Some additional info about the paper "A limit on behavioral plasticity in speech perception"

by Christophe Pallier, Laura Bosch & N. Sebastián-Gallés

The paper is available in pdf or postscript formats.

We provide here some figures which were not included in the original paper for reasons of space, and that show the data in more details.

Abstract

It is well attested that we perceive speech through the filter of our native language: a classic example is that of Japanese listeners who cannot discriminate between the American /l/ and /r/ and identify both as their own /r/ phoneme (Goto, 1971). Studies in the laboratory have shown, however, that perception of non-native speech sounds can be learned through training (Lively, Pisoni, Yamada, & Tohkura, 1994). This is consistent with neurophysiological evidence showing considerable experience-dependent plasticity in the brain at the first levels of sensory processing (Edeline & Weinberger, 1993; Kraus, et al., 1995; Merzenich & Sameshima, 1993; Weinberger, 1993). Outside of the laboratory, however, the situation seems to differ: we here report a study involving Spanish-Catalan bilingual subjects who have had the best opportunities to learn a new contrast but did not do it. Our study demonstrates a striking lack of behavioral plasticity: early and extensive exposure to a second language is not sufficient to attain the ultimate phonological competence of native speakers.


Stimuli

The stimuli consisted of seven vowels (s1 to s7) generated with CSRE4.2 implementation of Klatt's synthesizer. The stimuli differed only in the frequency of the first formant which increased from 404Hz (s1) to 584 Hz (s7) in steps of 30 Hz. F2 was set to 2084 Hz; the duration was set to 200 ms (with a dcrease in amplitude in the last 20 msec); the pitch increased from 112 Hz to 132 Hz in the first 50 msec, then decreased to 92 Hz for 100 ms and remained flat for 50 Hz.

These parameters were selected from goodness ratings obtained in a pilot work for synthetic vowels in the /e/-/E/ region of F1-F2 space.

Here are the stimuli:

s1.wav | s2.wav | s3.wav | s4.wav | s5.wav | s6.wav | s7.wav

Classification task (Experiment 1)

Fig. 1 Categorization of the /closed-e/-/open-e/ continuum (grouped results).

 

To assess individual variability we computed, for each subject, a "categorization indice", defined as "p7+p6+p5-p1-p2-p3" where pi=Prob(resp="pera" / stim=si).

Fig 1bis (not included in the paper) Histograms of categorization indices for Spanish (red) and Catalan (blue) subjects.


 


Discrimination task (Experiment 2)

Fig.2: Discrimination of vowel pairs (Grouped data).

To assess individual variations, we computed, for each subject, a discrimination indice defined as "p3-(p1+p5)/2" where pi is the score on the ith pair.

Fig 2bis (not included in the paper) Histograms of discrimination indices for Spanish (red) and Catalan (blue) subjects.


Correlation between Discrimination & Categorisation performances

On the following plot, each point represents a participant.

The correlation between the two tasks is significant (r2=.16; t(38)=2.7; p=.01)


Comment: It is clear from these histograms that there was an overlap between the performances of the Spanish-born and the Catalan-born subjects. Thus, one must not conclude from our paper that it is impossible for a second language learner to reach the performance of a native speaker. The difference between the two sub-populations is nevertheless remarkable.

We provide the file with individual data (excel format, text format) for those interested in further analyses. Email-me if explanations are needed.


Date: 19 Aug. 1997. Last update: 6/4/99 (stimuli added) Christophe Pallier.