1. Can a prosodic pattern induce/ reduce the perception of a lower- class suburban accent in French? Philippe Boula de Mareüil 1 & Iryna Lehka-Lemarchand 2 1 LIMSI-CNRS, Orsay, France 2 Université de Rouen, Rouen, France

2. Introduction: French context Linguistic features specific to lower-class youth in France lexicon = favoured level of analysis pronunciation peculiarities  Accent de banlieue (‘suburban accent’)  very different connotation from that of the Anglo world often associated with violence, unemployment, immigration

3. Previous phonetic studies Usually focusing on the segmental aspects of the banlieue accent pharyngealisation of /  / palatalisation/affrication of /t/ and /d/ before high front vowels Particular prosodic patterns relative lengthening of the penultimate syllable pitch rise followed by an abrupt fall on the final syllable  before a prosodic boundary observed in the suburbs of Paris and Rouen (80 miles west from Paris, in Normandy) most often, starting from a high (H) tone and not associated with lengthening

4. Present study: questions Investigating how this prosodic pattern in adolescents from the suburb of Rouen is perceived by listeners living in Normandy and listeners living in the Paris region What is the contribution of this sharp fall (referred to as HL fall) to an impression of suburban accent? Does its presence/absence entail the perception of a higher/lower degree of accentedness? To what extent does perception depend on subjects’ region and social proximity to the suburb of Rouen?

5. General outline Study based on prosody modification and resynthesis  Corpus and method Perceptual experiments Results and discussion

6. Corpus Spontaneous speech collected among 11 male adolescents (aged 14 to 17) who lived in the suburb of Rouen two speakers with metropolitan French parents another two from French overseas departments the others of African descent Recordings made during directed interviews For each speaker, 10-second prosodically marked and unmarked utterances (i.e. exhibiting HL falls or not) selected

7. Method HL fall = a slope steeper than -5 ST between the end and the beginning of the word-final syllable Using the PSOLA algorithm implemented in Praat prosodically marked utterances rendered unmarked by smoothing the HL falls (especially by lowering the H tone) prosodically unmarked utterances rendered marked by modifying their pitch contours (especially by raising the H tone in the word-final syllable onset)

8. Stimuli Test material composed of 44 utterances 11 marked original (MO) 11 unmarked synthesised (US) 11 unmarked original (UO) 11 marked synthesised (MS) As many high-low falls in the MO and MS utterances: 22 (i.e. 2 per utterance on average) 1 – 3 HL falls in MO utterances 1 – 4 HL falls in MS utterances

9. Illustration and measurements MO pitch curve excerpt, with two HL falls on word-final vowel nuclei -5 ST -9 ST StimulusMOUSUOMS Slope (ST)-8.4-2.0-0.7-8.8

10. Illustration and measurements MO pitch curve excerpt, with two HL falls on word-final vowel nuclei -5 ST -9 ST StimulusMOUSUOMS Slope (ST)-8.4-2.0-0.7-8.8

11. Perceptual experiment: task Through a web-based interface self-estimating the passive (indirect) and active (direct) exposure to the suburban accent familiarisation phase Test proper rating the degree of accentedness on a continuous 0–5 scale (0 = no accent; 5 = very strong accent) specifying if certain words or syllables sounded particularly marked by the suburban accent (optional task) Final comments and answers to a couple of questions

12. Listeners 20 listeners from Normandy (aged 19 on average) 20 listeners living in the Paris region (aged 28 on average) For most listeners rather frequent passive exposure to the suburban accent, even though the active exposure is rarer In age and geographical proximity, listeners from Normandy closer to the test speakers than listeners from the Paris region However, active exposure to the suburban accent some- what more frequent in the subjects from the Paris region

13. Perceptual results: ratings No main effect of listeners’ Group (Normandy vs. Paris region), but a significant effect of the Type of stimulus (MO, US, UO or MS) In both groups of listeners, MO stimuli perceived as more accented than US stimuli, and MS stimuli perceived as more accented than UO stimuli (significant differences according to pairwise t-tests) StimuliMOUSUOMS Normandy2.72.42.12.4 Paris region2.82.52.12.4 Average2.72.42.12.4

14. Perceptual results: salient cues Pinpointed words classified according to whether or not they were target words, whose pitch contours were modified Target words more often pointed out in the MO and MS stimuli Importance of prosody confirmed by final comments

15. Experimental results: final questions Q.: Where do these youngsters live, according to you? for 8 listeners from the Paris region and 4 listeners from Normandy, associations with the suburb of Paris  possible generalisations to other suburbs of the northern half of France, with Paris playing an emblematic role Q’: Did many (at least 10) utterances sound artificial? positive answer in only 4 subjects from the Paris region and 11 from Normandy  reassuring result as far as the naturalness of the 22 acoustically modified stimuli is concerned

16. Discussion and conclusion Study concentrated on the perception of an atypical prosodic pattern (a final HL fall) in French Robust results involving listeners from two French regions, suggesting that this HL fall is indexical of the suburban accent and may function as a marker of identity More and more associated with the ethno-geographic origin, especially from the Maghreb, in folk comments Caution needed about the role of Arabic in the French suburban accent

17. Acknowledgements Work partially financed by ANR within the framework of the PADE project Thank you for your attention

18. Speakers Two speakers set aside for a familiarisation phase For the test proper, two prosodically marked and two prosodically unmarked utterances (plus their counterparts modified by synthesis) kept for two speakers Only one prosodically marked utterance and one prosodically unmarked utterance (plus their counterparts modified by synthesis) kept for the remaining 7 speakers