ÁñÁ«ÊÓƵ

William Labov offers the first of three projected volumes on the nature of linguistic change; future volumes will deal with social and with cognitive factors. One longstanding controversy for which Labov attempts a new resolution is the question whether sound laws apply to phonemes or to words. Until recently English-speaking linguists accepted the Neogrammarian axiom that sound changes apply to every instance of a phonetic unit or grouping throughout the vocabulary of a language; but evidence from Chinese which became available in the 1970s gave fresh support to the contrary view, advocated by some French-speaking linguists, that changes diffuse gradually from word to word. Labov shows (though a later passage partly retracts the point) that the new data are best accounted for by Neogrammarian laws having the same stochastic character as the variable rules on which his sociolinguistic reputation is based: sound changes are conditioned by many phonetic factors, each of which affects the probability that the change will occur in a particular instance.

Another issue on which Labov brings new evidence to bear is the question whether language change is influenced by the need to maintain significant contrasts. Andre Martinet advocated the intuitively plausible idea that the robustness of a phonological distinction depends on how much useful work it does in keeping confusable words apart. Labov argues that Martinet was wrong. Language change stems not from speakers' need to make themselves understood, but from hearers' frequent failure to understand.

For Labov misunderstanding is central to the workings of language; using complex mathematical reasoning he argues that misunderstanding explains the puzzling phenomenon of drift, by which language change can operate in a constant direction over many generations.

Some commentators have found Labov's variable rules inherently implausible, wondering how a speaker could (or why he would) control his production of a sound by reference to running totals of alternative previous realisations. But Labov shows that such "probability matching" is widespread in animal behaviour; in a remark that deserves to become classic, he urges "We should not be embarrassed if we find that systemic readjustments in language are governed by the same cognitive faculty that governs the social behaviour of mallard ducks".

All this material, while not unproblematic, deserves careful consideration. But these sections of Labov's book are preceded by 400 pages which are less easy to admire. They are largely concerned with vowel shifts; Labov is particularly exercised by the question of how two vowels or diphthongs can exchange places in phonological space (as happened with the die and day diphthongs from Middle English to the present) without colliding and merging. His solution is that we must recognize a "peripheral" phonetic feature, distinguishing fully front or back vowels from centralized counterparts: this gave the die and day diphthongs room to slide past one another.

Labov has been such a doughty foe of aprioristic approaches in grammatical study that it is strange to find him taking a very different line in connexion with phonology: to describe phonological relationships in terms of features with discrete values is highly aprioristic. Labov knows, better than most linguists, that the physical realities underlying vowel quality are continuous parameters of tongue geometry and formant levels - vowel phonemes moving in opposite directions are like hikers entering a meadow from opposite sides, not like railway engines which must collide unless the track is double. Labov's distinctive features are constant through time but not space: he makes it quite explicit that they differ from one language-family to another, with language-specific switching points allowing phonemes to move from track to track.

When the New Zealand /e/ rose to the /i/ position, he argues, the /i/ phoneme was able to avoid merger by centralising, using an escape route available to the languages of one of August Schleicher's two fundamental divisions of Indo-European, namely the Slavo-Germanic branch that includes English, but not to the other "Aryo-Graeco-Italo-Celtic" branch, which is why so many distinct vowels and diphthongs of Ancient Greek have merged as Modern Greek /i/. But Labov does not pretend to explain what mechanism could maintain such constraints in particular language families over millennia, saying airily that this "remains an open question".

Futhermore, Labov's insistence on describing phonological systems in terms of discrete features contradicts another of his main themes, namely that (contrary to standard linguistic assumptions) phonemes may be almost but not quite merged. Discrete phonology seems irreconcilable with Labov's well-known finding, discussed at length here, of a near-merger between the voice and vice diphthongs on the Dengie peninsula in Essex, where some speakers consistently make the distinction yet cannot hear it.

This is a long and rather disorganized book. Several sections make important and novel points which merit careful debate but ultimately the author seems so swamped by rich and diverse data as to be unable to construct a consistent theoretical framework capable of accommodating them.

Anthony Fox's Linguistic Reconstruction is a textbook on the methods used to establish the properties of languages ancestral to those which now exist or which were spoken recently enough to have been recorded. Fox's survey is judicious and very well-informed; it draws on literature from the two-centuries-old beginnings of historical linguistics up to publications of the last year or two. Fox's subject was once central to linguistics, but became thoroughly unfashionable in recent decades; as a review written from the standpoint of modern understanding of language, Fox's book would be hard to beat.

It might perhaps be criticised as slightly Eurocentric: techniques such as the use of poetic rhyme to reconstruct earlier stages of Chinese phonology, which happen to be irrelevant in a European context, are not mentioned. Conversely, readers chiefly interested in European languages might hope for a chapter summarizing what is known or surmised about Proto-Indo-European; but Fox aims to describe the techniques of reconstruction: its results are mentioned only incidentally and sporadically.

Unlike Labov's book, which confuses the reader with numerous errors in phonetic transcription and diagram labels, Fox's book is produced as impeccably as one expects from Oxford University Press.

Geoffrey Sampson is reader in computer science and artificial intelligence, University of Sussex.