Multilevel modelling of the initial accent in French

Avanzi, Mathieu [UCL] Christodoulides, George [UCL]

In French, the minimal prosodic unit is the Accentual Phrase (AP). An AP is tonally marked by a high pitch movement associated with its rightmost full syllable and an optional high tone aligned with a syllable on its left (cf. the default pattern LHiLH*, identified by Jun & Fougeron 2002). These two tonal events are accompanied by durational cues: the final rise is marked by pre-boundary lengthening; and the initial rise is often marked by consonant onset lengthening. There is a general agreement in the literature regarding the fact that final rises are pitch accents, and the interactions of the numerous factors that influence their distribution are well-studied. Conversely, the phonological status of initial rises is still a question open to debate (Welby 2006), and the factors favouring its realisation are more poorly understood (Di Cristo 1998). Portes et al. (2012) recall that the initial rise (IR) does not have a fixed position: it can appear on a function word or on one of the first syllables of the content word of the AP, the first syllable of this content word being the most common position (Astésano et al. 2007). It has been claimed that the optional realisation of initial rises depends on metrical structure (the more syllables the AP contains, the greater chances for an IR to be realised), on speaking rate (the faster the speaker articulates, the greater the chances for an IR not to be realised), or on the position of the AP within the host intonational phrase (the IR would preferentially appear at the edge of pre-nuclear APs rather than at the edge of nuclear APs). It has also been claimed that speaking style has an effect on IR distribution: formal situations (such as political discourse, didactic speech, broadcast speech) and read-aloud speech would favour the production of initial rises (Lucci 1983). In addition, it has been shown that the geographic origin of the speaker is an important factor affecting the realisation of initial rises: Boula-de-Mareüil et al. (2012) show that speakers originating from the French-speaking part of Switzerland tend to realise more initial rises than Belgian of Hexagonal French speakers. To our knowledge, the way these constraints interact to define initial rise distribution in French remains unknown. The aim of this paper is to provide a multilevel modelling of initial rise realisation in spontaneous French, by evaluating the weight and the interactions of the factors mentioned above, based on corpus evidence. For this purpose, a 12-hour-long annotated corpus was used. The corpus consists of 14 regional varieties of French recorded in France, Belgium and Switzerland, with 112 native speakers (4 male and 4 female speakers per regional variety). Each speaker was recorded in a reading text task and a conversation task. The entire reading text and a stretch of 3 minutes of spontaneous speech for each speaker was orthographically transcribed and automatically aligned with the EasyAlign script within Praat (Boersma & Weenink 2014). The alignment was manually checked and corrected. Initial rises and pitch accents were identified independently by two experts, on the basis of their perceptual judgment only. AP boundaries (i.e. Clitic Groups carrying a pitch accent on their rightmost syllable) and intonation phrase boundaries (pitch accented syllables followed by a pause, associated with a major pitch rise and/or significant lengthening) were automatically identified and annotated on separate tiers. Results reveal that there is a significant effect of the grammatical status of the word (lexical words attract initial rise to a greater extent than function words), an interaction between the position of the syllable within the word and the number of syllables of the word (initial rises are preferentially associated with the first syllables of dissyllabic words), an effect of articulation rate (the mean shorter the mean syllabic duration of the host IP, the greater chances for IR to be realized), an effect of speaking style (a greater percentage of realised initial rises in the read speech sub-corpus than in conversation) and an effect of speakers’ origin: speakers from Switzerland realise more initial rises than those from France and Belgium. Surprisingly, no effect of the position of the AP within the host IP (pre-nuclear vs. nuclear) was found. These results and their impact on phonological status of the Initial Rise will be further discussed during the conference.