Download 2010 - Cornell University

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts

Herbert H. Clark wikipedia, lookup

Psychophysics wikipedia, lookup

Transcript
Syllable stress modulates articulatory planning:
evidence from a stop-signal experiment
Sam Tilsen
University of Southern California
[email protected]
Ladefoged, Silverstein, & Papçun (1973):
"there are some moments in the stream of speech when a speaker would find
it more difficult to interrupt himself than at other moments. Thus it might be
thought likely that a speaker might find it more difficult to interrupt himself
in the middle of a syllable than at the end; and perhaps that interruptions
might be much easier at the end of a word or phrase rather than in the
middle.”
“difficult” : reaction time (RT) to stop-signal
• contrary to hypothesis: no particular part of the sentence where speakers
found it more difficult to interrupt themselves.
• RTs were slower prior to utterance initiation
Syllable stress ?? Articulatory gestures
Unstressed
Shorter duration
Lesser magnitude
Less resistant to coarticulation
More prone to reduction
Stressed
Longer duration
Greater magnitude
More resistant to coarticulation
Less prone to reduction
? Easier to stop
? Harder to stop
(cf. Beckman & Edwards, 1994; Cho, 2002; Cho & McQueen, 2005; Cole et al.,
2007; Crystal & House, 1988; de Jong, 1995)
Simplified schematic model of gestural planning and inhibitory control:
gestural activation
σ-activation
Ed had ed- it- ed
inhibitory
process
Id
cessation
of speech
Ed had ed- it- ed
Id
gestural
activation
The planning activation hypothesis:
Hyp. 1a: Speakers will halt phonation more slowly when signaled just
prior to a stressed syllable than to an unstressed syllable.
The perceptual attention hypothesis:
Hyp. 1b: Speakers will halt phonation more quickly when signaled
during a stressed syllable than during an unstressed syllable.
Evidence from phoneme-monitoring tasks that perceptual attention is heightened during
stressed syllables:
Shields, McHugh, & Martin (1974), phoneme monitoring:
• faster RTs when the target occurred in a stressed syllable as opposed to an unstressed one.
 but this could be due to differences in acoustic cues.
Pitt & Samuel (1990), phoneme monitoring (w/stress-neutral stimuli)
• 24 ms faster RTs in stressed syllables
 but could be due to violation of acoustic expectancy by the "stress-neutral“ stimuli
Quene & Port (2005), Arantes & Barbosa (2006), phoneme monitoring:
• stress facilitates acoustic perception.
BUT the phoneme-monitoring task differs from the stop-signal paradigm in two important respects:
1. Effects of stimuli on attention are unimodal, i.e. effect of stress (an auditory percept) on RT to
auditory stimulus. Possible that perception of a visual stimulus will be unaffected.
2. Context stimulus is externally generated, whereas in the stop-signal paradigm, the context
stimulus (the utterance) is speaker-generated. Because it is pre-planned (and hence expected), it
may have negligible effect on the perception of a response stimulus.
METHOD
Sentence design
sw
Target duration:
Sally saw the men in Roma naming nine alarms
Sal- ly
saw
the
men in
Ro-
ma
na-
ming
nine a-
larms
s
s
w
s
s
w
s
w
s
w
s
w
w
2.3 s
sww Sally has seen that the women in Roma were naming eleven alarms
Sal- ly
has
seen that
the
wo-
men in
Ro-
ma
were na-
ming e-
le-
ven
a-
larms
s
w
s
w
s
w
s
w
w
w
s
w
w
s
w
w
w
s
w
2.9 s
mixed Sally saw that nine men in Roma were naming new mazes
Sal- ly
saw
that
nine men in
Ro-
ma
were na-
ming
new
ma-
zes
s
s
w
s
s
w
w
w
s
s
w
w
s
w
s
Metrical regularity: the degree to which the pattern of stressed and unstressed syllables in
an utterance is consistent. Can be quantified by assuming a sequence of syllables is a firstorder Markov process and calculating the entropy rate H according to the formula:
sw:
sww:
mixed:
0.00
0.67
0.78
Hyp. 2. The metrical regularity of an utterance will affect stop-signal RT.
2.5 s
METHOD
Sally saw the men in Roma naming nine alarms
Sally saw the men in Roma naming nine alarms
• 75% stop-signal trials, 25% catch trials
• feedback on catch trials and stop-signal trials
• stop-task instruction: cut off speech sharply
Participants: 12 native speakers of English (7 women, 5 men)
2 sessions x 2-3 blocks (per sentence) x 24 trials (18 stop-signal)
≈ 90 stop-signal trials / sentence / speaker
METHOD
Dependent variable:
Independent variable:
Perceptual attention:
Planning activation:
stop-signal RT
last pulse of modal voicing
location of the stop-signal relative to utterance…
stress of the syllable in which the stop-signal occurred
stress of the upcoming syllable
proximity of stop-signal to onset of nearest stressed syllable
RESULTS
pre-σstr
post-σstr
•
•
RTs are slower when signaled prior to the
onset of the nearest σstr
 supports planning activation hyp.
(effect sizes are 15-25 ms)
RTs are faster in the sw sentence.
 partly supports hyp. 2: metrical
regularity speeds RT
Test of perceptual attention hypothesis
ANOVA on RT z-score:
SSσstress
stress of the syllable that contains the stop-signal
SSσnext
stress of the upcoming syllable
(F = 0.53, p < 0.47)
(F = 8.89, p < 0.003)
sw
sww
mixed
Syllable-specific correlations between RT and stop-signal
location
ρ
men (in)
Ro(ma)
na(ming)
sw
φstress
-0.37 **
-0.19 **
-0.13
-0.37 **
-0.21 **
-0.15 *
δstress
sww
φstress
δstress
mixed
φstress
δstress
Ro(ma were)
-0.28 **
-0.30 **
men (in)
-0.38 **
-0.37 **
** p < 0.001, * p < 0.05
na(ming) e-0.32 **
-0.32 **
Ro(ma) were
-0.34 **
-0.35 **
le(ven) a-0.02
-0.04
na(ming)
-0.06
-0.06
RESULTS
sw
sww
RESULTS
mixed
MODEL
An extension of the task dynamic model of articulatory phonology (Browman &
Goldstein, 1988, 1990, 2000; Saltzman & Munhall, 1989; Nam & Saltzman, 2003; Saltzman & Byrd,
2003; Saltzman, Nam, Krivokapic, & Goldstein, 2008). In addition to phase coupled gestural
planning oscillators, the extended model (Tilsen, 2009a, 2008) incorporates:
1. syllable- and stress-planning oscillators
2. amplitude coupling between planning systems
MODEL
MODEL
Conclusion and future directions
 speakers halt phonation more slowly when signaled just prior to a stressed syllable than an
unstressed syllable.
1. the planning of syllables modulates articulatory planning and influences the timecourse of
inhibitory processes
2. current models of how stress interacts with articulation do not predict stop-signal effects, but
these effects can be predicted with amplitude-coupling between syllables and gestural
systems. also relates to:
• boundary phenomena & articulatory consequences of stress: increased duration,
gestural magnitude, resistance to coarticulation
• unified account of weight-based stress assignment, extraprosodicity, and
extrametricality
3. revived the stop-signal paradigm as a viable approach to investigating cognitive processes
related to speech.
additional factors in stop-signal RT:
• metrical regularity (familiarity, frequency)?
• speech-rate
• prosodic prominence (e.g. emphatic, contrastive focus)
• morphological and syntactic phrase structure
• spontaneous vs. prepared speech
• articulatory features
• supralaryngeal articulatory kinematics of stopping