730 likes | 849 Views
From perception to categorization and back again… 8/29. 1. A Short Reprise… Categories Treating variable stimuli equivalently… Exemplars A particular perceptual occurrence of a category. 2. A Short Reprise… Perceptual Categorization Continuous perceptual input Discrete Categories. 3.
E N D
A Short Reprise… • Categories • Treating variable stimuli equivalently… • Exemplars • A particular perceptual occurrence of a category. 2
A Short Reprise… Perceptual Categorization Continuous perceptual input Discrete Categories 3
boundary Within-category variation Between category variation • Exemplars Vary • Category Boundaries • Between vs. Within Category variation • … on same dimension Cup Bowl 4
Example: Speech Categories Why study speech? • Highly meaningful categorization. • Difficult categorization: lots of variation between speakers, speaking rates, phonation types • Difficult categorization: must happen very fast. • We understand the input well (from linguistics & phonetics) • Perhaps uses same principles as other types of categories. 5
Example: Speech Categories • What’s discrete? • What’s continuous? • How is this mapping formed? 6
Discrete Speech Categories • What is discrete? • Phonemes: Minimal contrastive unit. • Features: Voicing, Place, Manner • What is the relevant continuous perceptual detail? • How does this mapping take place? • What are the computations? 7
frequency amplitude Continuous Speech Cues 1) What’s continuous? Not correlated with categories. More abstract cues? Frequency & Amplitude? 8
formants frequency time At what frequencies is there energy at what time? amplitude frequency Speech is also temporal: changes over time. 9
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 1 10
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 2 11
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 3 12
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 4 13
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 5 14 .25
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 6 15
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 7 16
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 8 17
Continuous Speech Cues VOT: Time difference between opening of oral release (lips/tongue) and onset of voicing. Voicing: B vs. P D vs. T K vs. G 9 18
Continuous Speech Cues Formant Transition: Slope of formant frequency at word onset. Place: D vs. B vs. G S vs. Th vs. Sh 19
Continuous Speech Cues Formant Transition: Slope of formant frequency at word onset. Place: D vs. B vs. G S vs. Th vs. Sh 20
2nd Formant Transition Voice Onset Time Continuous cues. How are they categorized into features and phonemes? How can we test hypothesized cues? Speech Continuum. 21
Ba Da Ga Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Systematically vary 2nd formant from one end point to another… How do subjects identify each token? 22
Ba Da Ga Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Synthetic speech: Pattern Playback Random order of tokens 3AFC 25 Reptitions x step x subject 23
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Very consistent identification. 24
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Does categorization affect perception? continuous input to discrete categories 25
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Does categorization affect perception? Test discrimination within vs. between categories… ABX Task: Match third token to 1st or 2nd. 141 Within Same distance on continuum. Between 477 26
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) • Randomized order. • Counterbalanced: X=A | X=B • ABX before and after labeling. • Does categorizing sounds affect discrimination? • Predicted scores based on identification only. 27
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) 28
Within-category Between-category Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Discrimination is not uniform across the continuum. Some comparisons are really hard. Some are easy. 29
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Categorical Perception: 1) Sharp categorization. 2) Discrimination poor within category. 3) Discrimination good between categories. 30
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) 1 2 3 4 5 6 7 8 9 10 Continuous cues can be thought of as a dimension or psychological space. 3 - 2 <<<< 6 - 5 In a normal space, distance is the same everywhere. CP warps this space. 31
tall short tall short Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Categorical Perception 32
three-step two-step one-step Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) If perception was continuous: discrimination should only depend on distance in continuous measure. D G Discrimination = f( distance) 33
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) If perception was categorical: discrimination should only depend on distance in continuous measure. three-step two-step one-step D G Discrimination = f( location ) 34 .5
Categorical Perception: Liberman, Harris, Hoffman & Griffith (1957) Actual Results: Discrimination is a combination of both. three-step two-step one-step D G Discrimination = f( distance and location ) True CP: = f( location ) 35 .5
Is discrimination truly categorical? 1) Some effect of distance, but didn’t assess many “distances” 2) ABX Task is hard: converging evidence from 2nd task. 36
1) Assessing Distances Psychophysics: assess discrimination at a large numbers of stimulus differences. 100% Discrimination Gives complete picture of discrimination. 0% 0 300 600 900 1200 1500 1800 2100 2400 2700 3000 Frequency Difference 37
1) Assessing Distances Liberman et al assessed only 3 distances! 38
2) ABX Task and Difficulty ABX task requires listener to remember 3 things before making decision. • STM more efficient if items are chunked (G. Miller) • Syllables more efficient representation than raw acoustic form. • Subjects more likely to remember syllables. • Comparisons between syllables should be easier… • …Categorical Perception! 39
2) ABX Task and Difficulty Does the 3rd color match the 1st or 2nd? Task makes a huge difference. 1 2 ABX 40
2) ABX Task and Difficulty Does 1st color match left or right? Task makes a huge difference. 3 X[AB] 41
2) ABX Task and Difficulty Does 1st color match left or right? Task makes a huge difference. 3 X[AB] 42
2) ABX Task and Difficulty Are the two colors same or different? Task makes a huge difference. 5 6 AX 43
2) ABX Task and Difficulty Are the two colors same or different? Task makes a huge difference. 7 [AX] 44
2) ABX Task and Difficulty Are the two colors same or different? Task makes a huge difference. 8 [AX] 45
continuous categorical Within-category Discrimination? • Carney, Widden, Viemeister (1977) • Examine CP with range of tasks. • Examine CP with a range of stimuli. Category boundary 100% Discrimination Sample Data 0% 0 60 10 20 30 40 50 -40 -50 -20 -10 Comparison Base Voice Onset Time 46
Step 1: compute category boundaries. 2 AFC Identification. 47
Do we see within-category discrimination in a range of tasks? • Experiment 1 • Same/Different Task: YES • Training: Even better. 48
Do we see within-category discrimination in a range of tasks? • Experiment 2 • Oddity XAA, AXA or AAX YES • 3steps (30ms) between tokens. 49
Do we see within-category discrimination in a range of tasks? • Experiment 3 • Same different with fixed anchor: VERY GOOD 50