present experiment introduction coarticulatory timing and lexical effects on vowel nasalization in...
TRANSCRIPT
Present Experiment
Introduction
Coarticulatory Timing and Lexical Effects on Vowel Nasalization in English: an Aerodynamic Study
Jason BishopUniversity of California, Los Angeles
In English, vowels show significant nasalization when preceding a nasal consonant. The extent of this coarticulation, however, depends on other aspects of the phonetic context. For example, in (C)VNC sequences, nasalization is more extensive when:
initial C is an aspirated stop, or if NC are tautosyllabic (Cohn 1995). V is stressed (Krakow 1993). C is voiceless rather than voiced (Malécot 1960).
Beddor (2007) suggests the difference between a voiced C context (VND) and voiceless C context (VNT) is not due to more nasalization as such, but to the earlier onset of velum lowering when the oral articulation for N is shorter:
Constant-sized velum-lowering gesture begins later when N is longer, overlapping less with vowel
Constant-sized velum-lowering gesture begins earlier when N is shorter, overlapping more extensively with vowel
Less Nasalization
More Nasalization
Beddor (2007) presents acoustic evidence from VNC productions of 5 speakers of American English that this trading relationship between the durations of nasal consonants (N) and the duration of coarticulatory vowel nasalization (Ṽ), holds (R2 ranging from .27 to .45).
Goals 1.) Corroborate the findings in Beddor 2007 with aerodynamic data. 2.) Examine possible lexical effects as an additional source of variation in extent of coarticulatory vowel nasalization.
Methods 4 speakers of American English (2 male, 2 female; ages 21 – 28) Speakers read (C)VNC tokens embedded in the carrier sentence “Please say ______ with me”. Oral and nasal airflow collected simultaneously with audio recordings.
Variables
Duration of the nasal consonant (N) Duration of significant nasal airflow over the vowel (Ṽ) (i.e., nasal flow above the levels present in non-nasal context controls –e.g., sped for spend) Proportion of the vowel with significant nasal airflow (duration of significant nasal airflow over the vowel / total duration of vowel)
Vowel Nasalization and Nasal Consonant Duration: VNT and VND Contexts
0
50
100
150
200
250
300
0 20 40 60 80 100 120 140 160 180
Duration N (ms)
Dura
tion Ṽ
(m
s)
Results: Across Voicing Context
Figure 2. Scatterplot showing vowel nasalization as a function of the duration of the nasal consonant for (C)VNC test words (across voicing contexts for C). Although the trend is in the direction of an inverse relationship, the duration of N accounts for very little of the variation in Ṽ. R2 = .03(Speaker 1), .05(Speaker 2), .02(Speaker 3), .05(Speaker 4).
1. Evidence for an inverse relationship not found across voicing contexts:
2. N duration is a better predictor of Ṽ duration within voicing contexts:
VNT
0
50
100
150
200
250
0 20 40 60 80 100 120 140 160
Duration N (ms)
Dur
atio
n Ṽ
(ms)
VND
0
100
200
300
0 20 40 60 80 100 120 140 160 180
Duration N (ms)
Dur
atio
n Ṽ
(ms)
Figure 3. Vowel nasalization and nasal consonant duration in a voiceless obstruent context. R2 for trend lines = .61(Speaker 1), .16(Speaker 2), .25(Speaker 3), .10(Speaker 4).
Figure 4. Vowel nasalization and nasal consonant duration in a voiced obstruent context. R2 for trend lines = .29(Speaker 1), .28(Speaker 2) .23(Speaker 3), .37(Speaker 4).
Speaker 1High Frequency
0
50
100
150
200
0 20 40 60 80 100 120 140
Duration N (ms)
Dura
tion
? (m
s)
Speaker 1Low Frequency
R2 = 0.23
0
50
100
150
200
0 20 40 60 80 100 120 140
Duration N (ms)
Dur
atio
n ?
(ms)
Speaker 2High Frequency
R2 = 0.12
0
50
100
150
200
250
0 25 50 75 100 125 150
Duration N (ms)
Dura
tion
? (m
s)
Speaker 2Low Frequency
R2 = 0.19
0
50
100
150
200
250
0 25 50 75 100 125 150Duration N (ms)
Dura
tion
Ṽ (m
s)
Speaker 3High Frequency
0
50
100
150
200
250
0 25 50 75 100 125
Duration N (ms)
Dura
tion
Ṽ (m
s)
Speaker 3Low Frequency
R2 = 0.15
0
50
100
150
200
250
0 25 50 75 100 125
Duration N (ms)
Dur
atio
n Ṽ (
ms )
Lexical Effects: Neighborhoods
Lexical Effects: Frequency
Summary Within both the voiced and voiceless context, the duration of N accounted for considerable amounts of the variation in Ṽ; not true when voicing contexts were considerable together. This differs from what is reported in Beddor 2007, where the strongest relationship was found across
voicing contexts.
Scarborough (2004) measured anticipatory nasal coarticulation in two lexical confusability conditions, finding more extensively nasalized vowels in more confusable or, here, lexically “difficult” words:
Easy (or high R : high frequency within sparse phonological neighborhoods)
Difficult (or low R : low frequency within dense phonological neighborhoods)
Three Questions
1.) Does lexical frequency (independent of neighborhood size/lexical difficulty) influence the relation between N and Ṽ ?
2.) Does neighborhood size correlate with degree of nasalization ?
3.) How does lexical difficulty influence the relation between N and Ṽ ?
Subsets of the production data with the relevant lexical properties were examined to address these questions. Lexical statistics were taken from the Washington University Speech and Hearing Lab Neighborhood Database.
Speaker 1
R2 = 0.49
0.4
0.6
0.8
1
0 5 10 15 20
# of Neighbors
Prop
ortio
n V
Nas
aliz
ed
Speaker 4
R2 = 0.36
0.2
0.4
0.6
0.8
1
0 5 10 15 20# of Neighbors
Prop
ortio
n V
Nasa
lized
Speaker 2
R2 = 0.15
0.2
0.4
0.6
0.8
1
0 5 10 15 20
# of Neighbors
Prop
ortio
n V N
asal
ized
Speaker 3
R2 = 0.15
0.2
0.4
0.6
0.8
1
0 5 10 15 20
# of Neighbors
Prop
ortio
n V N
asal
ized
Word# of
Neighbors
spent 3
spend 5
grant 6
want 9
ant 12
tint 14
lent 17
cant 18
went 19
Speaker 1Difficult
0
50
100
150
200
0 20 40 60 80 100 120
Duration N (ms)
Dura
tion Ṽ (
ms)
Speaker 2Difficult
R2 = 0.33
0
50
100
150
200
0 25 50 75 100 125
Duration N (ms)
Dura
tion
Ṽ (
ms)
Speaker 3Difficult
0
50
100
150
200
250
0 25 50 75 100 125
Duration N (ms)
Dura
tion Ṽ
(m
s)
Speaker 1Easy
R2 = 0.34
0
50
100
150
200
250
0 20 40 60 80 100 120 140
Duration N (ms)
Dura
tion Ṽ (m
s)
Speaker 2Easy
R2 = 0.37
0
50
100
150
200
0 25 50 75 100 125 150
Duration N (ms)
Dura
tion Ṽ
(m
s)
Speaker 3Easy
R2 = 0.11
0
50
100
150
200
250
0 25 50 75 100 125 150 175
Duration N (ms)
Dura
tion Ṽ (m
s)
Speaker 4Easy
R2 = 0.3459
0
50
100
150
200
0 25 50 75 100 125 150
Duration N (ms)
Dura
tion Ṽ
(m
s)
Speaker 4Difficult
0
100
200
300
0 25 50 75 100 125 150
Duration N (ms)
Dura
tion Ṽ
(m
s)High Frequency Low Frequency
Easy Difficult
Word log freq# of
Neighbr R
rent 2.3222 17 0.066844
lend 2.1461 17 0.058803
bent 2.5315 17 0.061989
June 2.9685 16 0.090507
can't 3.2304 18 0.086955
bend 2.3802 18 0.057645
went 3.705 19 0.16318
and 5.4602 12 0.140092
MEAN3.09301
3 16.75 0.09
High Frequency
Word log freq# of
Neighbr R
dent 1.301 19 0.033844
lent 1.699 17 0.041772
rend 1 19 0.025448
scant 1.699 3 0.209611
wand 1 9 0.053691
canned 1.7782 16 0.041803
gent 1 17 0.027906
wend 1 18 0.028739
MEAN 1.30965 14.75 0.058
Low Frequency
Results: Within VNT and VND
Word log freq
# of Neighbr R
want 3.5198 9 0.199035
grand 2.6812 7 0.166447
scant 1.699 3 0.209611
spend 2.7243 5 0.216822
grant 2.6721 6 0.240051
spent 3.017 3 0.294879
MEAN 2.7189 5.5
0.221141
Easy
Wordlog
freq# of
Neighbr R
bent 2.5315 170.0619888
3
rent 2.3222 170.0668439
8
dent 1.301 190.0338437
2
lent 1.699 170.0417721
8
canned 1.7782 16
0.04180287
bend 2.3802 180.0576446
6
MEAN2.00201
7 17.333330.0506493
8
Difficult
1. N and Ṽ: trend towards inverse correlation for high frequency, positive
correlation for low frequency tokens
2. Neighborhood size and vowel nasalization positively correlated (R2 = .15 – .49)
3. N and Ṽ: inversely correlated for lexically easy words, unrelated or positively
correlated for lexically difficult tokens
Figure 5. Vowel nasalization and nasal consonant duration for high and low frequency subsets of the tokens (three repetitions of 8 tokens plotted for each frequency condition).
Speaker 4High Frequency
R2 = 0.1953
0
50
100
150
200
250
0 25 50 75 100 125 150
Duration N (ms)
Dura
tion
Ṽ (m
s)
Speaker 4 Low Frequency
0
50
100
150
200
250
0 25 50 75 100 125 150
Duration N (ms)
Dur
atio
n Ṽ
(m
s)
Figure 6. Proportion vowel nasalized (Ṽ/Vtotal) and neighborhood size (each point represents average values over three repetitions of each of nine tokens).
Figure 7. Vowel nasalization and nasal consonant duration for lexically easy and difficult words (high and low R, respectively). Three repetitions of 6 tokens plotted for each lexical type).
Word List
Word List
Word List
Conclusion
ReferencesBeddor, P. (2007). Nasals and nasalization: the relation between segmental and coarticulatory timing. Proceedings of the 16th International Congress of Phonetic Sciences, 249-254.
Cohn, A. (1990). Phonetic and phonological rules of nasalization. UCLA Working Papers in Phonetics 76, 1 – 224.
Krakow, R. (1993). Nonsegmental influences in velum movement patterns: syllables, sentences, stress, and speaking rate. In Huffman, M. & Krakow, R. (eds.): Nasals, Nasalization, and the Velum. New York: Academic Press, 87 – 113.
Malécot, A. (1960). Vowels nasality as a distinctive feature in American English. Language 36, 222 – 229.
Tatham, M. & Morton, K. (2006). Speech production and perception. New York: Palgrave.
The results presented here are in general agreement with those in Scarborough 2004:
Speakers produced more, not less, coarticulation for tokens from denser phonological neighborhoods.
Further, they have implications for patterns of gestural alignment discussed in Beddor 2007: A tendency to decrease vowel nasalization in the presence of a longer nasal consonant, as predicted by Beddor, is apparent in the productions of these 4 speakers.
But: this effect is restricted to just those lexical conditions expected to be easy for listeners: high-frequency words and words of a high frequency within sparse neighborhoods.
Such findings are highly suggestive of what we should expect in a more listener-oriented communication task (more coarticulation for low frequency/difficult words regardless of the duration of the nasal consonant), and as such are very difficult to understand under a view of coarticulation as reduction (e.g., Thatham & Morton 2006).
Figure 1. From top to bottom channels: Audio, oral airflow, nasal airflow for the sentence “Please say bend with me.” Black dotted line shows duration of the vowel with significant nasal flow; red dash dot line shows the duration of the nasal consonant. Zero nasal flow is referenced to nasal flow during the vowel of non-nasal [bɛt].
500 ml/s
1000 ml/s