SEPARATION OF COOCCURRING SYLLABLES: SEQUENTIAL AND SIMULTANEOUS GROUPING or CAN SCHEMATA OVERRULE P

1
SEPARATION OF CO-OCCURRING SYLLABLES SEQUENTIAL
AND SIMULTANEOUS GROUPINGorCAN SCHEMATA
OVERRULE PRIMITIVE GROUPING CUES IN SPEECH
PERCEPTION?

• William J. Barry, Georg Meyer Jacques
  Koreman

Institute of Phonetics, University of the
Saarland, Saarbrücken, Germany Dept. of
Communication and Neuroscience, Keele University,
Keele, UK
29th European Mathematical Psychologists Group
meeting (Keele, UK)28 August - 2 September 1998
2
OVERVIEW

• Introduction
• Primitives and schemata in ASA
• Double-vowel experiments
• Utterances
• Linguistic primitives and schemata
• Experiment 1
• Experiment 2
• Conclusions

3
PRIMITIVES AND SCHEMATA
SCHEMATA
PRIMITIVES
4
PRIMITIVES AND SCHEMATA
SCHEMATA
PRIMITIVES

• On- and offset (time domain)
• F0 and harmonicity (frequency domain)

5
DOUBLE VOWELS
SCHEMATA
STYLISEDSPECTROGRAM
PRIMITIVES

• On- and offset (time domain)

6
DOUBLE VOWELS
SCHEMATA
STYLISEDSPECTROGRAM
PRIMITIVES

• F0 and harmonicity (frequency domain)

7
UTTERANCES
SCHEMATA?
PRIMITIVES?
8
LINGUISTIC PRIMITIVES
SCHEMATA
ACOUSTIC CUES
9
LINGUISTIC SCHEMATA
SYLLABLES
ACOUSTIC CUES
10
LINGUISTIC SCHEMATA
PHONOLOGICAL FEATURES
ACOUSTIC CUES
11
LINGUISTIC SCHEMATA (1)
/al/
/en/
alveolar
nasal
PRIMITIVES
synthesised al
synthesised en
12
LINGUISTIC SCHEMATA
LINGUISTIC SCHEMATA (2)
/al/
/en/
/al/
/en/
PRIMITIVES
PRIMITIVES
synthesised al
synthesised en
13
EXPERIMENT 1
Stimuli pairs of simultaneous synthetic VC
syllables
/e/ /a/ /o/
/l/ /n/
Subjects 10 German listeners who were able to
identify more than 95 of the individual
synthetic VC syllables correctly
14
EXPERIMENT 1
Task identify the pair of synthetic VC
syllables
/e/ /a/ /o/
/l/ /n/

• Primitives available to the listeners
• 40-ms VC place transition with formants going
  from vowel to consonant target
• nasality cue starting early in vowel

15
EXPERIMENT 1
Phonetic background to the acoustic cues
16
EXPERIMENT 1
Task identify the pair of synthetic VC
syllables

• Inconsistent stimuli were also presented in which
• the nasalised vowel was linked with /l/
• Non-nasalised vowel was linked with /n/
• leading to conflicting nasality and place cues

BUT
17
EXPERIMENT 1
Task identify the pair of synthetic VC
syllables
We say a VC pair is identified correctly if the
place transitions are used for the decision. In
the case of inconsistent stimuli this means that
the nasality cue must be ignored!
NOTE
18
EXPERIMENT 1
Question 1 Do listeners use the nasality cue?
19
EXPERIMENT 1
Question 1 Do listeners use the nasality cue?
Answer A No.
Predicted experimental result No difference
between consistent and inconsistent stimuli.
20
EXPERIMENT 1
Question 1 Do listeners use the nasality cue?
Answer A No.
Answer B Yes.
Predicted experimental result Consistent stimuli
better identified than inconsistent ones.
21
EXPERIMENT 1
Question 1 Do listeners use the nasality cue?
50
45
40
35
30
25
20
15
10
5
0
consistent
inconsistent
CONDITION
Significant difference between consistent and
inconsistent condition (plt0.01)
22
EXPERIMENT 1
Question 1 Do listeners use the nasality cue?
Answer A No.
Answer B Yes.
Corresponding experimental result Consistent
stimuli better identified than inconsistent
ones. Whether listeners use nasality as a
primitive or schema is unclear.
23
EXPERIMENT 1
Question 2 Do listeners map place cues onto
phonetic feature schema?
24
EXPERIMENT 1
Question 2 Do listeners map place cues onto
phonetic feature schema?
Answer A No.
Predicted experimental result Given the
identification rate of the vowel, the distinction
in the identification of the consonant is above
chance. Reason the acoustically different place
transitions are used to link the vowel with the
correct consonant in VC schema.
25
EXPERIMENT 1
Question 2 Do listeners map place cues onto
phonetic feature schema?
Answer A No.
Answer B Yes.
Predicted experimental result Given the
identification rate of the vowel, the distinction
in the identification of the consonant is at
chance level. Reason the acoustically different
place transitions are mapped onto the same place
schema (both cons. are alveolar) and cannot
contribute to link the vowel with the correct
consonant.
26
EXPERIMENT 1
Question 2 Do listeners map place cues onto
phonetic feature schema?
100
90
80
VC pair correct
70
vowel pair correct
60
50
VC pair correct
40
30
20
10
0
consistent
inconsistent
CONDITION
Identification of VC pairs given correct vowel
pair is at chance level
27
EXPERIMENT 1
Question 2 Do listeners map place cues onto
phonetic feature schema?
Answer A No.
Answer B Yes.
Corresponding experimental result Given the
identification rate of the vowel, the distinction
in the identification of the consonant is at
chance level. Reason the acoustically different
place transitions are mapped onto the same place
schema (both cons. are alveolar) and cannot
contribute to link the vowel with the correct
consonant.
28
EXPERIMENT 1
CONCLUSIONS

• Listeners use acoustic continuity cues to link
  speech sounds lower VC identification rate for
  inconsistent than for consistent stimuli shows
  that the nasality cue is used.
• Nasality cue may be a primitive or a schema.
• Listeners map the place-of-articulation
  primitives onto a schema linking of the vowel
  with /n/ or /l/ is at chance level despite
  acoustically different place transitions, because
  the (phonetically important) place transitions
  are mapped onto the same schema for the two
  consonants.

29
EXPERIMENT 2
Stimuli the same pairs of simultaneous synthetic
VC syllables, but this time also at different F0s
Subjects 10 German listeners who were able to
identify more than 95 of the individual
synthetic VC syllables correctly
30
EXPERIMENT 2
Task identify the pair of synthetic VC
syllables

• Primitives available to the listeners
• 40-ms VC place transition with formants going
  from vowel to consonant target
• nasality cue starting early in vowel
• F0

31
EXPERIMENT 2
Note ? F0 provides primitive melody and
harmonicity link between V and C ? F0 cannot
trigger a schema at the level of the stimulus
offered for identification (VC syllable),
although it may trigger a schema at higher levels
of prosodic-phonological structuring, i.e. in
intonation
32
EXPERIMENT 2
Question Do listeners use the F0 cue more than
they do nasality?
33
EXPERIMENT 2
Question Do listeners use the F0 cue more than
they do nasality?
Answer A No.
Predicted experimental result Consistent stimuli
better identified in the different-F0 condition,
but distinction between consistent and
inconsistent stimuli maintained or even enhanced.

34
EXPERIMENT 2
Question Do listeners use the F0 cue more than
they do nasality?
Answer A No.
Answer B Yes.
Predicted experimental result F0 cue overrides
the distinction between consistent and
inconsistent stimuli in the different-F0
condition.
35
EXPERIMENT 2
Question Do listeners use the F0 cue more than
they do nasality?
PAIRS CORRECT
60
55
50
45
40
35
30
25
20
same F0 different F0
Significant difference consistent/inconstent only
for same F0 (plt0.01)
36
EXPERIMENT 2
Question Do listeners use the F0 cue more than
they do nasality?
Answer A No.
Answer B Yes.
Corresponding experimental result F0 cue
overrides the distinction between consistent and
inconsistent stimuli in the different-F0
condition.
37
EXPERIMENT 2
CONCLUSIONS

• F0 presents a strong continuation cue linking the
  V and C
• The F0 primitive cue overrides the nasality cue

38
very tentative... HYPOTHESES

• Primitives trigger schemata if possible

Place transitions in experiment 1 triggered same
place schema and therefore could not override
nasality cue. Problem we have no proof that
nasality triggers a schema.
39
very tentative... HYPOTHESES

• Primitives trigger schemata if possible

Place transitions in experiment 1 triggered same
place schema and therefore could not override
nasality cue. Problem we have no proof that
nasality triggers a schema.

• Competing primitives are weighted

In experiment 2, it is shown that the F0
primitive is a stronger cue than the inconsistent
nasality cue.
40
very tentative... HYPOTHESES

• Primitives trigger schemata if possible

Place transitions in experiment 1 triggered same
place schema and therefore could not override
nasality cue. Problem we have no proof that
nasality triggers a schema.

• Competing primitives are weighted

In experiment 2, it is shown that the F0
primitive is a much stronger cue than the
inconsistent nasality cue.
or
Streams are preselected on the basis of F0
In experiment 2, the inconsistent nasality cue
cannot be used by the listener because the F0
primitive has split the signal into two streams.
41
THANK YOU