LINGUIST List 13.48
Thu Jan 10 2002
Review: Bybee, Phonology and Use
Editor for this issue: Simin Karimi <siminlinguistlist.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
Simin Karimi at simin
linguistlist.org
or Terry Langendoen at terry
linguistlist.org.
Subscribe to Blackwell's LL+ at
http://www.linguistlistplus.com/ and
donate 20% of your subscription to LINGUIST! You get 30% off on
Blackwells books, and free shipping and postage!
Directory
Graham Horwood, Review, Bybee (2001)
Message 1: Review, Bybee (2001)
Date: Mon, 7 Jan 2002 13:23:21 -0500 (EST)
From: Graham Horwood <gvhrci.rutgers.edu>
Subject: Review, Bybee (2001)
Review of Bybee, Joan. 2001. Phonology and Language Use.
Cambridge Studies in Linguistics 94. Cambridge: Cambridge
University Press.
Reviewed by Graham Horwood, Rutgers University
OVERVIEW
After introducing the work with a brief overview of
generative phonology as the outgrowth of American
Structuralism, the author outlines the basic assumptions of
a model designed to overcome what is taken to be the
principle limitation of the generative approach: the fact
that it makes no provision for the effects of language use
on language structure. In the proposed, functionally
grounded model, lexical representations are taken to be
malleable and therefore subject to usage- and
frequency-based fluctuation and change over time.
Linguistic and non-linguistic objects share the same kind
of mental representation, and thus there is nothing
cognitively 'special' about language learning. Phonological
and morphological category is a function of structural
similarity, and grammatical principles/rules/constraints
are emergent from lexical representations themselves,
having no status independent of them. Thus phonology itself
is a procedural comprehension/production system, rather
than the abstract, psychological system argued for since
Chomsky & Halle (1968). These assumptions in place, the
author sets about exploring their foundations and
consequences in the chapters that follow.
The question of exactly how lexical items are stored
is an especially important one for a theory which eschews
generative grammar, and in Chapter Two, Bybee lays out the
basic representational system of her model. It is argued
that lexical representations are compilations of the
language user's experience. Thus an exemplar of a word
contains any number of actually occurring instances of
phonetic representation, and the lexicon, as in
connectionist theories, is taken to be a vast network of
phonetic, semantic, and structural co-occurance data in which
lexical items "overlap" (i.e., 'exaggerated' is already
partially stored as 'exaggerate'). Traditional phonological
units, such as the phoneme or syllable, emerge from the
interaction of phonetic principles and lexical coocurance
frequencies. Linguistic generalizations, formalized by
Bybee as _schemas_, are "discovered" as lexical items are
categorized for storage; they exist at any level of
generality--[$send$], [-end$], [-VNC$], [-C$], etc. are all
successively more general schemas of the word 'send'.
Complex morphological systems and broader phonological
regularities (i.e., rules/constraints/principles) are also
represented as highly embedded, overlapping structures.
Unlike symbolic rules, however, schemas are taken to a) have
no existence independent of the lexicon, b) be productive
as a gradient function of lexical frequency, and c) often be
particular to individual lexical items. The distinction
between a productive and an irregular schema is argued to be
a function of type and token frequency; thus past tense
/-ed/, for example, is 'regular' because it exists in the
larger majority of English verbal paradigms than do ablaut
alternations such as those in run~ran. Bybee argues that we
observe the properties of schemas in natural language
behavior and in psycholinguistic experimentation, and goes
on to support her model with the facts of such in the
chapters to follow.
Among the central tenets of generative phonology is
the understanding that predictable features and contrastive
ones receive a fundamentally different formal treatment:
contrastive properties of words are encoded in the lexicon
and predictable ones are generated by the formal grammar.
In Chapter Three, Bybee defends her stance that
'psychologically real' units of lexical organization crucial
to generative phonology (such as the phoneme) are in fact
emergent generalizations of the structured lexical network,
and that phonetic signals are recorded as perceived/uttered
in the lexicon with no cancellation of redundant material.
This position is supported with evidence from three broad
phenomenon types: word-specific phonetic variation; lexical
contrast shown by distributionally predictable features; and
morphologically variable sound change.
Chapters Four and Five address the question of how
phenomena traditionally explained by phonological rules are
accounted for in the system. The answer comes in the
understanding that schemas are mutable--due to variation
conditioned by phonetics or morphology--and are themselves
responsible for allophonic variation as they change over
time. Phonetic alternations, the matter of Chapter Four,
begin as phonetic tendencies which change high-frequency
words in the lexicon; these changes then generalize over
larger portions of the lexicon. Since speech is viewed as a
neuromotor activity in which words are stored as routines of
articulatory gestures, language-specific phonotactic and
phonetic patterns are simply often-repeated neuromotor
routines, and as such are subject to "tendencies of
compression and reduction". Traditional representational
units such as phoneme and syllable are not necessarily
atomic units of representation in the proposed system.
Bybee argues that, just because many vowels and consonants
emerge as discrete segmental units, we need not conclude
that all collocations of phonetic properties must do so, and
in fact that it is questionable whether all contextually
varying instances of a given consonant are members of the
same category.
Having treated the various gestural conditioning
factors for (purely phonetic) alternation, Bybee turns
in Chapter Five to alternations found in specific subdomains
of the lexicon, i.e., morphologized phonological processes.
In Bybee's lexical network model, variation in phonetic form
is tied to phonetic, semantic, lexical, and even social
context. Semantic associations are more critical than
phonological associations--we know this b/c 'go' and 'went'
are identified as part the same paradigm despite their
complete lack of phonological relatedness. Given the
primacy of semantic connection, then, it follows that
specification of phonetic schemas should be highly
restricted by semantic context, i.e., morphologized.
Morphologization cuts across semantic category, typically
in a lexically restricted manner. Thus a full phonetic
Middle English [f]~[v] alternation may occur in Modern
English nouns (wife~wives) as well as verbs (leaf~leaves),
but have lexical exceptions in either case (i.e.,
chief~chiefs (n.), brief~briefs (v.)). Bybee takes facts
such as these to be strong support for her theory, observing
that in standard generative models phonological alternations
are most typically accounted with a single rule or
constraint, and thus should be entirely regular across the
lexicon.
Chapters Six and Seven consider variously sized
units of storage and the psycholinguistic and diachronic
evidence for them. Bybee assumes that the atomic unit of
lexical storage is the word, rather than the morpheme.
Distinct instances of the same morpheme may become
diachronically 'frozen' to the lexical representations of
their respective context words, triggering what is referred
to in the generative paradigm as root-level allomorphy.
Since collocations of words are not stored together as a
single unit except in the event of high frequency, however,
alternations across word boundaries, i.e., word-level or
postlexical allomorphy, must be subject to the kind of
variation often resultant from highly regular phonetic
conditioning. This captures the fact that root-level
alternations tend to be both more common and more lexically
stable than word-level alternations. Bybee goes on to
argue, based on evidence from 'don't' reduction in English
and Liaison in French, that phonological alternations which
have traditionally been analyzed as word-level in actuality
generalize to high-frequency phrasal constructions.
Chapter Eight deals with one of the most obvious
challenges to Bybee's theory: its apparent inability to
capture linguistic universals that are not functionally
motivated. Human language is, in Bybee's model, essentially
a by-product of an intricate pattern-recognition system.
Independent of the conflicting drives to a) obey various
anatomical pressures which might condition a fairly uniform
set of phonetically-conditioned alternations and b) preserve
language-specific high-frequency words from reductive
phonetic drift, it obtains that *any* given pattern could be
captured by the system--Bybee consistently equates language
learning to the learning of other complex motor behaviours
such as playing the piano. Consequently, any linguistic
universal not functionally motivated would remain
unexplained. The author approaches this apparent problem by
challenging its underlying assumption: that linguistic
universals exist at all. If there are no linguistic
universals, there is no Universal Grammar, as such. Rather,
there exist certain "dynamic mechanisms" which cause
individual languages to change over time in patternable
ways. These mechanisms of change include: gestural
reduction; morphologization; lexical class formation;
morphological regularization; and the formation of high-
frequency sound sequences into constructions. If all spoken
languages share a property X, it is because all languages are
constrained by human physiology and a common set of
language-changing mechanisms which engender property X.
DISCUSSION
On the whole, the work is well-presented and well-reasoned.
It also provides an interesting alternative to generative
theories of phonology. A few minor criticisms do arise,
however. First--and on a relatively impressionistic
note--while high-level description of the system is
abundant, little is found in the way of precise formulation.
The model predicts that linguistic primitives and
lexicon-specific generalizations should arise from a complex
interaction of phonetic, semantic, and sociological factors
and both token and type frequency. However, the metric by
which each of these highly interconnected and opposing
factors is to be quantitatively measured remains somewhat
mysterious--perhaps some more direct implementation of the
model within a connectionist framework such as that of
Rumelhart & McClelland (1986) would give the reader a better
concept how such concepts as "strength of association" and
"lexical connection" are to be formalized independently of
the usage statistics which they describe. Similarly,
Bybee's arguments occasionally hinge on experimental results
which in many cases are subject to a considerable amount of
variation based on sociolinguistic factors. In her
discussion of Spanish coda reduction in Chapter Five, for
instance, Bybee admits that the phenomenon is subject to
variation based on such factors as speaker sex and age,
speaking rate, and register. In the face of such an array
of possible deviation from any controlled norm, it is
difficult to consider the strength of the results without
some suspicion.
On a more theoretical note, the author observes
that the proposed model is explanatorily adequate in the
Chomskian sense inasmuch as it identifies the factors
which contribute to the creation of synchronic grammars.
It is not immediately apparent, however, that the principles
of the proposed theory predict all *and only* the grammars
of natural languages, simply because of the psychological
generality the author consistently attributes to the system.
The author treats implicational universals and inventory
asymmetries with appeal to gestural reduction tendencies and
near-universal paths of diachronic change, but the avid
denial of any universal without functional grounding leaves
some cross-linguistically unattested phenomena within the
predictive scope of the theory. Taking an example from
prosodic morphology, for instance, consider the fact that
reduplicative "templates" can never be back-copied (e.g.,
Diyari /RED+tjilparku/ -> *[tjilpa-tjilpa<rku>], see
McCarthy & Prince 1999). The schemas upon which Bybee's
system is built are effectively templates of a highly
malleable and language-specific nature, and it is not
immediately apparent why the schema of a reduplicative
affix/clitic/word might not impose its (prosodically small)
schema on the (larger) reduplicative base, producing the
unattested form of copying. Similarly, at least one
universal I am aware of seems to run directly counter to
the implications of the author's model. It is widely
understood that roots and affixes differ in their
markedness properties cross-linguistically: the segmental
inventories of affixes are typically much reduced from
those of roots, and spreading processes such as vowel
harmony and assimilation typically target affixal material
before root material. This apparent universal runs counter
to the predictions of Bybee's model: since affixes
(particularly inflectional ones) are among the most
frequently used morphemes (or rather, schemas) in a
language, we would expect them to be highly resistant to
the types of reductive historical change which collapse
inventories and allow assimilatory variation.
In any event, these perceived shortcomings may well
fall to further elaboration of the model, and, overall, the
book is very well written and very clear in its goals,
arguments, and aims. The author outlines a complex,
sophisticated linguistic theory and provides considerable
evidence for each of her points, incorporating
psycholinguistic, computational, and diachronic findings
into a single, coherent model of lexical representation
presented in a terminology easily accessible to the
generative linguist. The model itself explains frequency
effects, lexical exception, and diachronic morphologization
in a coherent and fairly natural way, and as well makes a
number of interesting predictions about how and why
individual languages function and evolve as they do.
REFERENCES
- Chomsky, Noam and Morris Halle. 1968. The sound pattern
of English. New York: Harper and Row.
- McCarthy, John and Alan Prince. 1999. Faithfulness and
Identity in Prosodic Morphology. In R. Kager, H. van der
Hulst and W. Zonneveld, (eds.), The Prosody-Morphology
Interface, 2 8-309. Cambridge University Press, Cambridge.
[ROA-26.]
- Rumelhart, D. E. and J. L. McClelland. 1986. On learning
the past tenses of English verbs. Ch. 18, 216-271.
Cambridge, MA: MIT Press/Bradford Books.
- -------------------------------
Graham Horwood is a Ph.D. student at Rutgers University.
His current work explores processual morphology, issues in
morphological locality, and the implications of a
lexical-network based morpho-phonological architecture for
Optimality Theory.