A quantitative reanalysis of schwa realization in contemporary metropolitan French
The French schwa – also known as the e-muet and e-caduc – is perhaps one of the most studied phenomena in phonology. Its alternation with zero seems highly unpredictable; however, this alternation is clearly subject to the influence of phonological and sociolinguistic factors. Despite the thousands of pages that have been written about this vowel over the last five-hundred years, linguists disagree on virtually every aspect of this vowel, including its representation and definition. Yet, in spite of the variety in definitions and conceptualizations of this vowel, much of our understanding is dependent on generalizations of the schwa and its distributions that were written in the late nineteenth and mid-twentieth centuries (i.e. Delattre, 1939/1966, 1949, 1951; Fouché, 1959; Grammont, 1894, 1920). Following from Weinreich, Labov, and Herzog (1968) who argue that variation is the crux of any linguistic system, this dissertation proposes a purely quantitative reanalysis of the e-muet and its distribution. Before any sort of theoretical treatment of a variable phenomenon can be applied, a thorough data-driven description must be furnished (c.f. Laks & Griffiths, to appear).
Data from this study come from the Projet phonologie du francais contemporain corpus (PFC; Durand, Laks, & Lyche, 2002) a large corpus of spoken French with a coding protocol that specifies and annotates every schwa within (n = 207167). A general and adaptable program is written to clean and further specify the PFC query as part of pre- processing. Once pre-processed, data were fit to series of generalized linear mixed-effects regression models and random forest models that predict the realization and distribution of the e-muet.
I find that the traditional descriptions of the French schwa-zero alternation are largely accurate, and the multifaceted variability of this vowel is subject to segmental, prosodic, and phonotactic constraints. I also find that the variability of this alternation is subject to a large effect from the lexicon, particularly in Northern Metropolitan France. The primary motivation behind the schwa-zero alternation is found to be phonotactics in two varieties of Metropolitan French, but the granularity afforded by the pre-processing program allows for a more nuanced view into the phonological factors affecting this alternation. These include a reconsideration of the domain for phrase-final schwa realization from the intonational phrase to the accentual phrase, and an effect of fricatives, which were found to facilitate schwa deletion. I also find that lexical schwa and epenthetic schwa are subject to two distinct linguistic distributions; however, they curiously overlap in the word-final position resulting in an asymmetry between word-final schwa deletion and word-final schwa epenthesis in distinct lexemes. An age effect is also documented in Southern France whereby younger southerners are beginning to realize schwa at similar rates to northerners. Finally, I conclude with a call for a reconsideration of the e-muet not as one phenomenon, but instead as a collection of phonetically similar and phonologically distinct phenomena. The quantitative reanalysis furnished in this thesis reaffirms much of our understanding of the French schwa, while shedding light on new peculiarities.