Individual differences in phoneme categorization Effie Kapnoula, - PowerPoint PPT Presentation

Individual differences in phoneme categorization Effie Kapnoula, Bob McMurray, Eunjong Kong, Matthew Winn, & Jan Edwards 19th Mid-Continental Phonetics & Phonology Conference

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant frequencies) and the perceived phoneme 2 Hillenbrand, Getty, Clark & Wheeler, 1995

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant frequencies) and the perceived phoneme • One solution: categorical perception 3

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant frequencies) and the perceived phoneme • One solution: categorical perception par bar par bar 1 2 3 4 5 6 7 4

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant frequencies) and the perceived phoneme • One solution: categorical perception par +Simple solution +Fast commitment bar par bar 1 2 3 4 5 6 7 5

Two alternative forced choice (2AFC) 6 Werker & Tees, 1987; Joanisse et al, 2000; López-Zamora et al, 2010

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant transitions) and the perceived phoneme • One solution: categorical perception par +Simple solution +Fast commitment bar par bar 1 2 3 4 5 6 7 • Alternative: gradient perception par bar par bar 1 2 3 4 5 6 7 7

The problem of lack of invariance • There is no one-to-one relation between a sound (i.e. formant transitions) and the perceived phoneme • One solution: categorical perception par +Simple solution +Fast commitment bar par bar 1 2 3 4 5 6 7 • Alternative: gradient perception par +Flexibility +Late commitment bar par +Keep useful within-category information bar 1 2 3 4 5 6 7 8

Gradiency in speech perception • Evidence for gradiency from eye-movements Response = Response = 0.08 Competitor Fixations 0.07 Looks to 0.06 0.05 Looks to 0.04 Category 0.03 Boundary 0.02 0 5 10 15 20 25 30 35 40 VOT (ms) 9 McMurray, Tanenhaus & Aslin (2002)

Two alternative forced choice (2AFC) • Is gradiency good or bad for speech perception? ? 10 Werker & Tees, 1987; Joanisse et al, 2000; López-Zamora et al, 2010

Gradiency in speech perception • Measuring gradiency: Visual analog scaling (VAS) task bull pull bull 11

Gradiency in speech perception 12 Kong, E. J., & Edwards, J. (2011)

Gradiency in speech perception 13 Kong, E. J., & Edwards, J. (2011)

Summary and aims • Summary points:  Listeners are capable of gradient categorization of phonemes  The VAS task allows for this gradiency to be expressed in participants’ responses 14

Summary and aims • Summary points:  Listeners are capable of gradient categorization of phonemes  The VAS task allows for this gradiency to be expressed in participants’ responses • Where does gradiency come from? Is it good or bad for speech perception? 15

Summary and aims • Summary points:  Listeners are capable of gradient categorization of phonemes  The VAS task allows for this gradiency to be expressed in participants’ responses • Where does gradiency come from? Is it good or bad for speech perception?  Establish a way of quantifying gradiency via the VAS task 16

Summary and aims • Summary points:  Listeners are capable of gradient categorization of phonemes  The VAS task allows for this gradiency to be expressed in participants’ responses • Where does gradiency come from? Is it good or bad for speech perception?  Establish a way of quantifying gradiency via the VAS task 1. Investigate possible sources of gradiency (e.g. executive function) 2. Link gradiency to multiple cue use 3. Examine whether gradiency is good or bad for speech perception 17

Method • Stimuli: labial alveolar Real words bull-pull den-ten Nonwords buv-puv dev-tev CVs buh-puh deh-teh • Seven (7) VOT steps (primary cue) and five (5) F0 steps (secondary cue) 5 4 F0 steps 3 2 1 1 2 3 4 5 6 7 VOT steps 18

Method • Stimuli: labial alveolar Real words bull-pull den-ten Nonwords buv-puv dev-tev CVs buh-puh deh-teh • Seven (7) VOT steps (primary cue) and five (5) F0 steps (secondary cue) • Tasks: pull bull • Visual analog scaling (VAS) task bull pull • Two alternative forced choice (2AFC) 19

Method • Additional tasks:  Trail making task (cognitive flexibility)  N-Back task (working memory) non-speech cognitive processes  Flanker task (inhibition) 20

Method • Additional tasks:  Trail making task (cognitive flexibility)  N-Back task (working memory) non-speech cognitive processes  Flanker task (inhibition)  AZ-bio (sentences in babbling noise - 1:1 STN ratio) 21

Method • Additional tasks:  Trail making task (cognitive flexibility)  N-Back task (working memory) non-speech cognitive processes  Flanker task (inhibition)  AZ-bio (sentences in babbling noise - 1:1 STN ratio) • Participants: 130 undergraduates at the U of Iowa 22

Results 23

Results 40 40 Sub 8 Sub 7 30 30 20 20 10 10 0 0 0 10 20 30 40 50 60 70 80 90100 0 10 20 30 40 50 60 70 80 90100 40 40 Sub 9 Sub 68 30 30 20 20 10 10 0 0 24 0 10 20 30 40 50 60 70 80 90100 0 10 20 30 40 50 60 70 80 90100

Results: Quantifying gradiency 25

Results: Quantifying gradiency • Extracting gradiency from VAS data 5 5 4 F0 steps 4 F0 steps 3 3 2 2 1 1 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT steps VOT steps s F0 steps F0 steps s θ θ VOT steps VOT steps 26

Results: Quantifying gradiency • Extracting gradiency from VAS data 5 5 4 F0 steps 4 F0 steps Steep s slope 3 3 2 2 categorical 1 1 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT steps VOT steps Shallow s slope s F0 steps F0 steps s gradient θ θ VOT steps VOT steps 27

Results: Quantifying secondary cue use • Extracting F0 use from 2AFC data 1 p 0.8 2AFC response 0.6 F0 = 90 hZ 1 0.4 F0 = 125 hZ 2 0.2 b 0 1 2 3 4 5 6 7 VOT step 28

Results 29

Results: Stimulus and place effects VAS response p CVs Real words 100 b 2AFC response p 1 1 b 2AFC response p 80 0.8 0.8 60 0.6 0.6 NP 0.4 40 0.4 NW RW 0.2 0.2 20 0 0 b 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT step VOT step VOT step Nonwords p 1 b 2AFC response p 100 0.8 VAS response 80 90hZ 0.6 60 125hZ alv 0.4 40 lab 20 0.2 b 0 0 30 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT step VOT step

Results: Stimulus and place effects VAS response p CVs Real words 100 b 2AFC response p 1 1 b 2AFC response p 80 0.8 0.8 60 0.6 0.6 F<1 NP 0.4 40 0.4 NW RW 0.2 0.2 20 F<1 0 0 b 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT step VOT step VOT step Nonwords p 1 b 2AFC response p 100 0.8 VAS response 80 90hZ 0.6 60 125hZ alv 0.4 40 lab 20 0.2 b 0 0 31 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT step VOT step

Results: Place differences in F0 use F(1,250) = 27.8, p < 0.001 Labials Alveolars b 2AFC response p 1 1 0.8 0.8 0.6 0.6 1 90hZ 0.4 0.4 125hZ 5 0.2 0.2 0 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 VOT step VOT step 32

Results Do individual differences in gradiency derive from differences in 1. general cognitive function ? 33

Results Do individual differences in gradiency derive from differences in 1. general cognitive function ? gradient gradient gradient categorical categorical categorical 34  EF measures did not account for a statistically significant amount of variance in VAS slope, F(3,108)=1.75, p=.162, or F0 use, F<0

Results Do individual differences in gradiency derive from differences in 1. general cognitive function ?  EF measures did not account for a statistically significant amount of variance in VAS slope, F(3,108)=1.75, p=.162, or F0 use, F<0  Speech perception processes may be played out on a different level of processing than higher cognitive processes, such as working memory 35

Results Do individual differences in gradiency derive from differences in 1. general cognitive function ? Are individual differences in gradiency linked to multiple cue use ? 2. 36