Processing inter-speaker variability: the case of regional accents Floccia, C. Laboratoire de psychologie EA3188, Université de Franche-Comté, [email protected]

Goslin, J. Laboratoire de psycholinguistique expérimentale, Université de Genève

& Girard, F. Laboratoire de psychologie EA3188, Université de Franche-Comté

Abstract Two experiments investigated regional accent perception in adults. Subjects performed lexical decision on the last items of sentences, which were uttered with a familiar or an unfamiliar regional accent. In the first experiment, it was found that adults were faster in recognising a word in the familiar accent conditions than in the unfamiliar accent one. A second experiment was conducted in order to determine whether this effect was due to the processing of the word itself, or to that of the preceding sentence.

1

Introduction

We are all experts at identifying people’s geographical origin by listening to their accent, especially regional accents in our native language. This suggests that we have acquired a remarkable representation of the possible phonological and prosodic variations in our language. How do we process and integrate these variations? Are they encoded in short-term memory and used during lexical access, or are they processed independently of our typical word extraction routines? Accent perception offers a privileged insight not only on the phonological and prosodic representations used in speech perception, but also how it copes with variability. Contrary to other types of cross-speaker variability (such as voice quality), accent variations can be classified according to phonetic, phonological and prosodic features. Until recently accent perception has been largely neglected, with most studies focussing upon production issues (Long, 1993; Chambers, 1992), or intelligibility (e.g. Munro & Derwing, 1995). However, of late interest in this area has increased, with two studies that examine children’s remarkable abilities in the acquisition of regional accents (Nathan, Wells, & Donlan, 1998), and possible impairments of these abilities (Nathan & Wells, 2001). In our study, initiated in the year 2000, we aim to explore regional accent processing during both development and adulthood. In this report we present two experime nts designed to test for the existence of a perceptual cost in word recognition processes due to a regional accent in adults. These data would be the pre-requisite for future research aimed at identifying the mechanisms responsible for accent integration. These experiments used a lexical decision task in which the to-be-detected word (or nonword) was placed at the end of sentences uttered by speakers with different regional accents. Our first hypothesis was that an unfamiliar accent would elicit a global processing delay in word identification that would spread over the sentence until its last word. The other hypothesis was that, as in speaker normalisation, only a few syllables or phonemes would be necessary for the listener to become attuned to an unfamiliar accent (see Jusczyk & Luce, 2002, for a review of similar effects in speaker-voice adaptation), and thus no perceptual cost should be observed on the last word of the sentence.

2 2.1

Experiment 1 Participants

Twenty psychology students (aged 23;11) were tested. They were all monolingual French speakers, natives of the Franche-Comté region.

2.2

Stimuli

Nineteen disyllabic words and ten non-words were selected. Each item appeared in a total of 12 sentences, 4 sentences for each of the regional accents under analysis. For each of the three regional accents, Franche-Comté (Besançon) (FC), Aix-en-Provence (AP), and Paris (P), two female speakers were used to produce the stimuli. This resulted in a total of 348 sentences, each with a final word selected for unpredictability, as asserted by a pilot sentence completion test. Words Bateau Cahier Bonnet Soleil Robot Serpent Forêt Mouton Poulet Poisson

Matin Ballon Sapin Maison Souris Melon Bougie Lapin Vélo

Non-words Lupi Rimo Duti Tulo Firchou Souchu Bulu Sirma Futa Nufou

Table 1. List of stimuli used in Experiment 1

2.3

Procedure

Lexical decision was to be performed on the last item of each sentence. No answer was requested for nonword stimuli.

2.4

Results

There was a main effect of accent, by item (F(2,36)= 8.21; p=0.0012), and by subject (F'(2,38)= 17.29; p=0.0000). This effect was due to a longer delay in processing words uttered with an AP accent (747 ms), than those with a P (712 ms) or a FC accent (709 ms) (see figure 1).

TR(ms) 750 740 730 720 710 700 690 Paris

Franche-Comté

Aix-enProvence

Figure 1. Mean lexical decision times as a function of accent in Experiment 1

2.5

Discussion

The first interpretation of these results is that processing time is related to participant’s familiarity with the accents. Naturally, all participants were familiar with their own, FC accent, with widespread exposure to the P (Parisian) accent due to media coverage. It is unlikely that the participants would have had equal exposure to the AP accent. However, it is also possible that other factors could be responsible for this effect. Firstly, it is possible that, even with careful selection of speakers, there was a significant difference in the speaking speed of AP speakers when compared with those of the other two accents. Post-hoc analyses showed a strong correlation between word lengths and RT’s (the longer the word, the slower the answer). However this was the case for all accents, and AP words were not found to be significantly longer than P or FC words. Secondly, it is possible that the disparity between the accents could have been due to lexical decisions being made on words uttered by different speakers. Therefore it is not possible to determine whether the effect was due to the influence of the accent carried by the test item itself, or to the influence of accent carried by the preceding sentence, which is the effect we were looking for. To differentiate between these factors we have repeated the task used in experiment 1, but have used a single speaker to produce all of the final words in the sentences.

3

Experiment 2

This experiment is a repetition of the task used in Experiment 1, diverging only in that a single male speaker with a neutral P accent produced the last word of the stimuli.

3.1

Participants

Forty three participants were tested (5 men and 38 women) (aged 21:7, from 18:6 to 47:2), all selected from the FC region as in Experiment 1.

3.2

Stimuli

Ten disyllabic test items were selected, all starting by an unvoiced plosive, to facilitate cross splicing. Each word appeared in 12 different sentences, split between two accents (FC and Toulouse: T), with two female speakers per accent. Six legal nonwords were also constructed. This resulted in a 192 sentences list. Words Poisson Couteau Carotte Camion Piano Cochon Canard Cadeau Pinceau Tortue

Non-words Timo Pufoul Kirchou Touli Cavane Paro

Table 2. List of stimuli used in Experiment 2

3.3

Results

There was no main effect of accent, in both analysis by subject (F(1,42) < 1) and item (F(1,9) < 1). Mean reaction times were 511 ms for both accents (510.6 ms for FC and 510.9 ms for T).

4

Discussion

A possible reason for the lack of the accent effect in this experiment could be due to differences in average reaction times between experiments 1 and 2. As reaction times were slower in Experiment 1 than 2 it is possible that the effect in the former experiment was due to the slower subjects. However, this possible explanation can be discarded, when we divided Experiment 1 subjects into a fast and slow group we found the effect of accent in both groups. Another explanation is that the perceptual cost due to accents measured in Experiment 1 might be due to the accent carried by the test item itself, and not to the accent carried by the preceding sentence. This would suggest that, contrary to our hypothesis, the cost of accent variability is not propagated onto proceeding speech segments. To investigate this issue in greater detail we are now conducting lexical decision experiments using isolated words. Finally, a last possibility is that the voice change at the end of the sentence in Experiment 2 (from female to male) is very perturbing for the subjects. They could adopt a strategy consisting of ignoring the first part of the sentence, and concentrate on the male voice. This attentional shift would lead to a decrease of the accent effect. The isolated word experiment results will also allow us to conclude on this matter.

References Chambers, J. K. (1992) Dialect acquisition. Language, 68, 673-705. Jusczyk, P. W. & Luce, P. A. (2002). Speech perception and spoken word recognition : Past and present. Ear and Hearing, 23, 2-40. Labov, W. (1989). The limitations of contexts: evidence from misunderstandings in Chicago. Chicago Linguistics Society, 25, 171-200. Long, M. H. (1993). Second language acquisition as a function of age: research findings and methodological issues. In K. Hyltenstam & A. Viberg (Eds.): Progression and Regression in Language: Sociocultural, Neuropsychological and Linguistic Perspectives. CUP. Munro, M. J., & Derwing, T. G. (1995). Processing time, accent and comprehensibility in the perception of native and foreign-accented speech. Language and Speech, 38, 289-306. Nathan, Liz, & Wells, Bill (2001). Can children with speech difficulties process an unfamiliar accent? Applied Psycholinguistics, 22, 343-361. Nathan, Liz; Wells, Bill & Donlan, Chris (1998). Children’s comprehension of unfamiliar regional accents: a preliminary investigation. Journal of Child Language, 25, 343-365.

Acknowledgements This research is part of a Cognitique project entitled « Ecole et Sciences Cognitives. Apprentissage des langues : dysfonctionnements et remédiations » (2002-2004), directed by Pr. Gabrielle Konopczynski and the first author. Many thanks to Céline Saget and Anne Bardin.