Neurophysiologic correlates of cross-language

phonetic perception

LING 7912Professor Nina Kazanina

perception of native and nonnative phonetic categorie• Hindi and English Listeners•

• The native Hindi listeners identified the stimuli as belonging to two distinct phonetic categories (/ba/ and /pa/). English listeners discriminated the same stimulus pair at a chance level.

In the electrophysiologic experiment N1 and MMN Were measured. • The changes in N1 latency were not significantly different for Hindi

and English listeners.• in response to the /ba/–/pa/ stimulus contrast, MMN was seen

robustly only in Hindi listeners and not in English listeners. • These results suggest that neurophysiologic levels of stimulus

processing reflected by the MMN and N1are differentially altered by linguistic experience.

Perceptual deficit in discriminating non native

phonetic contrasts • listeners’ speech perception abilities are altered by experience with

a particular language

Perceptual deficit: deficit in discrimination of some non-native speech contrasts for adult second lang. learners. (6–8 months old infants VS. adults)

• Certain non-native contrasts appear to be easier to discriminated or captured compared to others, which can be because of some shared phonetic features with our language.

• The prevailing view: The deficit is due to re-alignment of cognitive categories resulting from ‘‘higher-level’’ attentional biases rather than a neural-sensory loss.

• Is underlying neural sensory altered by Linguistic experience?

investigating neural correlatesof speech processing.

• MMN occurs only when there is the deviant stimulus presented in the context of a sequence of standard stimulis

• The N1 is an evoked response whose latency and morphology reflects the time of onset of acoustic events within speech

• N1 and MMN are considered as neurophysiologic indices of pre-attentive processing (are recorded without active participation of the subject)

• Is MMN measure of just a “sensory” or non-cognitive processing?

Experiment

N1 and MMN evoked potentials were obtained from Hindi and English speakers in response to

Stimulus: speech sounds varying in voice-onset-time which constitutes different phonetic categories in Hindi and but not in English.

I-BEHAVIORAL EXPERIMENT

• Identification Task

• Discrimination Task

A.Method

1. Subjects

Ten native Hindi speakers &ten monolingual speakers of American–English with no previous exposure to the Hindi language.

All subjects were paid $10/hour for their participation.

2. Stimuli

Speech sounds bilabial CV stimuli that differed in VOT

from -90 to 0 ms. Here they were ba/ and /pa/ recorded into a digital signal processing system.

• The vowels were produced with a slight /r/ coloring. The consonants were produced with pre-voicing as is appropriate in Hindi.

• The syllables were acceptable to native-English speakers as /ba/ and were acceptable to native-Hindi speakers as ‘‘baar’’ (which means ‘‘again’’ in Hindi) and ‘‘paar’’ (which means ‘‘side’’ in Hindi)

• The naturally produced stimuli were manipulated to creat stimuli with pre-voicing durations of 0 to 90 ms (see Fig.1)

3. Procedures

a) Identification Task

• Hindi subjects listened to the stimuli to classify each stimulus either as a /pa/ (or ‘‘paar’’) or /ba/ (or ‘‘baar’’) and to respond by clicking a computer mouse on panels marked PA and BA

• ten repetitions of each of the nine stimuli were presented to the subject in a random order.

b) Discrimination Task

• the stimulus pair chosen for the discrimination experiment had -10 and -50 ms VOT.

• Each trial consisted of two stimuli with (ISI) of 500 ms.

• Randomized subjects were asked to determine whether the randomized stimuli in the pair were ‘‘same’’ or ‘‘different.’’ by clicking on panels labeled ‘‘same’’ or ‘‘different’’ on the computer screen.

• The experimental session consisted of a total of 100

trials (50 same and 50 different) for each stimulus pair.

B. Results

Identification scores

• The group mean identification scores for each stimulus token are shown in Fig.2.

Hindi listeners consistently (i.e. more than 75% of the time) identified stimuli with VOT’s of 0 and -20 as /pa/ and -50 and -90 ms as /ba/.

• English listeners reliably identified all the stimuli as /ba/.

Discrimination Scores

• The mean discrimination scores for the -10 and -50 ms VOT stimulus pair are shown in Fig. 3.

• Hindi subjects: discrimination of the stimuli pair was with a high degree of accuracy

• English listeners discrimination of the stimulus pair was close to chance.

• A paired t-test revealed that Hindi subjects’ discrimination of the stimuli was significantly more accurate than their English counterparts (t=10.8; d f 18; p<0.0001).

Hindi subjects:

identification &discrimination results are consistent with previous reports of categorical perception for similar stimulus.

• English Subjects: identification & discrimination results are consistent with previous reports; changes in pre-voicing are not phonemically perceived by native English speakers.

II. ELECTROPHYSIOLOGIC EXPERIMENT

A.Method

1.Subjects:

subjects watched a videotaped movie of their choice. Subjects were asked to ignore the test stimuli that were presented through an insert earphones.

2. Stimuli

• a. MMN• For the MMN recordings, the stimuli is the same as those in the

behavioural discrimination experiment (-10 standard and -50 deviant ms VOT (using oddball paradigm).

• The deviant stimulus had a probability of occurrence=15%.• The stimuli off-set to on-set ISI of 510 ms

• b. N1. • For the N1, the stimuli is the same stimulus as in identification

experiment. Each stimulus separated by the ISI of 800 ms, –

3.Recording procedures

• MMN and N1 evoked potentials were recorded

a. MMN. • In the oddball paradigm, 2000 sweeps of the

response to the standard and 300 sweeps of the response to the deviant stimulus were collected

•

b.N1 300 sweeps elicited in response to stimulus were collected

4.Data analysis

• a. MMN. For individual subjects, sweeps were averaged to compute an individual average waveform elicited by the deviant stimulus when it occurred in the context of the standard stimulus (i.e., in the oddball paradigm) and by the deviant stimulus when it was presented alone.

The point of maximum negativity of the MMN component was noted and the adjacent relative positive peaks were selected as the MMN onset and offset.

Area of the MMN: a line was drawn from the onset to the offset of the MMN in the difference wave. The area was defined as a summation of the point by point multiplication (ms*µV) of the enclosed difference wave.

A significant negativity in the mean difference wave was defined as the group MMN.

b. N1.

In the group mean waveforms, N1 was identified visually as prominent negative peak within the first half of the time window.

• Group averaged waveforms were computed by averaging across the individual average waveforms.

B. Results

• 1. MMN

• The group averaged waveforms: Fig. 4. • A robust and statistically significant MMN was

obtained for Hindi listeners but not for English listeners.

• Furthermore, a t-test revealed that the MMN area was significantly larger for the Hindi compared to the English group (t=2.3; df=18; p<0.05).

• To summarize, the native (Hindi) speakers who could behaviourally discriminate the stimulus contrast exhibited a robust MMN to the contrast. (The effect of pre-voicing for Hindi speakers here which makes phonemically relevant categories)

• Non-native (English) listeners whose perceptual discrimination of the same contrast was at a chance level, a significant MMN in response to the stimulus contrast was not observed.

N1

• N1 latencies systematically reflected the acoustic change from the pre-voiced portion to the voiced portion within the syllables equally well for both the English and Hindi groups. (same N1)

• neurophysiologic representation of the duration of pre-voicing and voicing onset within a syllable was equally robust in native and non-native speakers.

• English listeners were unable to discriminate short versus long VOTs, But the N1 (showing changes in duration of pre-voicing) has not changed for them compared to Hindi speaker

Coclusion:

• Neural encoding of acoustic changes ongoing in a speech (N1) provides necessary but not sufficient information for behavioural discrimination of speech sounds.

• N1 and MMN reflect functionally different levels of processing.

• The N1 reflects stimulus processing which occurs at a sensory level that is not modified by exposure to the phonetic categories of a language.

• The MMN reflects a level of processing in which language specific categories play a role.

• The Future studies should systematically examine the changes in MMN that accompany relearning of nonnative contrasts in adult second-language learners.

• The characterization of the MMN as reflecting an entirely sensory level of processing is not tenable in light of the present results.