French liaison and linguistic theory By Jürgen Klausenburger (review)

BOOK NOTICES 721 discuss this fact. From the data, we can't tell whether some clitic object pronouns trigger rafforzamento , like the neuter article—or whether certain consonants, such as [b], are always lengthened in initial position (or whether some other explanation is correct). (r) G's translations into Italian also arouse curiosity. At times he translates with a construction different from that of the dialect, even though a corresponding construction exists in standard Italian, thus, for fveks a idda ka skapps] (63), he gives Lo vedo fuggire (scappare ) ? see him run away', even though one can say Lo vedo che scappa. In sum, this is a descriptive work with a wealth of information, much of which demands further investigation. [Donna Jo Napoli, University ofMichigan.] French liaison and linguistic theory. By Jürgen Klausenburger. Stuttgart : Steiner, 1984. Pp. 83. K aims here primarily to present a history of the analysis of French liaison, mostly since the 1960's, with an account of present controversies . Only as a secondary consideration does he offer his own analysis. K's first goal does not seem to be satisfactorily achieved. Though he does discuss interesting work not previously given attention in the literature (52), he fails to do justice to the complexity of much of the work on liaison in a generative framework. One of the fascinating facts about liaison is that its conditioning factors are not strictly phonological or morphological, and an extensive literature is devoted to the question ofprecisely what the conditioning factors above the word level are. K alludes to the debate, but only briefly; and in his concluding remarks, he calls the behavior of liaison 'elusive' (77), attributing that quality principally to 'stylistic factors , notoriously inappropriate candidates for scientific systematization'. This conclusion is unfortunate: if it were true, it would render liaison of no interest except as a phonological phenomenon. Yet the occurrence of liaison is not arbitrary. Although K claims that a listing approach to the non-phonological conditioning factors for liaison is about the best we have been able to achieve, he is aware of studies on sandhi phenomena in other languages (such as 'raddoppiamento ' in Italian) which point out strong similarities between the non-phonological factors governing those phenomena and the ones governing liaison. A list approach, with no attempt at generalization to predict which sort of item will occur, is an implicit statement of the arbitrary nature of the information on the list; it has no chance of accounting for the existence of these cross-language similarities in sandhi rules. I take the position that similarities such as these are not coincidences; and thus I would want any discussion of liaison to enter into the debate about non-phonological conditioning factors. K's second goal is achieved: he gives us his analysis. After discussing the generative analyses ofdeletion and insertion, K opts for a suppletion analysis. He argues that a large amount of data can be accounted for in this way: yet he sets aside a category of examples upon which he suspends judgment (26). In any case, his argument loses force if data exist which call for a generative analysis, since it could then be extended to cover all the data. Later, K argues that other morpho- (phono)logical phenomena are also best accounted for by suppletion; he suggests that ideas about what constitutes an adequate theory have changed enough so that the value of such analyses can now be seen. This value is said to lie primarily in their simplicity (since other rules necessary to support generative analyses are no longer needed) and in the fact that the phenomena which K considers are idiosyncratic—and, as such, should be handled in the lexicon (634 ). The suppletion analysis of liaison, however, may lack some of the explanatory value of a generative analysis such as deletion. For example , with a suppletion analysis, ifonly one of the two forms of a given lexical item in the lexicon is used for derivational processes, the choice of which form is used by each process for each lexical pair could be random. But it is not: one form prevails across the lexicon for each process. K ends with a test...