View
228
Download
2
Category
Tags:
Preview:
Citation preview
Identification and discrimination of the relative onset time of two component
tones: Implications for voicing perception in stops
David B. Pisoni (1976-77)
Past Studies
• Lisker and Abramson (1964, 1970)• Liberman et al.• Mattingly, Liberman, Sydral, and Halwes• Eimas (1971)• Kuhl Miller (1975)• Lasky• Streeter• Miller et al
Lisker and Abramson
• They investigated – Voicing and aspiration differences shown across different
languages (last class)
– Differences in timing and glottal activity
• They discovered 3 modes of voicing (1964):– Pre-voiced stops = voicing onset precedes the release burst
(negative onset, -VOT)– Short-lag voiced stops = voicing onset is simultaneous or
briefly lags behind the release burst (0 VOT)– Long-lag voiceless stops in which the voicing onset lags
behind the release burst (positive onset, +VOT)
Liberman
• In perceptual experiments done with synthetic stimuli* they found (1961)…– Subjects ID and discriminate differences in VOT in
a categorical-like manner that reflects the phonological categories* of their language
– Consistent labeling with sharp crossover points – Discontinuities in discrimination that are correlated
with the abrupt changes in the labeling functions– Better at discriminating 2 synthetic stimuli from 2
different phonological categories vs. from the the same
According to Several
• Empirical Findings– Non-speech signals are perceived in a continuous
mode– No other categorical perception studies had been
done with synthetic stimuli– Non-monotonic discrimination functions are the
result of labeling processes associated with phonetic categorization
• Interpretation– Evidence for the operation of a special mode of
perception…Speech Mode
Liberman et al. (1961) & Mattingly, Liberman, Syrdal
and Halwes (1971)• Are discontinuities in speech discrimination functions
due to the acoustic or psychophysical* attributers of the signals themselves rather than some speech related labeling process?
• Found no peaks in the non-speech discrimination functions at phoneme boundaries so…
• Conclusion = Speech Mode – discrimination of speech stimuli were attributable
to phonetic categorization resulting from the stimuli being perceived as speech.
Eimas (1971)• 2 and 3 month old infants• Found that they can discriminate synthetic
speech sounds varying in VOT much like English speaking adults
• Implication = infants have access to mechanisms of phonetic categorization
• Innate mechanisms• Responding to phonetic coding VS.
psychophysical differences• Environment plays a secondary role
Kuhl and Miller (1975)
• Study done with chinchillas• Trained to respond differently to the consonants /d/ and
/t/ (human voice)• Used synthetic stimuli varying in VOT with a sharp
crossover point• The discrimination functions were similar to English
speaking human data…but chinchillas don’t have spoken language
• Suggests a psychophysical basis VS. phonetic basis for the labeling behavior
• Results = the boundary for voiced and voiceless labial stops occurs at about +25 msec…threshold
Lasky et al. (1975)• Cross-language studies• 4 to 6 1/2 month old infants born to Spanish-speaking
parents• Found evidence for 3-categories in discrimination
– Boundary occurred in the region of +20 msec and +60 msec (corresponds to the English voiced/voiceless times)
– And at -20 msec and -60 msec • Spanish only has one phoneme boundary b/w voiced
and voiceless stops and it does not coincide with the boundaries they found
• Conclusion: – Environment plays minor role
– Responding to psychophysical attributes
Streeter (1976)
• Kikuyu infants (Kenya)• Show evidence of 3 categories of voicing for
labial stops• Kikuyu have no voicing contrasts for labial
stops (but they exist at other articulation places)
• Conclusion:– They had not been exposed to these before– Responding to psychophysical attributes– Similar to the Laskey et al. research
Miller et al. (1976)• Non-speech control signals
– Using VOT in the form of a noise bust and a buzz
• Adults• Results: discrimination functions that were
similar to those found with stop consonants differing in VOT– Discrimination was excellent for stimuli selected
from b/w categories and poor for stimuli within a category
• Perceptual threshold• Psychophysical account
Pisoni
• Independent from Miller et al. but at the same time
• Used stimuli that varied in temporal order of the onsets of 2 component tones at 2 frequencies (Figure 1):– 500 Hz– 1500 Hz– -50, 0, +50 msec VOT (ranging in 10 msec
increments)
Pisoni
• Goals:– To learn something about how the timing
relations in stop consonants are perceived– To provide a more general account of the
diverse findings obtained with adults, infants and chinchillas on VOT stimulus
– To provide an account of the results obtained with non-speech stimuli
Pisoni: Experiment I• 8 paid volunteers from ad in student paper• All were right handed and native English
speakers• Stimuli (Figure 1):
– 11 digital two-tone sequences– Lower tone = 500 Hz– Higher tone = 1500 Hz– Variable is VOT
• -50• 0• +50
Pisoni: Experiment I
• Stimuli was presented at 80 dB SPL• 2 one-hour sessions done over 2 days• Day one:
– Identification training sequences– Presented with the endpoint stimuli (-50 & +50)– Told to learn (w/their own strategy) which one of
the 2 buttons was associated w/ea sound– Immediate feedback for correct responses
Pisoni: Experiment I
• Day two:– Tested for identification
• 11 stimuli presented in random order• No feedback
– Tested for ABX discrimination*• 9 two-step pairs along the continuum• Feedback provided for correct responses• Told to determine whether the 3rd sound (X) was most
like the first (A) or second (B) sound• Chance performance
Pisoni: Experiment I
• Figure 2 (p.1355)– Filled in circles = labeling functions response to 2
end points• Sharp and consistent for some
• Crossover points for the category boundary for 6 of the 8 are not at 0 but are displaced towards the lagging (+50) stimuli– Why?
Pisoni: Experiment I
• Possibly due to limitation on the processing of temporal information or…
• Due to Masking of the high frequency (1500 Hz) by the low (500 Hz)– So, they accounted for that by running a pilot study
• Pilot study (p.1355-56)– Results: they found no shift in boundary location
so…the Limitation on the Processing of Temporal Information is the more like cause of asymmetry
Pisoni: Experiment I
• ABX-discrimination results– Open circles Fig. 2– Categorical-like discrimination– Peaks and troughs– S2 ideal
Pisoni: Experiment I
• Results from ID and ABX:– Categorical perception with non-speech signals – This form of perception is not unique to speech
signals– Removes one positive line of evidence for the
Speech Mode theory
• Questions:– Are the findings due to labeling process brought
about by the training process?– Or is it do to a simpler psychophysical explanation?
Pisoni: Experiment II
• Goal (in order to answer the previously asked questions):– To obtain ABX-discrimination functions before any
training experience (label training)– If peaks in discrimination exist there will be reason
to suspect a psychophysical basis for the observed discrimination functions from E1
Pisoni: Experiment II
• 12 volunteers• Same 11 stimuli used in E1• 2 one-hour sessions held on separate
days (no label training)• 360 ABX trials done ea. day with
feedback• 9 two-step stimuli comparisons were
responded to 80x by ea subject
Pisoni: Experiment II• Results:
– Figure 3 (p. 1357)– 2 patterns shown (except S1 = chance)
• Single peak @ approximately +20 msec• Double peak @ approximately +20 and -20
– Natural categories are present at places along the stimulus continuum marked by narrow regions of high sensitivity (thresholds)
– 3 categories corresponding with the temporal events• Lower tone leading by 20 msec or more (-)• More or less simultaneously within the -20 to +20 msec region• Lower tone lags by 20 msec or more (+)
Pisoni: Experiment II
• These results contrast:– Liberman et al. (1961)– Mattingly et al. (1971)
• The above both found:– Marked differences in discrimination
between speech and non-speech signals– Why?
Pisoni: Experiment II
• The lack of familiarity with the stimuli used (Liberman = synthetic spectograms of /do/ and /to/; Mattingly = 2nd formant transitions were isolated from the rest of the
stimulus pattern) • The absence of any feedback during the
discrimination task• With complex multidimensional signals it may
be difficult for subjects to attend to the relevant attributes that distinguish these stimuli
Pisoni: Experiment II
• Patterns of categorical perception are seen when using speech and non-speech stimuli
Pisoni: Experiment III
• Goal:– To demonstrate that subjects can classify
these same stimuli into three distinct categories whose boundaries occur at precisely these regions on the continuum
Pisoni: Experiment III
• Same training procedure– Except…3 responses instead of 2
• 8 additional subjects were recruited
• Same set of 11 tonal stimuli
• Took place on 2 separate days
Pisoni: Experiment III
• Day 1– Shaping and identification training with the 3
stimuli (-50, 0, +50 msec)– Subjects were free to adopt their own coding
strategies– Immediate feedback was provided
• Day 2– Labeling tests were conducted
Pisoni: Experiment III
• Figure 4 (p. 1358)– All subjects partitioned the stimulus
continuum into three well-defined categories
– Boundaries found at approximately -20 and +20 msec
– Perceptual threshold– Ability to discriminate temporal
differences
Pisoni: Experiment IV
• Goal:– Simultaneous vs. non-simultaneous….– Having the subjects determine whether
there are one or two distinct events at stimulus onset
Pisoni: Experiment IV
• 8 additional volunteers• None had participated previously• Same 11 tonal stimuli• A single 1-hour session• 11 stimuli presented randomly• Told to listen to ea sound carefully and then
determine whether they could hear one or two events at stimulus onset
• No feedback was given
Pisoni: Experiment IV
• Figure 5 (p. 1359)– All subjects showed similar U-shaped functions
with sharp crossover points between categories
• Results:– The presence of 3 natural categories that may be
distinguished by the relative discriminability of the temporal order of the component events
Pisoni Findings
• A perceptual effect for processing temporal order information which may also underlie the perception of voicing distinctions in stop consonants in initial position
• There is a perceptual threshold (consistent with studies done by Hirsh, Hirsh and Sherrick, and Stevens and Klatt) of about 20 msec
Pisoni Findings• We know…that VOT (in terms of onset of voicing)
must be judged in relation to the temporal attributes of other events (release from closure)
• So, these events are ordered in TIME, therefore highly distinctive and discriminable changes will be produced at various regions along the temporal continuum
• Phonological systems apparently have exploited the principle of discriminating discrete attributes (natural categories) during the evolution of language
• In other words, we’ve positioned our phonemes on either side of the natural auditory boundary provided by the threshold
Recommended