50 Bernard Laks, Basilio Calderone and Chiara Celata
individual realizations, distributed over a socially and geographically stratified col-
lection of speakers. The analysis of enacted liaison in the PFC sample thus suggests
that corpus analysis may reveal the existence of subtle variations even for those
phenomena generally considered as ‘monoliths’ in a linguistic repertoire and with
very few internal sociophonological variation.- General discussion
At the conclusion of this quantitative analysis of liaison in the PFC corpus, the
image which emerges is one which differs significantly from that put forward by
the generative or post-generative phonologies which postulate the presence in all
cases of a latent consonant in an abstract representation. We have shown that if
we decide to analyze not the absence of liaison (as a consequence of the erasure
process) but rather the process that positively produces liaison, we find that a
relatively small number of <word> linked <word> constructions may account for
up to the 50% of total productions. In agreement with Bybee’s suggestions about
the statistical organization of the mental lexicon (Bybee 2005, 2006; Bybee and
McClelland 2005) and the theories of Usage Grammars (Barlow and Kemmer
2000; Langacker 2000), we believe that these constructions should be regarded
as ‘frozen’, i.e., stored as such in the mental lexicon (see also Chevrot et al. 2011
for an acquisitional point of view). However, we have seen that mnesic storage
of recurrent <word> linked <word> constructions, probably to be considered as
single words with lexicalized liaison, is not in itself sufficient to explain liaison and
its complexity. We must also postulate the existence of a more marginal process,
allowing the generation of a very large number of those rare liaisons that represent
the remaining 50% of all enacted liaisons. Mnesic storage corresponds to a gen-
eralized linkage dynamic that is typical of cursus languages (Pulgram 1970). Such
a process erases the boundaries between words and confers upon the prosodic
group a central role in oral production (Grammont 1914). By contrast, general-
ization of rare liaisons is a tendency driven by the orthographical demarcation
between words. This process, in contrast to mnesic storage, reaffirms the ortho-
graphical integrity of words.
It is also important to remark that traditional lexicalist analyses of French
liaison have been based on the analysis of “constructions”, i.e., “groupes de mots
qui ont une ‘forte cohesion syntaxique’ ” (Bybee 2005: 24). These are therefore
grammatical constructions, instantiated by specific lexical sequences with their
own degree of syntactic cohesion (usually strong to very strong) and their own
frequency of co-occurrence (varying from low to high). Syntactic cohesion
and frequency of co-occurrence are the epiphenomena of lexical properties of