Comparer des données de corpus : évidence, illusion ou construction ?Françoise Gadet (UPO & MoDyCo) et Sandrine Wachs (Sorbonne Nouvelle & DILTEC) Cet article étudie la propriété de comparabilité, trop souvent prise comme une évidence. Il s'arrête à la relation entre recueil de (grands) corpus et travail de terrain, quant à ce que les différents gestes induisent comme questionnements sur la langue et sa variabilité. Après avoir rappelé les démarches de grands corpus de français, il s'arrête au cas d'un corpus récent recueilli en région parisienne, MPF, pour finir par montrer quelles grandes questions sociolinguistiques pourraient être ouvertes à partir de la comparabilité. Mots-clés : Corpus, terrain, variation, variabilité, comparabilité, français.This article studies the property of comparability, which is too often taken as self-evident. It investigates the relationship between the collection of (large) corpora and fieldwork, and the questions induced by each about language and language variability. After summarizing some approaches adopted by large corpora of French, one particular corpus recently collected in the Paris region, the MPF, is discussed in greater detail. The study concludes by showing what broad sociolinguistic questions may be raised through the notion of comparability.

Rhythm plays an essential role in the segmentation of speech, a fundamental process in encoding and decoding messages for speakers and listeners. This study compared productions of French stressed syllables at different rhythmic boundaries, by French native speakers (NS) and Chinese learners (CL). The study started by retrieving duration of vowels and syllables within the examined utterances, with special focus on the last stressed syllables in Accentual Phrase (AP), for which temporal and non-temporal parameters were examined. Results concerning the temporal patterns showed that both NS and CL produced final lengthening in AP final syllables. However, CL's durations were much longer than the NS'. At the utterance level, CL with low proficiency had a wide temporal variation among segments, while high proficient CL had an isochrony tendency, similar to NS' performance. Results concerning the non-temporal parameters showed that low proficient CL produced a pitch fall between rhythmic units, which was not the case in NS' production. We considered this finding as a negative transfer from resetting of pitch register in Mandarin Chinese. Distribution of pauses was also analyzed, and some inappropriate pauses inserted within sense groups were found in CL. These pauses may weaken listener's decoding process in speech.