parsing acoustic variability as a mechanism for feature abstraction
DESCRIPTION
Parsing acoustic variability as a mechanism for feature abstraction. Jennifer Cole Bob McMurray Gary Linebaugh Cheyenne Munson University of Illinois University of Iowa. www.psychology.uiowa.edu/faculty/mcmurray. Phonetic precursors to phonological sound patterns. - PowerPoint PPT PresentationTRANSCRIPT
Parsing acoustic variability as a mechanism for feature abstraction
Jennifer Cole Bob McMurrayGary Linebaugh Cheyenne MunsonUniversity of Illinois University of Iowa
www.psychology.uiowa.edu/faculty/mcmurray
Phonetic precursors to phonological sound patterns
• Many phonological sound patterns are claimed to have precursors in systematic phonetic variation that arises due to coarticulation
• Assimilation– Vowel harmony from V-to-V coarticulation
(Ohala 1994; Beddor et al. 2001)– Palatalization from V-to-C coarticulation
(Ohala 1994)– Nasal Place assimilation (-mb, -nd, -ŋg) from C-to-
C coarticulation (Browman & Goldstein 1991)
• Assimilation• Epenthesis
– Epenthetic stops from C-C coarticulation: sen[t]se (Ohala 1998)
• Assimilation• Epenthesis• Deletion
– Consonant cluster simplification via deletion from C-C coarticulation: perfec(t) memory
(Browman & Goldstein 1991)
The role of the listener
Phonologization: • when acoustic properties that arise due to
coarticulation are interpreted by the listener as primary phonological properties of the target sound.
• generalization over variable acoustic input that results in a new constraint on sound patterning.
Speaker 3 All vowels
2
3
4
5
6
76.07.08.09.010.011.012.013.014.015.016.0
F2 (Bark)
F1 (B
ark)
iuoe
The role of the listener
• From V-to-V coarticulation …
ɛ
ʌ
ɑ
i
Speaker 3 All vowels
2
3
4
5
6
76.07.08.09.010.011.012.013.014.015.016.0
F2 (Bark)
F1 (B
ark)
iuoe
The role of the listener
• From V-to-V coarticulation … […ɛi…i…]
ɛ
ʌ
ɑ
i
[…ɛɑ…ɑ…]
Speaker 3 All vowels
2
3
4
5
6
76.07.08.09.010.011.012.013.014.015.016.0
F2 (Bark)
F1 (B
ark)
iuoe
The role of the listener
• Perception may yield vowel assimilation […ɛi…i…]
ɛ
ʌ
ɑ
i
[…ɛɑ…ɑ…]i ɑ
Speaker 3 All vowels
2
3
4
5
6
76.07.08.09.010.011.012.013.014.015.016.0
F2 (Bark)
F1 (B
ark)
iuoe
The role of the listener
• But – distinct factors can produce similar variants: […ɛi…i…]
ɛ
ʌ
ɑ
i
[…ɛ ŋ…]
From perception to phonology
• What is the mechanism for mapping from continuous perceptual features to phonological categories?
ɛi mid and highcentral and front-peripheral
ɛɑ mid and low central and back
From perception to phonology
• What is the mechanism for mapping from continuous perceptual features to phonological categories?
ɛi mid and highcentral and front-peripheral
ɛɑ mid and low central and back
The problem: • The perceptual system is confronted with uncertainty due to variation arising from multiple sources.
• Yet, patterns of variation must get associated with individual features of the context vowel (e.g,. high, front) if coarticulation serves as a precursor to phonological assimilation.
• How do lawful, categorical patterns emerge from ambiguous, variable input?
…the lack of invariance problem!
Our claims
• What is the mechanism for mapping from continuous perceptual features to phonological categories?
Our claims: Variability is retained. Acoustic variability is parsed
into components related to the target segment and the local context.
Feature abstraction through parsing. Acoustic parsing provides a mechanism for the emergence of phonological features from patterned variation in fine phonetic detail.
Variability is retained
• Listeners are sensitive to fine-grained acoustic variation. (Goldinger 2000; Hay 2000; Pierrehumbert 2003)
Variability is retained, not discarded
Consistent with exemplar models of the lexicon, phonetic detail is encoded and stored, and can inform subsequent categorization of new sound tokens.
• Variability due to coarticulation is subtracted to identify the “underlying” target sound.
(Fowler 1984; Beddor et al. 2001, 2002; Gow 2003)
Variability is retained
• Variability is useful for the identification of sounds in contexts of coarticulation.
• The perceptual system uses information about variability to identify a sound and its context, in parallel.
• Variability due to coarticulation is exploited to facilitate perception.
-- Listeners benefit from the presence of anticipatory coarticulation in predicting the identity of the upcoming sound.
(Martin & Bunnell 1982; Fowler 1981, 1984; Gow 2001, 2003; Munson, this conference)
Variability and perceptual facilitation
Perceptual facilitation from V-to-V coarticulation is expected to occur only if:
• The effects of coarticulation are systematic—an influencing vowel conditions a consistent acoustic effect on target vowels;
• The listener can recognize coarticulatory effects on the target vowel;
• The listener can isolate the effects of context vowel from other sources of variation, and attribute those effects to the context vowel.
Feature abstraction through parsing
More specifically…under coarticulation of vowel height and backness,
• The listener must parse out the portion of the variance in F1 and F2 that is due to coarticulation, and base their perception of the target vowel on the residual values.
• Acoustic parsing isolates the effects of context vowel on F1 and F2.
Feature abstraction through parsing
• The parsed acoustic variance defines features of the context vowel, over which new generalizations can be formed. phonologization
[ɛ] + [i] [ɛ] + [i]
[ɛ] + [high]
ɛi
Feature abstraction through parsing
• The parsed acoustic variance defines features of the context vowel, over which new generalizations can be formed. phonologization
[ɛ] + [i] [ɛ] + [i]
[ɛ] + [high]
phonologized to [i]
i
Feature abstraction through parsing
• The parsed acoustic variance defines features of the context vowel, over which new generalizations can be formed. phonologization
Question: Why phonologization? If target and context vowels can both be identified from the fine phonetic detail…. What’s the force driving phonologization?
Testing the model
The acoustic parsing model of speech perception requires that there is a robust and systematic pattern of acoustic variation from V-to-V coarticulation.
This paper: we present supporting evidence from an acoustic study of coarticulation.
• We examine a range of V-to-V coarticulatory effects in VCV contexts that cross a word boundary, where coarticulation cannot be attributed to lexicalized phonetic patterns.
Key Questions
Extent of phenomenon• Does V-to-V coarticulation cross word boundaries?• Does V-to-V coarticulation affect both F1 and F2?• Relative strength of V-to-V effects vs. other forms of
coarticulation?
Usefulness of phenomenon• How could V-to-V effects translate to perceptual
inferences?• Is the information by V-to-V coarticulation different
when other sources of variation are explained?
Methods
Target vowels: ɛ ʌ
Measure coarticulation
Context vowels: i æ ɑInduce Coarticulation
i
æ ɑʌɛ
• /u/ excluded from contexts (rounded + fronted)• intervening consonant varied in
- place (labial, coronal, velar)- voicing- /ɛg/ excluded (tends to be raised)
Methods
bed actor tech afternoon web addicteagle evening ecologistevergreen elevator educatorostrich Oxygen Offer
wet Afro deck alligator step AdmiralEaster Bunny easter basket eastEskimo elephant exitOxen octopus obstacle
mud apple bug astronaut pub advertisementeater evil easelumpire underwear undergradobservation optician operator
cut abdomen duck athlete cup appetizerevenly eating eavesdroppingonion usher ovenOlive officer occupant
Methods
Methods• 10 University of Illinois students.• 48 phrases x 3 repetitions.• Sentences embedded in neutral carrier sentences
/ɛ/ He said ‘_______’ all the time/ʌ/ I love ‘_______’ as a title
Coding• F1, F2, F3
- Converted to Bark for analysis• LPC (Burg Method)• Outliers / misproductions inspected by hand
Analysis
Target x Voicing x Context
F1 F2Voicing p=.033 p=.001Target p=.005 p=.001Context p=.001 p=.001Interactions n.s. n.s.
Target x Place x Context
F1 F2Place n.s. p=.001Target p=.01 p=.001Context p=.001 p=.001Interactions some some
V-to-V coarticulation crosses word boundaries.
Clear effects of coarticulatory contexton both F1 and F2.
Analysis
400
500
600
700
800
900
1000100015002000
F2 (Hz)
F1 (H
z)
æi
ɑSame
High
Low
Front Back
Male
Female
A lot of unexplained variance…
• How does the perceptual system “get to” the V-to-V coarticulation?
• How useful is V-to-V coarticulation?
• Does accounting for other sources of variance in the signal improve the usefulness of V-to-V?
Strategy
Need to systematically account for sources of variance prior to evaluating V-to-V coarticulation.
F2
ɛʌ
1431 hz 1801 hz
iɑ ?
ɑ-coarticulated ɛ?or
i-coarticulated ʌ?
Strategy
F2
ɛʌ
1431 hz 1801 hz
i?
A slightly i-coarticulated ɛ? or
A really i-coarticulated ʌ?
Need to systematically account for sources of variance prior to evaluating V-to-V coarticulation.
Strategy
F2
ɛʌ
1431 hz 1801 hz
iɑ ?
If you knew the category…If ʌ, then expect iIf ɛ then expect ɑ
? - ʌ: Positive (more i-like)? - ɛ: Negative (more ɑ-like)
F2? – F2category = coarticulation direction
Need to systematically account for sources of variance prior to evaluating V-to-V coarticulation.
Strategy
Target – F2? = coarticulation direction
F2
ɛʌ
1431 hz 1801 hz
iɑ ?
F2
ɛʌ
1431 hz 1801 hzF2
ɛʌ
1431 hz 1801 hz
iiɑɑ ??
Strategy: 1) Compute mean of a source of variance 2) Subtract that mean from F1/F23) Residual is coarticulation direction.4) Repeat for each source of variance (speaker, target
vowel, place, voicing).
Strategy
F1predicted = 1 * target + 0
If target = 0 for /ʌ/ and 1 for /ɛ/… ʌ) F1predicted = 1 * 0 + 0 Mean /ʌ/ = 0 ɛ) F1predicted = 1 * 1 + 0 Mean /ɛ/ = 0 + 1
Hierarchical Regression can do exactly these things.
1) Compute mean of a source of variance
Strategy
Hierarchical Regression can do exactly these things.
1) Compute mean of a source of variance.2) Subtract that mean from F1/F23) Residual is coarticulation direction.
ResidualF1actual - F1predicted = F1actual - (1 · target + 0)
ʌ) Residtarget = F1actual - 0ɛ) Residtarget = F1actual - (0 + 1)
Strategy
Hierarchical Regression can do exactly these things.
1) Compute mean of a source of variance.2) Subtract that mean from F1/F23) Residual is coarticulation direction.4) Repeat for each source of variance (speaker, target
vowel, place, voicing).
Residtarget = 2 * Place + 0
Residplace = 3 * Voicing+ 0
F1 = 0 * Target+ 0
Residvoicing = 4 * V-to-V + 0
Strategy
Construct a hierarchical regression to systematically account for known sources of variance from F1 and F2
• Speaker• Target vowel• Place (intervening C)• Voicing (intervening C)• Interactions between target, place & voicing
After partialing out these factors, how much variance does vowel context (V-to-V) account for?
3
4
5
6
7
889101112131415
F2 (Bark)
F1 (B
ark)
iæɑSame
Regression F2
1) Raw DataMale
Female
Regression F2
1) Raw Data
Partialed Out2) Subject
-1.5
-1
-0.5
0
0.5
1
1.5-3-2-1012
F2 Resid (Bark)
F1 R
esid
(Bar
k)i
æɑSame
ʌɛ
Regression F2
1) Raw Data
Partialed Out2) Subject3) Target Vowel
-1.5
-1
-0.5
0
0.5
1-2-1012
F2 Resid (Bark)
F1 R
esid
(Bar
k)iA
oSame
Regression F2
1) Raw Data
Partialed Out2) Subject3) Target Vowel4) Consonant
-1
-0.5
0
0.5
1-1-0.500.51
F2 Resid (Bark)
F1 R
esid
(Bar
k)iæɑSame
Regression F2
1) Raw Data
Partialed Out2) Subject3) Target Vowel4) Consonant5) Interactions
-1
-0.5
0
0.5
1-1-0.500.51
F2 Resid (Bark)
F1 R
esid
(Bar
k)iæɑSame
t
Regression F1
Step Variables R2change P
1 Subjects (10) .824 ***
Regression F1
Step Variables R2change P
1 Subjects (10) .824 ***2 Vowel .009 ***3 Voicing .018 ***4 Place (2) .003 **
Regression F1
Step Variables R2change P
1 Subjects (10) .824 ***2 Vowel .009 ***3 Voicing .018 ***4 Place (2) .003 **5 Vowel x Voicing .000 -6 Vowel x Place (2) .002 *7 Voicing x Place (2) .012 ***
Total R2=.884
Post-hoc analysis: height only.
Regression F1
Step Variables R2change P
1 Subjects (10) .824 ***2 Vowel .009 ***3 Voicing .018 ***4 Place (2) .003 **5 Vowel x Voicing .000 -6 Vowel x Place (2) .002 *7 Voicing x Place (2) .012 ***8 ContextVl (3) .012 ***9 ContextVL interactions (12) .003 -
Total R2=.884
Post-hoc analysis: height only.
Regression F2
Step Variables R2change P
1 Subjects (10) .409 ***
Regression F2
Step Variables R2change P
1 Subjects (10) .409 ***2 Vowel .412 ***3 Voicing .034 ***4 Place (2) .050 ***
Regression F2
Step Variables R2change P
1 Subjects (10) .409 ***2 Vowel .412 ***3 Voicing .034 ***4 Place (2) .050 ***5 Vowel x Voicing .008 ***6 Vowel x Place (2) .015 ***7 Voicing x Place (2) .004 ***
Regression F2
Step Variables R2change P
1 Subjects (10) .409 ***2 Vowel .412 ***3 Voicing .034 ***4 Place (2) .050 ***5 Vowel x Voicing .008 ***6 Vowel x Place (2) .015 ***7 Voicing x Place (2) .004 ***8 ContextVl (3) .008 ***9 ContextVL interactions (12) .001 -
Total R2=.940
Post-hoc analysis: height + backness.
Regression Summary
Progressively accounting for variance is powerfulF1: 88% of varianceF2: 94% of variance
using only known sources of variance
V-to-V coarticulation is readily apparent when other sources of variance are explained.
How useful would this be?
Effect of V-to-V coarticulation has a similar size to place/voicing effects.
Predicting Vowel Identity
Multinomial Logistic Regression (MLR)• Classification algorithm• Predict category membership from multiple variables.• Categories do not have to be binary
Samei
Sameɑ
Sameæ
Context Vowel
Predicting Vowel Identity
Samei
Sameɑ
Sameæ
Context Vowel
Samei
Samei
Sameɑ
Sameɑ
Sameæ
Sameæ
Context Vowel
• Assumes optimal listener.
• Computes % correct.• How much well could a
listener do under ideal circumstances with information provided.
Multinomial Logistic Regression (MLR)• Classification algorithm• Predict category membership from multiple variables.• Categories do not have to be binary
Predicting Vowel Identity
0
10
20
30
40
50
60
i ɑ æ Same
Vowel
% C
orre
ct
Partialed outSubjectVowelPlaceVoicingInteractions
Model does quite well at predicting all vowels but the identity.
Predicting Vowel Identity
-12
-10
-8
-6
-4
-2
0
2
4
6-42814202632
F2 (Z)
F1 (Z
)
ʌ-i
ʌ-æʌ-ɑ i
æ
ɑ
-10
-8
-6
-4
-2
0
2
4
6-18-12-6061218
F2 (Z)
F1 (Z
)
ɛ-i
ɛ-æɛ-ɑ i
æ
ɑ
Predicting Vowel Identity
Does partialing out other sources of variance improve the utility of V-to-V coarticulation?
- Use linear regression to partial out variance. - Use F1, F2 residuals to predict vowels.
FULL: Partial out everythingRAW: No parsingSPEAKER: Partial out speaker variation only. Assume
speaker normalization, but no interactions between consonant, or vowel and V-to V.
VOWEL: Partial out effects of everything heard at the target vowel (speaker + target)
NO-SPKR: Assume no normalization, but interactions between consonants.
Predicting Vowel Identity
FULL: about 4% better than others.VOWEL: parsing out consonant may not be necessarySPEAKER: Effect of speaker and phonetic cues similar.RAW: V-to-V not useful without some parsing.
2527293133353739414345
FULL VOWEL SPEAKER NO-SPKR RAW
% C
orre
ct
Predicting Vowel Identity
2) Regressively compensate for consonant coarticulation
targetvowel consonant context
vowelpreceding
context
3) Use residuals to predict context vowel
1) Parse out speaker effects on target
Suggests a 3-stage parsing process to maximally useV-to-V modifications.
Key Questions
Extent of phenomenon• Word boundaries?• Both F1 and F2?• Relative strength of V-to-V effects?
Usefulness of phenomenon• Perceptual inferences?• Parsing our variability?
Summary: Extent
• Clear evidence for V-to-V coarticulation across word boundaries—not lexicalized.
• V-to-V in both formants (height + backness).
• Strength is similar to that of place and voicing.
• Known sources of variance (speaker, vowel, consonant, V-to-V) can account for most of the variability in vowel production.
- Problem of lack of invariance?
- Identifying multiple categories at once may be easier than identifying one.
Summary: Usefulness
• Idealized listener (+ parsing) could identify upcoming vowel at 40% correct given only V-to-V coarticulation.
- Near 50% for /i/ and /ɑ/
• Parsing dramatically improves predictive power of V-to-V coarticulation
• Do you need perfect categorization of variance sources (e.g. speaker, target vowel, voicing…)?
- Imperfect categorization enhances need for multiple cues.- Simultaneously evaluating multiple features (e.g. V1, C, V2)
yields correct parse.
• How do you determine the order of parsing?- Temporal order of information arrival?
Future Directions
How do you identify the components you will be parsing?• See Toscano poster.
Does the model actually describe perception?• Parsing is a temporal process.• Visual world paradigm to time-course of processing
(e.g. McMurray, Clayards, Tanenhaus, in prep; McMurray, Tanenhaus & Aslin, 2002; McMurray, Munson & Gow, submitted).
Parsing as part of word recognition.• Lexical structure can contribute to inferences.• Interactive activation models (McClelland & Elman,
1986) could implement this.
Conclusions
Where do features come from?• Emerge out of progressively accounting for sources of
variance from signal.• Any “chunk” (segment) of the input can provide
multiple features.• Speaker normalization may work by same process.
Why phonologize?• Eliminates one step of parsing.
How does the system balance need for features with utility of fine-grained detail?
• Features provide tag to parse variance and utilize continuous detail.