6. The Phonological Basis of Sound Change.pdf

6. The Phonological Basis of Sound Change : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:32 PM

6. The Phonological Basis of Sound Change

PAUL KIPARSKY

Subject

Linguistics » Historical Linguistics

DOI:

10.1111/b.9781405127479.2004.00008.x

Tout est psychologique dans la linguistique, y compris ce qui est mécanique et matériel .

F. de Saussure 1910/1911

[…] The Neogrammarians portrayed sound change as an exceptionless, phonetically conditioned process

rooted in the mechanism of speech production. 1 This doctrine has been criticized in two mutually

incompatible ways. From one side, it has been branded a mere terminological stipulation without empirical

consequences, on the grounds that apparent exceptions can always be arbitrarily assigned to the categories

of analogy or borrowing. 2 More often though, the Neogrammarian doctrine has been considered false on

empirical grounds. The former criticism is not hard to answer (Kiparsky 1988), but the second is backed by

a formidable body of evidence. Here I will try to formulate an account of sound change making use of ideas

from lexical phonology, which accounts for this evidence in a way that is consistent with the Neogrammarian

position, if not exactly in its original formulation, then at least in its spirit.

The existence of an important class of exceptionless sound changes grounded in natural articulatory

processes is not in doubt, of course. It is the claim that it is the only kind of sound change that is under

question, and the evidence that tells against is primarily of two types. The first is that phonological

processes sometimes spread through the lexicon of a language from a core environment by generalization

along one or more phonological parameters, often lexical item by lexical item. Although the final outcome of

such lexical diffusion is in principle

[By permission of the author and the publisher, this chapter, originally published in John

Goldsmith (ed.) Handbook of Phonological Theory (Blackwell, 1995), is reprinted here with

minor changes; the author felt that this piece constituted as definitive a statement of his views

on sound change as there could be, so that reprinting it here was deemed appropriate by all

concerned.]

indistinguishable from that of Neogrammarian sound change, in mid-course it presents a very different

picture. Moreover, when interrupted, reversed, or competing with other changes, even its outcome can be

different.

Against the implicit assumptions of much of the recent literature, but in harmony with older works such as

Schuchardt (1885) and Parodi (1923: 56), I will argue that lexical diffusion is not an exceptional type of

sound change, nor a new, fourth type of linguistic change, but a well-behaved type of analogical change.

Specifically, lexical diffusion is the analogical generalization of lexical phonological rules . In the early

articles by Wang and his collaborators, it was seen as a process of phonemic redistribution spreading

randomly through the vocabulary (Chen and Wang 1975; Cheng and Wang 1977). Subsequent studies of

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274798

Page 1 of 22

6. The Phonological Basis of Sound Change : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:32 PM

randomly through the vocabulary (Chen and Wang 1975; Cheng and Wang 1977). Subsequent studies of

lexical diffusion have supported a more constrained view of the process. They have typically shown a

systematic pattern of generalization from a categorical or near-categorical core through extension to new

phonological contexts, which are then implemented in the vocabulary on a word-by-word basis. In section 1

I argue that lexical diffusion is driven by the rules of the lexical phonology, and that the mechanism is

analogical in just the sense in which, for example, the regularization of kine to cows is analogical. In fact,

the instances of “lexical diffusion” which Wang and his collaborators originally cited in support of their

theory include at least one uncontroversial instance of analogical change, namely, the spread of retracted

accent in deverbal nouns of the type tórmènt (from tormént ). In most cases, of course, the analogical

character of the change is less obvious because the analogy is non-proportional and implements

distributional phonological regularities rather than morphological alternations. For example, the item-by-

item and dialectally varying accent retraction in non-derived nouns like mustache, garage, massage,

cocaine is an instance of non-proportional analogy, in the sense that it extends a regular stress pattern of

English to new lexical items. What I contend is that genuine instances of “lexical diffusion” (those which are

not due to other mechanisms such as dialect mixture) are all the result of analogical change. To work out

this idea I will invoke some tools from recent phonological theory. In particular, radical underspecification

and structure-building rules as postulated in lexical phonology will turn out to be an essential part of the

story.

The second major challenge to the Neogrammarian hypothesis is subtler, less often addressed, but more

far-reaching in its consequences. It is the question how the putatively autonomous, mechanical nature of

sound change can be reconciled with the systematicity of synchronic phonological structure. At the very

origins of structural phonology lies the following puzzle: if sound changes originate through gradual

articulatory shifts which operate blindly without regard for the linguistic system, as the Neogrammarians

claimed, why don't their combined effects over millennia yield enormous phonological inventories, which

resist any coherent analysis? Moreover, why does no sound change ever operate in such a way as to subvert

phonological principles, such as implicational universals and constraints on phonological systems? For

example, every known language has obstruent stops in its phonological inventory, at least some unmarked

ones such as p, t, k . If sound change were truly blind, then the operation of context-free spirantization

processes such as Grimm's law to languages with minimal stop inventories should result in phonological

systems which lack those stops, but such systems are unattested.

With every elaboration of phonological theory, these difficulties with the Neogrammarian doctrine become

more acute. Structural investigations of historical phonology have compounded the problems. At least since

Jakobson (1929), evidence has been accumulating that sound change itself , even the exceptionless kind, is

structure-dependent in an essential way. Sequences of changes can conspire over long periods, for example

to establish and maintain patterns of syllable structure, and to regulate the distribution of features over

certain domains. In addition to such top-down effects, recent studies of the typology of natural processes

have revealed pervasive structural conditioning of a type hitherto overlooked. In particular, notions like

underspecification, and the abstract status of feature specifications as distinctive, redundant, or default, are

as important in historical phonology as they are synchronically. The Neogrammarian reduction of sound

change to articulatory shifts in speech production conflicts with the apparent structure-dependence of the

very processes whose exceptionlessness it is designed to explain.

A solution to this contradiction can be found within a two-stage theory of sound change according to which

the phonetic variation inherent in speech, which is blind in the Neogrammarian sense, is selectively

integrated into the linguistic system and passed on to successive generations of speakers through language

acquisition (Kiparsky 1988). This model makes sound change simultaneously mechanical on one level

(vindicating a version of the Neogrammarian position), yet structure-dependent on another (vindicating

Jakobson). The seemingly incompatible properties of sound change follow from its dual nature.

My paper is organized as follows. In the next section I present my argument that lexical diffusion is

analogical and that its properties can be explained on the basis of underspecification in the framework of

lexical phonology. I then spell out an account of sound change which reconciles exceptionlessness with

structure-dependence (section 2). Finally in section 3 I examine assimilatory sound changes and vowel shifts

from this point of view, arguing that they too combine structure-dependence with exceptionlessness in ways

which support the proposed model of sound change, as well as constituting additional diachronic evidence

for radical underspecification in phonological representations.

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274798

Page 2 of 22

6. The Phonological Basis of Sound Change : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:32 PM

1 Lexical Diffusion

1.1 “It walks like analogy, it talks like analogy …”

If lexical diffusion is not sound change, could it be treated as a subtype of one of the other two basic

categories of change? Clearly it is quite unlike lexical borrowing: it requires no contact with another

language or dialect (i.e., it is not reducible to “dialect mixture”), it follows a systemic direction set by the

language's own phonological system (it is a species of “drift”), and it involves a change in the pronunciation

of existing words rather than the introduction of new ones.

Table [6].1

Sound

change

Borrowing Lexical analogy

Lexical diffusion

Generality

Across the

board

Item by

item

Context by context, item by

item

Context by context, item by

item

Gradience

Gradient

Quantal Quantal

Quantal

Origin

Endogenous Contact Endogenous

Endogenous

Rate

Rapid

Slow

Effect on:

Rule system

New rules

No change Rules generalized

Rules generalized

Sound /phoneme

inventory

New inventory Peripheral No change

No change

Vocabulary

No change

New

words

No change

On the other hand, it does behave like lexical analogy in every respect, as summarized in [ table 6.1 ]. 3

It seems to be the case that lexical diffusion always involves neutralization rules, or equivalently that lexical

diffusion is structure preserving (Kiparsky 1980: 412). This has been taken as evidence for locating lexical

diffusion in the lexical component of the phonology (Kiparsky 1988). Being a redistribution of phonemes

among lexical items, it cannot produce any new sounds or alter the system of phonological contrasts. Its

non-gradient character follows from this assumption as well, since lexical rules must operate with discrete

categorical specifications of features.

An important clue to the identity of the process is its driftlike spread through the lexicon, by which it

extends a phonological process context by context, and within each new context item by item. This is of

course exactly the behavior we find in many analogical changes. An example of such lexical diffusion is the

shortening of English /ū/, which was extended from its core environment (1a), where it was categorical, by

relaxing its context both on the left and on the right (Dickerson 1975). In its extended environments it

applies in a lexically idiosyncratic manner. The essential pattern is as follows:

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274798

Page 3 of 22

6. The Phonological Basis of Sound Change : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:32 PM

We can provide a theoretical home for such a mechanism of change if we adopt lexical phonology and

combine it with a conception of analogical change as an optimization process which eliminates idiosyncratic

complexity from the system –- in effect, as grammar simplification. 4 The mechanism that drives such

redistribution of phonemes in the lexicon is the system of structure-building rules in the lexical phonology.

The direction of the phonemic replacement is determined by the rule, and its actuation is triggered jointly by

the generalization of the rule to new contexts, and by the item-by-item simplification of lexical

representations in each context. When idiosyncratic feature specifications are eliminated from lexical entries,

the features automatically default to the values assigned by the rule system, just as when the special form

kine is lost from the lexicon the plural of cow automatically defaults to cows . The fact that in the lexical

diffusion case there is no morphological proportion for the analogy need not cause concern, for we must

recognize many other kinds of non-proportional analogy anyway.

To spell this out, we will need to look at how unspecified lexical representations combine with structure-

building rules to account for distributional regularities in the lexicon. This is the topic of the next section.

1.2 The idea behind underspecification

The idea of underspecification is a corollary of the Jakobsonian view of distinctive features as the real

ultimate components of speech. All versions of autosegmental phonology adopt it in the form of an

assumption that a feature can only be associated with a specific class of segments designated as permissible

bearers of it (P-bearing elements), and that such segments may be lexically unassociated with P and acquire

an association to P in the course of the phonological derivation. But in phonological discussions the term

“underspecification” has come to be associated with two further claims, mostly associated with lexical

phonology, namely that the class of P-bearing segments may be extended in the course of derivation, and

that lexical (underlying) representations are minimally specified.

How minimal is minimal? There are several alternative versions of underspecification on the market which

differ in their answers to this question. 5 The most conservative position, restricted underspecification , is

simply that redundant features are lexically unspecified. On this view, the feature of voicing in English would

be specified for obstruents, where it is contrastive, but not for sonorants, which are redundantly voiced. An

entirely non-distinctive feature, such as aspiration in English, would not be specified in lexical

representation at all.

Radical underspecification (the version which I will assume later on) carries the asymmetry of feature

specifications one step further, by allowing only one value to be specified underlyingly in any given context

in lexical representations, namely, the negation of the value assigned in that context by the system of lexical

rules. A feature is only specified in a lexical entry if that is necessary to defeat a rule which would assign the

“wrong” value to it. The default values of a feature are assigned to segments not specified for it at a stage in

the derivation which may vary language-specifically within certain bounds.

A third position, departing even further from SPE , and currently under exploration in several quarters, holds

that the unmarked value is never introduced, so that features are in effect one-valued (privative).

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274798

Page 4 of 22

6. The Phonological Basis of Sound Change : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:32 PM

Contrastive and radical underspecification both posit redundancy rules such as:

(2) [+ sonorant] → [+ voiced]

Radical underspecifications in addition posits default rules , minimally a context-free rule for each feature

which assigns the unmarked value to it:

(3) [ ] → [-voiced]

The following chart summarizes the theoretical options, and exemplifies them with the values of the feature

[voiced] which they respectively stipulate for voiceless obstruents, voiced obstruents, and sonorants, at the

initial and final levels of representation:

As (4) shows, fully specified representations and privative representations are homogeneous throughout the

phonology. Contrastive underspecification and radical underspecification both make available two

representations , by allowing an underlying minimal structure to be augmented in the course of the

derivation.

Radical underspecification moreover assumes that default values are assigned by the entire system of

structure-building lexical rules. For example, in a language with a lexical rule of intervocalic voicing such as

(5), 6 the lexical marking of obstruents in intervocalic position would be the reverse of what it is in other

positions, with voiced consonants unmarked and voiceless ones carrying the feature specification [-voiced]

to block the rule:

(5) [ ] → [+voiced] / V___V

At what point are default values and redundant values to be assigned? I will here assume that default feature

values are filled in before the first rule that mentions a specific value of that feature. 7 Many assimilation

rules do not mention a specific feature value, but simply spread the feature itself, or a class node under

which that feature is lodged. Such rules can apply before the assignment of default values, yielding the

characteristic pattern “assimilate, else default.”

To summarize:

(6) a. For each feature F, a universal default rule of the form [ ] → [αF] applies in every language.

b. In each environment E in underlying representations, a feature must be either specified as [αF] or

unspecified, where E is defined by the most specific applicable rule R, and R assigns [-αF].

c. Default feature values are filled in before the first rule that mentions a specific value of that

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274798

Page 5 of 22

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: