Behavioral and electrophysiological evidence for the impact of

Behavioral and electrophysiological evidence for the impact of regional variation on phoneme perception A. Brunellière, S. Dufour, N. Nguyen, U. H. Frauenfelder Cognition 111 (2009) 390– 396 1

Introduction • Phonologies vary in time and space. • Flexibility of speech perception & existence of variation learning mechanisms (Mc. Queen, Norris, & Cutler, 2006) – Exposure to different varieties and to phonetic variation results in accommodation of different accents/speakers. – • Speakers' production changes over time (e. g. the Queen) Do “adjustments to other accents influence the perception of speech sounds in the listener’s own accent”? (p. 390) – Vowel mergers in France French vs. unmerged vowels in Switzerland French. 2

Introduction: Vowel discrimination • Two components related to vowel discrimination in ERPs: – Mismatch Negativity (MMN) component: • • – Frontocentral negative component at 100 -250 ms Sensitive to acoustic and phonological differences P 200 • It reflects processing of phonemic categories. 3

The Mismatch Negativity - I (Pulvermüller & Shtyrov, 2006) • It indicates the (mis)match between a stimulus and its memory trace in the brain. • Elicited by infrequent acoustic events (deviant stimuli) – • Bilateral or right-dominant distribution It also reflects higher level cognitive processes in the auditory system. – Lateralized to the left hemisphere • Semi-automatic brain response • It also shows sensitivity to lexical status, grammaticality, meaning. 4

The Mismatch Negativity - II (Pulvermüller & Shtyrov, 2006) • The MMN is appropriate for studying the brain basis of language because. . . – It is (semi-)automatic. – It occurs early. – It is suitable for studying individual linguistic items. It is not affected by the physical and psycholinguistic variance problems. • Oddball paradigm A relatively pure measure of memory trace activity can be obtained. • – 5

The P 200 (Dehaene-Lambertz, 1997) • Three conditions: control, within category, across category. – Native (/b/-/d/) and non-native (/d/-/ /) • Synthetic stimuli were acoustically equidistant (6 of a 16 syllable continuum). • Results: – P 200 and MMN for the native contrast, but not for the non-native contrast 6

The P 200 (Dehaene-Lambertz, 1997) Native phonemic contrast Within cat. Across cat. Foreign phonemic contrast Within cat. Across cat. 7

The French vowel system (Fagyal, Kibbee, & Jenkins, 2006) e Merger vs. non-merger Difference only in vowel height Albert: Is the contrast really preserved in French? Standard French & Swiss French: épée 'sword' /epe/ vs. épais 'thick' /ep / Southern French: /epe/ 8

Materials and methods Participants • 14 right-handed French-speaking students from the University of Geneva (Switzerland) – Lynn: Number of men/women? – Lynn & Albert: How much exposure? – Israel & Albert: Lack of a control group or 3 -group comparison. • Handedness assessed using the Edinburgh Inventory • No neurological or hearing impairment • Error rate in the same-different task < 20% 9

Materials • Stimuli were recorded from 4 female and 1 male French speakers. – Focus on phonological processing – Israel & Albert: What variety of French? – Israel & Pawel: Speaker variation • Target syllables: /be/, /by/ • “Best” tokens were selected. – • The male's tokens that best matched the females' in f 0 and amplitude envelope were selected. Duration was adjusted to 170 ms (50 ms prevoicing + 120 ms of plosion and vowel). – Israel: How good is the resulting token? (Grosjean et al. 2007) 10

Materials • Average euclidean distance in the F 1 -F 2 plane at vowel midpoint: – /be/- /b /: 2. 09 Barks – /b / - /be/: 1. 77 Barks – /b / - /by/: 1. 51 Barks – /by/ - /b /: 2. 75 Barks Critical band (bark) scale: An auditory scale based on the idea that the human auditory system below 16 k. Hz is composed of 24 internal bandpass filters (critical bands). (Syrdal & Gopal, 1986) 11

Materials The “Auditory Image model of peripheral auditory processing (Patterson, Allerhand, & Giguère, 1995) (. . . ) simulates the activity that complex sounds elicit in the auditory pathways”. (p. 392) For each syllable, an auditory spectral profile was computed at vowel midpoint. The members of each contrast are deemed to differ to a similar extent, as shown by the average euclidean distance between the auditory spectral profiles of the context vowel and the test vowel. 12

Materials 13

Procedure • Stimuli were presented binaurally via headphones in an acoustically shielded chamber. • Trials: – 5 syllables separated by 600 ms of silence. – The 1 st four were tokens of the same phoneme each uttered by a different woman. – The 5 th stimulus was produced by a man and could be the same as (control condition), or different from (deviant condition) the rest. – Total of 320 trials (80 trials x 2 conditions x 2 contrasts) – Inter-trial interval: 3000 ms 14

Procedure • Same/different bimanual response • RTs measured from the onset of the 5 th syllable • 16 practice trials + 2 blocks of 160 experimental trials 15

Recording system The EEG (sample rate 1024 Hz) was recorded from the scalp with a 64 -channel Bio. Semi Active-Two AD-box. Individual electrodes were adjusted to a stable offset of <20 m. V. The EEG epochs (from 100 ms before test syllables to 500 ms after them) were averaged for each experimental condition and for each participant. The data were filtered offline by a bandpass (0. 7– 20 Hz) and corrected by a baseline of 100 ms before the test syllable onset. Epochs were accepted under an artefact rejection criterion of 70 μV. 16

Recording system Accepted trials > 50 for each experimental condition for all participants The epochs made for each experimental condition and for each participant were calculated independently of whether the behavioral responses were correct or incorrect. Data from bad channels for each participant were interpolated (Perrin, Pernier, Bertrand, Giard, & Echallier, 1987) and the EEG signal was transformed using the average reference. – Lynn: What is a bad channel? 17

Data analysis • Behavioral data: – • EEG data: – • ANOVA on error rates and RTs with phonemic contrast (merged vs. non-merged) and condition (control vs. deviant) as variables. ANOVA around each ERP component with phonemic contrast, condition and sites (frontocentral, centroparietal, L temporal, R temporal, L posterior and R posterior) as factors. The Greenhouse-Geisser correction was applied. 18

Data analysis EEG data: – Time windows (after onset of test syllable): • • N 100: 73 -113 ms P 200: 190 -230 ms MMN: 270 -310 ms Later effects: 372 -486 ms – – Lucy: Can foreign phonemes elicit N 400? Sites: frontocentral (F 1, Fz, F 2, FC 1, FCz and FC 2), centroparietal (CP 1, CPz, CP 2, P 1, Pz and P 2), left temporal (AF 7, F 5, FT 7, T 7 and AF 3), right temporal (AF 8, F 6, F 8, FT 8, T 8 and AF 4), left posterior (TP 7, CP 5, CP 3, P 7, P 5 and P 3) and right posterior (TP 8, CP 6, CP 4, P 8, P 6 and P 4). 19

Behavioral results • Error Rates: – Significant contrast x condition interaction • • CC: no differences between contrasts DC: more errors for the /e/-/ / contrast (8% vs. 1%) – • Albert: Is the contrast really preserved in French? RTs: – Significant contrast x condition interaction • • CC: No differences between contrasts DC: Longer latencies for the merged contrast (837 ms vs. 702 ms) 20

EEG results • N 100: – – • Effect of sites only: Frontocentral and centroparietal sites had more negative values than the remaining four. FC more negative than CP P 200: – Trend effect of condition (p=. 06): more negative values in CC than DC – Contrast x condition interaction: more positive values for / /-/y/ in CC than DC, but no effect of condition was found for /e/-/ /. 21

EEG results • MMN: – Condition x sites interaction: More negative values for DC than CC at frontocentral sites More negative values for CC at L temporal sites Centroparietal sites showed more positive values for CC than DC. No interaction between contrast and condition on any site • • – • Late time window: – Contrast x condition interaction at the L temporal sites: / //y/ elicited more negative values for CC. – No effect of condition for /e/-/ / 22

EEG results 23

EEG results 24

Discussion • Clear processing differences between the two contrasts: – /e/-/ / induced only a MMN, but / /-/y/ elicited a MMN and electrophysiological differences between C and D stimuli on the P 200. – / /-/y/ is discriminated earlier and more easily than the merged contrast. – Results of the behavioral experiment are in line with those of EEG: /e/-/ / elicited more errors and slower responses. – Becoming attuned to other accents has an impact on the native speakers' perception of their own accent. • Lynn: baseline MMN 25

Discussion • The effects observed are phonemic, not simply acoustic. – – • Contrasts were chosen so that their members were auditorily equidistant. Different speakers were used, introducing acoustic variability and forcing participants to rely on abstract representation when making their decisions. The /e/-/ / contrast was more difficult to discriminate. – Greater variability of these vowels may result in memory traces that are close together in the phonemotopic map. • Pawel: changes in phonemotopic map 26

References Dehaene-Lambertz, G. (1997). Electrophysiological correlates of categorical phoneme perception in adults. Neuroreport, 8, 919– 924. Mc. Queen, J. M. , Norris, D. , & Cutler, A. (2006). The dynamic nature of speech perception. Language and Speech, 49, 101– 112. Pulvermüller, F. , & Shtyrov, Y. (2006). Language outside the focus of attention: The mismatch negativity as a tool for studying higher cognitive processes. Progress in Neurobiology, 79, 49– 71. Syrdal, A. K. & Gopal, H. S. (1986). A perceptual model of vowel recognition based on the auditory representation of American English vowels. Journal of the Acoustical Society of America, 79: 4, 1086 -1100. 27