24.964
Phonetic Realization
Tone
Readings for next time:
• Kenstowicz, Abu Mansour and Törkenczy (2000). Two
notes on laryngeal licensing.
• Kang, Y-J. (2003) Perceptual similarity in loanword
adaptation: English postvocalic word-final stops in
Korean. Phonology 20.
Realization of lexical tone - Mandarin
160
F
140
120
F0 (Hz)
60
L
L
H
H
H
H
H
F
140
120
F
H
H
L
H
H
L
H
(b)
H
R
R
100
60
R
(a)
160
80
H
R
R
100
80
F
H
L
F
(c)
H
H
H
Normalized Time
R
(d)
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Speech Melody as Articulatory Implemented Communicative Functions."
Speech Communication 46 (2005): 220-251.
• Substantial variation in tone on syll3 depending on syll2 tone (carryover coarticulation).
• Variation in tone on syll3 is larger earlier in the syllable.
– transition from syll2, converging on a target for syll3.
Realization of lexical tone - Mandarin
160
F
140
120
F0 (Hz)
60
L
L
H
H
H
H
H
F
140
120
F
H
H
L
H
H
L
H
(b)
H
R
R
100
60
R
(a)
160
80
H
R
R
100
80
F
H
L
F
(c)
H
H
H
Normalized Time
R
(d)
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Speech Melody as Articulatory Implemented Communicative Functions."
Speech Communication 46 (2005): 220-251.
• Minimal variation in H tone on syll 1 (minimal anticipatoty
coarticulation).
• Main anticipatory affect seems to be dissimilatory raising of H
preceding L.
Realization of lexical tone - Mandarin
1-2
m
a
3-2
m
4-2
a
140
140
120
120
100
100
80
80
0
50 100 25 50
75 100 50 100 25 50
2-1
m
160
f0(Hz)
f0(Hz)
160
2-2
75 100
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Contextual
Tonal Variations in Mandarin." Journal of Phonetics 25 (1997): 61-83.
0
a
2-2
50 100 25 50
2-3
m
2-4
a
75 100 50 100 25 50
75 100
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Contextual
Tonal Variations in Mandarin." Journal of Phonetics 25 (1997): 61-83.
Same points based on data from Xu (1997)
• Substantial carryover coarticulation.
• Minimal anticipatory coarticulation.
Syllable-synchronized sequential target approximation
Onset F0 determined by: default neutral
value, or preceding pitch target
Turning points
(delayed due to inertia)
Approaching [low]
with weak effort
Approaching [rise]
Acceleration
t
ge
ar
ct
]
se
[ri
F0
i
am
n
Dy
Deceleration+
Acceleration
Approaching [low]
[low]
Static target
time
Image by MIT OpenCourseWare. Adapted from Xu, Y. “Speech Melody as Articulatorily Implemented Communicative Functions.”
Speech Communication 46 (2005): 220-251. Also, Xu, Y., and Q. E. Wang. "Pitch Targets and Their Realization: Evidence
from Mandarin Chinese." Speech Communication 33 (2001): 319-337.
Xu and Wang (2001), Xu (2005)
• Linear tone targets.
• Tone targets are coextensive with syllables.
• f0 approaches tone target asymptotically while target is
active (no look-ahead).
Target approximation - Mandarin
160
F
140
120
F0 (Hz)
60
L
L
H
H
H
H
H
F
140
120
H
F
H
L
H
H
L
H
(b)
H
R
R
100
60
R
(a)
160
80
H
R
R
100
80
F
H
L
F
(c)
H
H
H
Normalized Time
R
(d)
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Speech Melody as Articulatory Implemented Communicative Functions."
Speech Communication 46 (2005): 220-251.
• Dashed lines show estimated tone targets for third
syllables.
Syllable synchronization
N-V ratio:
Top:
Bottom:
N<V
fan
` nao
nan
` ling
N=O
hao
` miao
le
`I niao
N>V
hong
` ma
ling min
`
300
200
F0 (Hz)
100
0
300
200
100
0
V-N boundary
0
100
200
300
400
500
600
-250
-150
-50
50
150
250
350
Time (ms)
Image by MIT OpenCourseWare. Adapted from Xu, Y. "Speech Melody as Articulatory Implemented Communicative Functions."
Speech Communication 46 (2005): 220-251.
•
•
The same tone sequence (RL) results in very similar f0 contours
regardless of the relative duration of V and coda N in syllable 1.
Right panels show less consistency when contours are aligned at the
offset of V1.
Anticipatory vs. carryover coarticulation in Thai
• Thai is also reported to show more carryover than
anticipatory coarticulation, but there is anticipatory
coarticulation (Gandour et al 1994).
• Data from one speaker in Nuttakorn (2002) suggests that
anticipatory coarticulation primarily affects overall pitch
range of tones, whereas carryover coarticulation involves
a transition from the endpoint of the preceding tone.
Nuttakorn (2002)
m
aa
m
aa
300
300
250
250
200
150
100
100
0 50 100
25
50
75
100 50 100
25
50
75
aa
50
100
0 50 100
25
50
Duration (% of each segment)
M-M
350
m
F-M
H-M
aa
R-M
m
M-L
aa
350
300
300
250
250
200
150
100
100
50
25
50
75
100 50 100
25
50
75
100
M-L
M-F
M-H
L-L
m
50
0 50 100
Duration (% of each segment)
M-M
75
100 50 100
25
50
75
100
F-L
aa
H-L
m
R-L
aa
200
150
0 50 100
aa
Duration (% of each segment)
F0 (Hz)
F0 (Hz)
Anticipatory
L-M
m
200
150
50
m
350
F0 (Hz)
Carryover
F0 (Hz)
350
25
50
75
100 50 100
25
50
75
100
Duration (% of each segment)
M-R
L-M
L-L
L-F
L-H
L-R
Images by MIT OpenCourseWare. Adapted from Thubthong, Nuttakorn, Boonserm Kijsirikul, and Sudaporn Luksaneeyanawin.
"Tone Recognition in Thai Continuous Speech Based on Coarticulation, Intonation and Stress Effects." International
Conference on Spoken Language Processing (ICSLP2002), Colorado, USA, September 16-20, 2002, pp. 1169-1172 .
Nuttakorn (2002)
m
aa
m
aa
350
300
300
250
250
F0 (Hz)
Carryover
F0 (Hz)
350
200
150
100
100
0 50 100
25
50
75
100 50 100
25
50
75
100
aa
50
0 50 100
25
Duration (% of each segment)
M-F
F-F
m
H-F
aa
m
aa
50
75
100 50 100
25
50
75
100
Duration (% of each segment)
R-F
M-H
aa
350
300
300
250
250
F0 (Hz)
Anticipatory
• Note F-F
F0 (Hz)
350
L-F
m
200
150
50
m
200
F-H
m
H-H
aa
R-H
m
aa
200
150
150
100
100
50
L-H
50
0 50 100
25
50
75
100 50 100
25
50
75
100
0 50 100
Duration (% of each segment)
F-M
F-L
F-F
F-H
25
50
75
100 50 100
25
50
75
100
Duration (% of each segment)
F-R
H-M
H-L
H-F
H-H
H-R
Images by MIT OpenCourseWare. Adapted from Thubthong, Nuttakorn, Boonserm Kijsirikul, and Sudaporn Luksaneeyanawin.
"Tone Recognition in Thai Continuous Speech Based on Coarticulation, Intonation and Stress Effects." International
Conference on Spoken Language Processing (ICSLP2002), Colorado, USA, September 16-20, 2002, pp. 1169-1172 .
Anticipatory vs. carryover coarticulation in Cantonese
Normalized F0
Tone 4
Tone 1
1.2
1.2
1
1
0.8
0.8
Tone 4
Tone 4
1.2
1.2
1
1
0.8
0.8
Tone 4
Tone 5
1.2
1.2
1
1
0.8
0.8
Normalized time
Tone 1
Tone 4
Tone 4
Tone 4
Tone 5
Tone 4
Normalized time
Examples of co-articulated tone contours of disyllabic words.
Image by MIT OpenCourseWare. Adapted from Li, Yujia, Tan Lee, and Yao Qian. "Analysis and Modeling of F0 Contours
for Cantonese Text-to-Speech." ACM Transactions on Asian Language Information Processing (TALIP) 3, no.
3 (September 2004): 169-180.
• Cantonese also shows carryover articulation but
minimal anticipatory coarticulation (Li et al 2004).
Li et al (2004)
1.4
Normalized F0
1.4
1
1.2
1
2
3
0.8
5
6
4
1
Normalized F0
1.4
1
5
6
4
LL: low-level left context
1.4
1
1
1.2
1
3
6
0.8
2
3
0.8
LH: high-level left context
1.2
1
1.2
2
2
3
5
4
RH: high-level right context
5
0.8
4
6
RL: low-level right context
Image by MIT OpenCourseWare. Adapted from Li, Yujia, Tan Lee, and Yao Qian. "Analysis and Modeling of F0 Contours
for Cantonese Text-to-Speech." ACM Transactions on Asian Language Information Processing (TALIP) 3, no.
3 (September 2004): 169-180.
• Averaged tone contours, divided according to height of preceding/following tone.
• “for almost all tones, the beginning section has a much larger variance than the ending one. This suggests that co-articulation from the left tone is more significant than that from the right. In general, the third and the fourth sections have the smallest variance”
Anticipatory vs. carryover coarticulation
• Predominance of carryover coarticulation is reported in
Vietnamese (Brunelle 2003).
• Apparently in Igbo also (Akinlabi & Liberman ??, NELS
31), although with an interesting twist (below).
Asymmetry in coarticulation
Carryover
not: anticipatory
symmetrical
• Xu attributes asymmetry to an imperative to keep tonal gestures in
phase with syllables (cf. Goldstein, Nam etc).
– i.e. onset of tone gesture coincides with onset of syllable
– but what is the syllable gesture? Jaw?
• In Browman & Goldstein (1990), there is no gesture
coextensive with the syllable.
– Why is the offset of the gesture coordinated with the offset of
the syllable?
Asymmetry in coarticulation
Carryover
not: anticipatory
symmetrical
Alternatives:
• Syllable rhyme provides a better basis for realization of tonal
contrasts because onset may be obstruent,
– Only codas in Mandarin are nasals. Different patterns with obstruent
codas?
– In Mandarin, same pattern obtains with glide onset (Xu ??).
• Basic idea: tonal targets are weighted according to position - late
targets have higher weights.
– variant: tonal targets are only assigned the ends of syllables (cf.
Akinlabi & Liberman) - doesn’t work for contour tones.
Asymmetry in coarticulation
• Xu’s model of tonal realization seems very compatible
with a weighted constraint formulation.
• Implementation is easier if coarticulatory asymmetries
follow from positional weighting of targets.
• Targets with extended duration (as opposed to point
targets) are required.
– Possibly also dynamic targets, but cf. Potisuk et al (1997) who model
tone implementation in Thai with sequences of level targets.
– Slope targets are motivated by evidence that tone slope is perceptually
salient (Gandour 19
Anticipatory tonal coarticulation in Taiwanese -
Peng (1997)
Low-rising tone (24)
Low-falling tone (21)
F1
F1
55
33
21
(h)
-20 30 80 130 180 230 280 330
Duration (ms)
21
55
33
(i)
-20 30 80 130 180 230 280 330
Duration (ms)
Image by MIT OpenCourseWare. Adapted from Peng, Shu-hui. "Production and Perception of Taiwanese Tones
in Different Tonal and Prosodic Contexts." Journal of Phonetics 25 (1997): 371-400.
• Only examined anticipatory tonal coarticulation.
• Substantial effects for one speaker, smaller but significant effects with others (especially for high falling 51 tone).
Anticipatory tonal coarticulation in Kinyarwanda
Myers (1998) • Kinyarwanda contrasts L, H, LL, HL, LH
• Contour tones only permitted on long vowels
umwáana
‘child’
umwaámi
‘king’
umuunhu
‘person’
ururími
‘language’
• Rise for H, HL begin during preceding syllable.
C
H
HL
LH
170
130
110
90
ke
170
130
110
90
ke
170
130
110
90
ke
V
350
700
1050
1400
1750
350
700
1050
1400
1750
310
600
910
1200
1500
C V
Image by MIT OpenCourseWare. Adapted from Myers, S. "F0 Timing in Kinyarwanda." Phonetica 60 (2003): 71-97.
Kinyarwanda - Myers (1998)
• Rise for H, HL begin during preceding syllable.
• Timing of onset of rise as a proportion of the preceding
syllable duration:
Speaker
H
HL
LH
1
0.317
0.457
1.560
2
3
0.402
0.185
0.182
0.323
1.606
1.325
4
0.336
0.511
1.419
Image by MIT OpenCourseWare. Adapted from Myers, S. "F0 Timing in Kinyarwanda." Phonetica 60 (2003): 71-97.
• Rise begins about 100 ms before f0 peak.
• With word-final, phrase-final H syllables, f0 peak can
precede the syllable.
Kinyarwanda - Myers (2003)
• Kinyarwanda tones also differ from Mandarin tones in
that the alignment of the f0 peak within a high-toned
syllable varies susbtantially:
– Earlier in phrase-final words than in phrase-medial words.
– Earlier in phrase-final syllables than in phrase-medial syllables.
– H peak tends to be earlier the shorter the distance to the next H
(‘crowding’) (no such effect for HL, LH).
u|u|i!mi |gwaa≠ ÔeÓ
No following H
i|a|u!m umNaanaÓ
H follows after 1 syllable
!
u|u|i!mi |uniniÓ
H follows after 2 syllables.
!
• Similar factors affect the timing of f0 peaks for H* pitch
accents in English (Silverman & Pierrehumbert 1990) and Spanish
(Prieto et al 1995).
Kinyarwanda vs. Mandarin • Why does Kinyarwanda work so differently from
Mandarin and the other SE Asian languages cited above?
• Fewer tones? Only two levels HL.
• Asymmetry between L and H - L is phonologically inert.
Analyzed by Myers as phonologically absent (cf.
Chichewa, Myers 1998).
– Realization of L has a low weight?
• Behaviour similar to Mandarin can be observed in a 3toned African language, Yoruba (Akinlabi & Liberman).
– L, M, H. – M often analyzed as absence of tone.
Yoruba
• A&M: Simple tones
are assigned a target at
the end of the syllable.
• If there is a sequence
of syllables with the
same tone, a target is assigned to the end of
the last syllable.
• Initial mid target.
• Interpolate between
targets.
• Igbo is much the same.
290
280
270
260
250
240
230
220
210
200
190
180
170
160
150
140
130
120
110
a1_68
o
w
349.5
a
l
a
349.75
n
i
r
350
o
r
un
350.25
o wa la ni ro run 'he drove Alani to heaven'
H(
L) H L M
Image by MIT OpenCourseWare. Adapted from Akinlabi, A., and
M. Liberman. "Tonal Complexes and Tonal Alignment." NELS 31 (2000).
Tonal anticipation in intonation
5000
0
0
L+H*L- Time (s) H*
L-L%
Ma r i a nn a m adethe m ar ma l a de
1.70603
• Near-linear interpolation from L- to H* across [meIdD´].
– If H* were synchronous with its syllable, no
movement should occur until onset of [mA].
• Note also movement from H* to final L- is ‘early’.
Phonetics and phonology of tone • Some patterns that have been analyzed as tone spread are
better analyzed in terms of the details of tonal realization.
– Kinyarwanda has been described as having lefwtards spread of
H tone, e.g. /umusóre/ -> umúsóre (Sibomana 1974).
• Many phonological processes have been described that resemble progressive tonal coarticulation, e.g. Yoruba (a) Yoruba rising example
ala (LH)
ala (L LH)
'dream'
(b) Yoruba falling example
rara (HL)
rara (H HL)
'elegy'
(c) Edo (Bini) falling example
ekpo (H HL)
ekpo (HL)
'bag'
Image by MIT OpenCourseWare.
• H tone spread tends to be rightwards (Myers).
Phonetics and phonology of tone • These processes may well be phonological - in some cases
they are neutralizing. But they are also probably
motivated by the same constraints that motivate tonal
coarticulation.
• E.g. In Kinyarwanda, f0 peak for H is early in phrasefinal syllables. In Chichewa, final H retracts to the penult
in phrase-final position (Myers 1998).
References •
•
•
•
•
•
•
•
•
•
Peng, S.-H., 1997. Production and perception of. Taiwanese tones in different tonal
and prosodic contexts. Journal of Phonetics, 25, 371-400.
Potisuk, Gandour & Harper (1997). Contextual variations in trisyllabic sequences of
Thai tones. Phonetica 54.
Prieto, P.; Santen, J. van; Hirschberg, J. (1995) Tonal alignment patterns in Spanish. J.
Phonet. 23: 429–451.
Silverman, K.; Pierrehumbert, J. (1990) The timing of prenuclear high accents in
English; in Kingston, Beckman, Papers
in Laboratory Phonology I: Between the grammar and the physics of speech, pp. 72–
106. Cambridge University
Press, Cambridge.
Xu, Y. (1997). Contextual tonal variations in Mandarin. Journal of Phonetics 25: 61
83.
Xu, Y. (2005). Speech melody as articulatorily implemented communicative functions.
Speech Communication 46: 220-251.
Xu, Y. and Liu, F. (2007) Determining the temporal interval of segments with the help
of F0 contours. Journal of Phonetics 35: 398-420.
Xu, Y. and Wang, Q. E. (2001). Pitch targets and their realization: Evidence from
Mandarin Chinese. Speech Communication 33: 319-337.
MIT OpenCourseWare
http://ocw.mit.edu
24.964 Topics in Phonology: Phonetic Realization
Fall 2006
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.