Contenu du sommaire : La liaison : de la phonologie à la cognition, sous la direction de Jean-Pierre Chevrot, Michel Fayol et Bernard Laks

Constraint-based theories of phonology have prompted the revival of an early seventeenth conception of liaison as a means to break hiatuses between words. Liaison would be the French response to the resolution of conflicts between various universal constraints, in particular word invariance and optimal syllable onsets. The specific interplay between various constraints, it is claimed, explains dialectal and social variations, language acquisition and evolution (Tranel 2000: 43). This paper examines the historical development of liaison in French and finds, however, no evidence that it ever was motivated by constraints against hiatuses or empty syllable-onsets. It has its source in the loss of some word-final consonants before word- initial consonants, and appears to be the result, in most cases, of imperfect transmission across generations of various consonant sequences. Later developments from the seventeenth century onwards were mostly motivated by what appears to be a cognitive bias for word invariance that was actually responsible for the steady increase of hiatuses between words in utterances. During the same period, various analogical processes lead to the extension of [t] liaisons before consonants after quand, vingt (Morin 1990) and, in some formal forms of speech, after third person verb-forms, as in Il faut [t] faciliter...

This paper views the alternation between liaison consonants and their absence as due neither to insertion nor deletion, but rather to the existence of alternant morphosyntactic constructions containing the liaison consonant. These alternate constructions can be viewed as irregular in much the same way as certain morphological paradigms are considered irregular. A certain level of token frequency is necessary to maintain these irregularities (as confirmed with data from Âgren, 1973), and the less frequent of such constructions tend to be lost. High frequency constructions (such as determiner + noun) are also felt to have a higher degree of syntactic cohesion; thus frequency can account for the intuition that liaison is maintained where there is tighter syntactic cohesion.

Lexical status of the liaison consonant are proposed to account for this later stage characterized by the formation of more abstract constructions including liaison.

This paper presents a new analyse of acquisition of liaison in French, supporting that liaison errors that are systematically observed between 2 and 3 years in the context of obligatory liaison (determiner + noun, i.e. un-n-ours [ènuRs] a bear' could be produced like un-z-ours [èzuRs], or un-t-ours [êtuRs]) are consecutive to the segmentation of the determiner which is originally produced as a proto-form completely integrated to the lexical unit. Analyses of data suggest that children procède in a "template" that is the domain of their generalisations ant that they apply systematically the Maximal Onset Principle to perform their segmentation. The paper supports a conception of phonological acquisition that is guided by general universal principles of grammar and morphological bootstrapping and not by a lexical storage of supplétive forms that the child should find out in the speech signal during the processing.

One of the issues raised by liaison concerns the lexical status of liaison consonants (LCs). LCs are generally assumed to be lexically attached to the preceding word. It has also been proposed that they are epenthetic and lexically independent, or that they belong to the following word. This issue is revisited in light of new or neglected data, notably from acquisition and phonetics. In the general case LCs are argued to be epenthetic, but in marginal cases they correspond to (fixed) initial or final consonants belonging to distinct allomorphs of the following or preceding word. The model of liaison that emerges from these different categories of LCs differs from the traditional view in several respects. First, liaison is not considered a uniform process. Second, "floating" segments are evacuated. Third, liaison appears to be driven by a constraint requiring lexical invariance, as it involves no modifications of lexical forms, and a constraint against allomorphy. This approach is characterized by the unicity, simplicity, and invariance of lexical forms. By contrast, previous analyses have viewed liaison as a simple and unified process, but they involve either numerous lexical exceptions, widespread allomorphy, or complex autosegmental representations.

How do French listeners recognise words in liaison environments? This article presents a literature review on liaison processing by French adults. First, we will see that despite phonological variations and resyllabification associated to the liaison phenomenon, lexical access is not impaired for bound words. We will then examine various factors that are involved in lexical access for words that undergo liaison. We will see that the resolution of liaison depends partly on acoustic cues that distinguish liaison consonants from non-liaison consonants, and partly on the availability of lexical support for a liaison interpretation.

Liaison in French is commonly described as involving a "latent" consonant whose realization or suppression is determined by contextual factors. This is an oversimplified view of the phenomenon, but liaison consonants (when pronounced) do have distinctive properties that call for a special treatment. Furthermore, the same notion of latent consonant appears to be relevant for morphological derivation. In this paper we offer a new formalization that captures the special status of latent consonants, by representing them as an "appendix" to the phonological "body" of a linguistic sign. We present an analysis of liaison that makes use of phonological representations of this type (while remaining general enough to accommodate cases where liaison does not involve a latent consonant). We explain in detail how the phonology is constructed in phrasal combinations, incorporating syntactic and lexical constraints, giving rise to the observed inventory of obligatory, optional, and impossible liaison contexts. We also show that the phonological appendix can be used to account for the realization of latent consonants in morphological derivation (even when these consonants do not participate in the liaison alternation).

Arguing both in synchrony and in diachrony, the paper shows that French liaison is not to be considered as a unique or unified phenomenon. The relationship between phonology and orthography is at play in the stabilization of contemporary liaison. We underline the interplay of three different dynamics. The first is a purely phonological one aiming at simplifying codas by dropping final consonants. The second is morphological in nature as its main purpose is the preservation of number and person marks. The third one is (ortho)graphical. It promotes the visual form of written words. Contradiction between those three dynamics leads to a state of equilibrium where liaison on a main category word is morphologically motivated and tends to be categorical, when liaison from a main category word is phonologically motivated and tends to be variable. This analysis argues for a constructional approach of the grammar and of the lexicon where graphical form, graphical endings and graphical segmentation of words are added to constructional representations.