LLF – Bât. ODG – 5e étage – Salle du conseil (533)
Philippe Martin (Université de Paris)
Intonation et grammaire de dépendance/ Intonation and dependency grammar
The dominant Autosegmental-Metric model describes the sentence intonation as a succession of tonal targets obtained by percolation from the syntactic structure, targets of which a grammar must account for the well-formed sequences. In the absence of appropriate specific syntactic models, this approach shows some difficulty in describing the intonation of spontaneous utterances.
Conversely, by reversing the order of syntax-intonation operations, i.e., by assuming that sentence intonation is put in place before the syntactic structure (in chunks of 2 to 3 seconds), we can describe the prosodic structure independently of the syntax, based solely on its definition: merging stress groups in several levels, stress groups being groups of words containing only one stressed syllable, excluding emphatic stress, and in final position in French.
The melodic variations carried only by vowels of stressed syllables (excluding emphatic stress), classified according to the threshold of perception of their melodic change (i.e., vs. perception of a static tone), then appear as indicators of dependence relations between stress groups. From then on, the intonation grammar of French, but also of English, Italian, Dutch or Korean, becomes quite simple and boils down to a few dependency rules as indicated by melodic contours located on stressed vowels, while accounting for the intonation of both read and spontaneous sentences.
The data supporting this model was obtained by a dedicated software program, allowing reliable prosodic annotation even for noisy or poor-quality recordings. This software also has, in addition to the automatic segmentation API for more than 42 languages, a graphic simulator of the prosodic structure operating from a sequence of melodic contours, editable by the user and allowing an instant replay of the sentences melodically modified.