of GP's two levels of
representation. This destroys the main advantage DCP had over the
two-levelled standard version of GP (section 2). In addition, DCP's
declarations are more explanatorily arbitrary and inadequate than
GP's constraints (even compared to a two-levelled version of GP;
section 3). Finally, in section 4, I will explain why DCP
self-destructs by virtue of providing unacquirable analyses.
1. GP's restrictiveness
One of the problems of transformational theories is, says Coleman,
that they allow deletion and substitution: It is surprising that he
does not discuss GP's Empty Category Principle ('ECP') which
eliminates any phonological deletion or substitution in cases of
vowel-zero alternation (cf. e.g. Kaye, Lowenstamm & Vergnaud
('KLV') 1990). All Coleman says about the ECP (in Coleman 1995:344)
is that it is a constraint/configuration and that it "is a matter
of speculation ... whether these principles [including the ECP]
combine to define the phonology of even a single language''. He
does not discuss in what way or to what extent the numerous PhD
theses and phonologies of languages written in GP-type frameworks
do or do not succeed in such a 'definition'. His only other
statement regarding GP that we find in the same paragraph (or
elsewhere, for that matter) is that GP employs "systematic phonetic
representations", which is simply wrong. The same goes for GP's
Projection Principle (KLV 1990): Coleman only mentions but does not
discuss the Projection Principle. Importantly, this principle
disallows any phonological (i.e. derived) resyllabification of
lexical prosodic structure, which makes GP far more restrictive
than other derivational models. Furthermore, since GP proposes
formal asymmetrical licensing constraints (e.g. government) -
universally holding between certain prosodically defined positions
- in combination with a number of universal and only to some extent
language-specific substantive constraints on what melodic units may
be linked to such formally restricted positions -, GP is
metatheoretically more restrictive than DCP which simply declares
what it finds but contains comparatively very little explanatory
power regarding phonotactic restrictions (cf. also below).
In addition, GP is more restrictive than classical transformational
approaches with rule ordering and a quite unlimited number of
levels available because it only assumes two levels of
representation, one lexical and one derivational level. In GP,
phonology is viewed as a function 'phi' that is applied to a
lexical input representation (i.e. the addresses in the hearer's
lexicon) deriving the phonological representation. Since there is
no rule ordering, all phonological processes happen simultaneously,
resulting in a situation in which any order of application of a
given set of processes have the same result. This is expressed by
Kaye's impressively simple "minimalist hypothesis" (Kaye 1995:291)
regarding derivations: "Processes apply whenever the conditions
that trigger them are satisfied" (ibid.). Self-evidently, this
approach is more restrictive than multi-levelled transformational
theories (i.e. those with rule ordering, e.g. Bromberger & Halle
1989). So Coleman has to pick relatively weak opponents in order to
make his case. (This is not meant to offend 'weak' proponents of
rule ordering; GP's restrictiveness would not have been possible
without earlier SPE- and AP-based research.)
GP's restrictiveness is also evident in the number of smallest
melodic unit it employs: The most restricted version of element
theory only stipulates six monovalent elements (A, I, U, H, L,
Glottal, cf. Ploch 1999, chapter 4, for a summary) instead of
Coleman's 10+ (at least) binary features. For a reply to Coleman's
claim that element theory (in its earliest version as charm theory
with 11 elements) and by implication the even more restricted
modern versions are too restricted, cf. Kaye (1995). Note also that
Coleman's claim is based on his opinion that in GP as a
derivational model "systematic phonetic representations are derived
from the lexical representations after [derivations]'' (Coleman
1995:344). Since GP does not agree with any phonologically relevant
level of 'systematic phonetic representations' Coleman may well
disagree with GP's non-interest (not in phonetics but) in phonetic
levels of phonological representations, but it makes no sense to
accuse GP of being highly unrestrictive in relation to
phonologically relevant phonetic representations which it neither
proposes nor predicts to exist. Other published statements about GP
on the part of Coleman (i.e. in Coleman 1990, 1995) are typically
not backed up by evidence or even a few tentative arguments. Kaye
1995) has already pointed out that Coleman's "version of GP bears
little resemblance to anything that has been proposed" (Kaye
1995:325; cf. ibid.:324-328 for further arguments). It appears that
in order for Coleman's claim that transformational approaches are
too unrestrictive to work it has to be based on frameworks that are
less restrictive than GP and on misinterpretations of GP.
Finally, let me mention GP's Uniformity Condition (Kaye 1995:292)
which stipulates that "Phonological representations are directly
interpretable at every level", i.e. there is no need for any
phonetic level of phonological representation or redundancy rules
filling in missing feature specifications - be it within classical
pre-underspecification theories with redundancy rules with
completely specified matrices in surface representations or,
empirically worse since more unrestricted, within
underspecification, i.e. with lexical AND surface
underspecification. (For a more detailed discussion of redundancy
rules and underspecification, cf. chapters 1 and 2 in Ploch 1999.)
2. Lexical GP
In opposition to Kaye's (1995) two-levelled derivational version of
GP (but in agreement with his interfaces), GP could be turned into
a theory with only one (lexical) level without loss of explanatory
power but with additional restrictiveness, relative economy (of the
Occam's Razor type) and the main advantage of DCP, i.e.
computational straightforwardness. Each and every phonological
process in standard GP can be reformulated as lexical constraint.
For example, I-spreading (i.e. the derivational process in Turkish
that assigns each nucleus in a word (domain) an I-element iff the
nucleus left-adjacent at the nuclear projection contains I) can be
reformulated as a non-derivational lexical constraint: "In Turkish
(lexical) addresses, a nucleus that immediately follows a nucleus
dominating I at the nuclear projection contains I." Again, Coleman
does not try to construct a worthy opponent for DCP, and it is
because of this metatheoretical negligence that his claim that DCP
is more restrictive than transformational theories does not amount
to much: In order for Coleman's argumentation against the
unrestrictiveness of transformational frameworks to be competent,
he would have to include a argumentation IN FAVOUR OF the most
restricted transformational (GP-like and, possible, GP-unlike)
framework he could formulate and a comparative analysis between
this opponent and DCP. If DCP THEN still worked better than the
opponent, he would have an argument (barring other problems). (An
excellent example of such a metatheoretically sound approach can be
found in Popper's argumentation FOR and Deconstruction of certain
historicist doctrines in Popper 1961; for a phonological attempt at
this kind of deconstruction of a competing hypothesis, cf. my
discussion of the hypothesis that the phonology is motivated by the
phonetics, in part 1 of my thesis, Ploch 1999.)
3. DCP's arbitrariness and unrestrictiveness
DCP's rules are highly arbitrary. For example 'O[-back] --> /l/'
(cf. CLM:178); this rules excludes dark l from onset positions. It
is not difficult to see that this rule explains nothing because it
is only descriptive. Would DCP universally exclude a rule like
'O[+back] --> /l/' in a language with clear and dark 'l'? It is
also very unclear in what way the categories (or any other
proposal) Coleman employs explain the phonological behaviour of
such categories. For example, in an implicational rule such as
'[+nasal] logically implies [+voice]' - which is logically
equivalent to the declarative '-([+nasal] & -[+voice])' and
expresses the traditional redundancy rule '[+nasal] --> [+
voice])', the co-occurrence of [+nasal] and [+voice] are simply
stipulated (CLM:190f.). It contains no explanation of the fact that
there are many logically possible combinations of features employed
by Coleman that could be used in this kind of implicational rule
which are not attested, e.g. there is no
'-([+nasal] & -[+strident])' (285ff.) Declarative rules merely
describe distributional observations but are not set up to explain
and thus cannot explain any of these properties. Another example
can be found in the rule 'O(nset) --> /h/' (CLM:294); this rule for
English states that /h/ (like affricates (/tsh/, /dzh/)) must occur
as onset by itself and can therefore not occur after /s/ (like
other obstruents can); so while /st/ is possible, /s-h/ is not.
However, Coleman's rule has no explanatory power: one could ask why
there is no rule 'O --> /m/'? Coleman can of course answer this by
pointing out that such a rule would be proven wrong by
/sm/-clusters. This does not solve the problem: So why is there no
/s-h/-cluster in any of the languages with sC-clusters? Why can /m/
but not /h/ be preceded by /s/? Coleman observes (which is not in
itself objective, cf. Popper 1972) and states what he sees.
Declarative rule are of the 'I see ...' type, have no explanatory
power, are arbitrary and in this way more unrestrictive than rules
in theories like AP (let alone element theory).
In addition, Coleman needs filters that must be placed on certain
rules. For example, the rule 'O --> Cl[-nasal] Glide' declares that
"Only closure which does not contain a nasal ... can be combined
with a glide'' (CLM:294), prohibiting e.g. */mr/, */snl/. Firstly,
the unrestrictiveness and high degree of arbitrariness of DCP
reveals itself in Coleman's 'filters' which must be placed on this
rule (CLM:294f.). It might be an interesting topic to investigate
whether a phonological analysis has to cater for negative
generalisations like Coleman's filter #2 '*labial-labial' because
it is not at all self-evident that a child acquiring English
actually cognitively tests all combinations predicted to exist by
tentative hypotheses based on attested data (without the addition
of filters). So does the phonology of an English speaker contain
such a preclusion or are such labial-labial sequences grammatical
but simply unattested? (Note that Jensen 1999 links our varied
'feelings' of ungrammaticality towards certain combinations not
with negative filters but with a theory of positive attestation,
i.e. where the set of attested forms in a speaker's grammar define
an average phonological form; the further away from this average a
new form, e.g. French <foie> for a native English speaker, is the
more ungrammatical it 'feels'.) However, Coleman's framework also
allows him to exclude specific sequences, e.g. filter #1 '*/stw/,
*/skl/'. It remains open why there is no filter #6 '*spr, *str'
preventing sequences which do occur. Apparently, Coleman's approach
is descriptive only, it makes no predictions and has thus zero
explanatory power. We see that his claim that, opposed to all these
sadly unrestricted transformational context-dependent rewrite
rule-based approaches, his declarative rules have the great
advantage that they are highly restrictive is
(metatheoretically/explanatorily but not computationally) mistaken.
Even though GP is more restrictive than other transformational
approaches, the only time Coleman mentions GP in CLM other than in
his reference (CLM:24) to his misinterpretative argumentation
against charm theory (Coleman 1990), is when he discusses his
filter #5 on rule */sr/, */shl/, */shw/ (CLM:295) which contains at
least some generalisation in that it can be viewed as a
dissimilatory constraint preventing [+high][+high] and
[-high][-high] sequences. His argument against GP (KLV 1990) is
that the dissimilatory nature of filter #5 provides evidence
against the spreading-based (and thus in his view presumably
dissimilation-phobic) nature of AP (Goldsmith 1976) and GP.
However, dissimilation in GP is not precluded simply because
elements can spread in standard GP with two levels or are subject
to (non-phonetically motivated) 'assimilatory' co-occurrence
constraints holding at the one existing (lexical) level in a
one-levelled version of GP. (Let me use 'spreading' as somewhat
misleading term for one- or two-levelled approaches to element
co-occurrence.)
For example, I-spreading (palatal harmony) in Turkic languages
(Charette & G�ksel 1994, 1996, Ploch 1996, 1998, 1999) is a
derivational constraint that forces nuclei following the harmonic
head (i.e. the first nucleus of a domain) to contain an I-element
iff the harmonic head contains I. This does not per se preclude
dissimilatory constraints in the same framework. A constraint
preventing geminate consonants in English would in GP typically be
formulated as dissimilatory constraint on melodic expressions
precluding a postnuclear rhymal position ('coda') and a following
onset position from dominating the same melodic expression. In part
2 of Ploch (1999), I have proposed dissimilatory constraints on
elements (L(ow tone) and H(igh tone)) to explain Dahl's Law in a
number of Bantu languages and in certain South German varieties of
German (specifically Augsburg Swabian) and Meinhof's Law (also in
Bantu languages). (I neglect here that the dissimilatory
constraints on L and H were more specific than element-specific. In
element theory, elements within a phonological expression may be of
one of two status: head or operator. The dissimilatory constraints
proposed refer specifically to the ungrammaticality of the
co-occurrence of H- or L-HEADS.) So Coleman simply circumvents any
discussion of the restrictive character of transformational or
'declarative' GP by constructing an unfounded weak version of GP
that no GPist employs. Therefore, I find that Coleman does not try
to let his DCP compete with competent but weak transformational
theories. As Kaye has already pointed out in 1995, since Coleman
(1995), Coleman has made "much of the alleged derivational nature
of GP although it is not clear what exactly [or even in vague
terms, for that matter! SP] would change if some of the procedural
language in some GP formulations were replaced by declarative
ones'' (Kaye 1995:325) (cf. also Kaye's further discussion in Kaye
1995:325ff.).
4. DCP's unacquirability problem
As already apparent in Coleman (1995), CLM presents a theory which
ensures that analyses put forward within it are not acquirable. The
argument for this claim goes as follows: In order to arrive at a
certain (synchronic) constraint relating to a string X in a
so-called variety or dialect of English it is permissible for
Coleman (and apparently even a great explanatory advantage) to use
analyses of phonological correspondences of string X in other
'varieties/dialects' as evidence. For example, since Orton &
Dieth's (1962) survey of English dialects has [tl] "in all
test-words beginning with /kl/ and [dl] in ... the /gl-/ test-words
for some speakers'' (CLM:295) in certain (non-standard) 'dialects'
of English, Coleman views the standard restriction excluding */tl/,
*/dl/, */thl/ as neutralisation of alveolar AND velar place of
articulation in stop+/l/ clusters. Similarly, the phonological
category of the u-vowel in English <took> "should be regarded as
distinctively 'long' in non-transformational phonology, even though
its phonetic interpretation is relatively short" (CLM:172) and in
fact indistinguishable from the (phonetic) vowel [u] in put (which
would be short phonologically AND phonetically). Coleman's
'evidence' for this claim is that the Great Vowel Shift did affect
the vowel in <took> (because it was long) but not the one in <put>
(because it was short). So synchronically short [u] can for Coleman
correspond to phonologically long or short /u/. Even though I have
no problem with non-one-to-one mappings between the phonology and
the phonetics, I disagree with Coleman's argument for
distinguishing short /u/ in <took> from (phonetically identical)
short /u/ in /put/ because there is no way how an acquirer of a
'variety' of English with identical realisations for these two
vowels could find out that the vowel in /took/ was long once. Also
so-called evidence from other 'dialects', e.g. the English spoken
in Newcastle, where the vowels in /book/ might still be long can be
no evidence since a child in London cannot be expected to travel to
Newcastle to make sure that not one or more of his/her vowels
correspond to two or more vowels in some other 'variety' of
English. We see that the incorporation of such 'dialectal' or
'historical' evidence results in a framework that provided
unacquirable analyses. In like manner, Coleman (1995:376) motivates
declarative statements by alternations like <drink> vs. <drench>,
<church> vs. <kirk>, <bridge> vs. <brig>. It remains open what
about <drink> and <drench> would tell the child acquiring English
that s/he is supposed to view these forms as 'related'. (Cf. my
discussion of the French forms like <fin> 'nice (masculine)' vs.
<fine> 'nice (feminine)' in Ploch 1999:141ff., 146ff.; there is
just as little evidence to derive <fin> phonologically from <fine>
or vice versa or to derive them from a common source (even though
these forms are historically linked) as there is to derive <fin->
in <finlande> 'Finland' from any of these forms.)
Summary: Coleman's claim that DCP is as opposed to transformational
context-sensitive approaches more restrictive is computationally
sound. However, his conclusion that 'therefore' all
transformational theories are too unrestrictive for the processing
of natural language is baseless. This situation stems from 1) his
refusal to construct a strong competing transformational theory
next to DCP, 2) his refusal to deal with GP, the most restrictive
transformational framework, 3) his omission to discuss whether or
not and if, in what way a declarative version of certain
transformational frameworks (e.g. GP) would get access to some of
DCP's 'advancements', and 4) the high degree of arbitrariness
apparent in DCP declarative and non-explanatory descriptions/rules.
So even though DCP's formalism is to some extent more restrictive
than SPE's AP's or even (two-levelled) GP's formalism, there is no
evidence to suggest that DCP would beat all or even most
transformational theories as theories of natural language
processing. In addition, Coleman proposes analyses that are
unacquirable. It also remains somewhat unclear in what way a strong
competitor of DCP would gain or loose by adopting his (interesting)
thoughts on the differentiation between notation and denotation.
There are, however, three aspects of Coleman's book that I find
thought-provoking or noteworthy: 1) His overview over the
historical development of notations and denotations in phonemic and
transformational phonology is good teaching material. 2) Reading
his book inspired me to investigate how GP or whatever theory I
might want to employ could be formulated declaratively or
non-procedurally and in more precise mathematical terms, and what
could be gained from taking this step. 3) Coleman's argumentation
against the unrestrictiveness of and the invertibility problem in
transformational approaches made the importance of such
computational concerns for phonology more apparent to me.
Abbreviations
AP autosegmental phonology
BBR Barton, Berwick & Ristad (1987)
CLM Coleman (1998), Phonological Representations, i.e. the book
reviewed here.
DCP declarative phonology
ECP Empty Category Principle
GP government-phonology
KLV Kaye, Lowenstamm & Vergnaud
PSG (classical) phrase-structure grammar
UPSG unification-based phrase-structure grammar
E) Bibliography
Barton, G.E., Berwick, R.C. & Ristad, E.S. (1987), Computational
Complexity and Natural Language, MIT Press, Cambridge (MA).
Brockhaus, W. (1995), Final Devoicing in the Phonology of German,
Niemeyer, T�bingen (Germany).
Bromberger, S. & Halle, M. (1989), "Why phonology is different",
Linguistic Inquiry, 20, pp. 51-71.
Charette, M. & G�ksel, A. (1994), "Switching and vowel harmony in
Turkic languages", Cobb, M. & Jensen, S. (eds.), SOAS Working
Papers in Linguistics and Phonetics. Volume 4, School of Oriental
and African Studies (University of London), London, pp. 31-52.
Charette, M. & G�ksel, A. (1996), "Licensing constraints and vowel
harmony in Turkic languages", Ploch & Simpson, pp. 1-25. Also
published in: Kardela, H. & Szymanek, B. (eds.), A Festschrift for
Edmund Gussmann, The University Press of the Catholic University of
Lublin, Lublin (Poland), pp. 29-56.
Chomsky, N. (1957), Syntactic Structures, Mouton, The Hague
(Holland).
Coleman, J.S. (1990), "Charm theory defines strange vowel sets",
Journal of Linguistics, 26, pp. 165-174.
Coleman, J.S. (1995), "Declarative lexical phonology", Durand &
Katamba, pp. 333-382.
Dresher, B.E. & Kaye, J.D. (1990), "A computational learning model
for metrical phonology", Cognition, 34, pp. 137-195.
Durand, J. & Katamba, F. (eds.) (1995), Frontiers of Phonology:
Atoms, Structures, Derivations, Longman, London & New York.
Goldsmith, J.A. (1976), Autosegmental Phonology, Indiana University
Linguistics Club, Bloomington.
Harris, J. (1997), "Licensing inheritance: an integrated theory of
neuralisation", Phonology, 14, pp. 315-370.
Henderson, E.J.A. (1971), The Indispensable Foundation: a Selection
from the Writings of Henry Sweet, Oxford University Press, London.
Jensen, S. (1999), "Meta-phonological speculations", unpublished
manuscript.
Kaye, J.D. (1993), "Current issues in phonology", lectures held at
the School of Oriental and African Studies (University of London).
Kaye, J.D. (1995), "Derivations and interfaces", Durand & Katamba,
pp. 289-332.
Kaye, J.D., Lowenstamm, J. & Vergnaud, J.-R. (1985), "The internal
structure of phonological elements: a theory of charm and
government", Phonology Yearbook, 2, pp. 305-328.
Kaye, J.D., Lowenstamm, J. & Vergnaud, J.-R. (1990), "Constituent
structure and government in phonology", Phonology, 7.2, pp.
193-231.
Orton, H. & Dieth, E. (1962), Survey of English Dialects. (B) The
Basic Material, vol. 1: The Six Northern Counties and the Isle of
Man (3 parts), E.J. Arnold, Leeds.
Ploch, S. (1996), "The role of parsing", in Ploch & Simpson, pp.
76-105.
Ploch, S. (1998), "Non-switch harmony in Yawelmani (and Turkish and
Sakha)", Ploch, S. & Scott, G.-J. (eds.), SOAS Working Papers in
Linguistics and Phonetics. Volume 8, School of Oriental and African
Studies (University of London), London, pp. 209-238.
Ploch, S. (1999), Nasals on My Mind. The Phonetic and the Cognitive
Approach to the Phonology of Nasality, PhD thesis, School of
Oriental and African Studies (University of London), London.
Ploch, S. & Simpson, A. (eds.) (1996), SOAS Working Papers in
Linguistics and Phonetics. Volume 6, School of Oriental and African
Studies (University of London), London.
Popper, K.R. (1961), The Poverty of Historicism, Routledge & Kegan
Paul, London. First published 1957.
Popper, K.R. (1972), Objective Knowledge: an Evolutionary Approach,
Oxford at the Clarendon Press, Oxford. Reprinted (with corrections)
1973.
Scobbie, J.M. (1993), "Issues in constraint violation and
conflict", Ellison, T.M. & Scobbie, J.M. (eds.), Computational
Phonology, Edinburgh Working Papers in Cognitive Science, 8,
University of Edinburgh Centre for Cognitive Science, pp. 37-54.
Scobbie, J.M., Coleman, J.S. & Bird, S. (1996), "Key aspects of
Declarative Phonology", Durand, J. & Laks, B. (eds.), Current
Trends in Phonology. Models and Methods, vol. 2, University of
Salford Publications, CNRS, Paris X & University of Salford,
Salford (UK), pp. 685-709.
Williams, G. (1998), The Phonological Basis of Speech Recognition,
PhD thesis, School of Oriental and African Studies (University of
London), London.
F) Reviewer's biography
Stefan Ploch, School of Oriental and African Studies (University of
London).
PhD 1999, School of Oriental and African Studies (University of
London).
Main research areas include: phonology, parsing, licensing,
derivations, nasality and nasalisation, phonetics versus phonology,
Turkic languages, South American languages, philosophy of science.