Laryngeal Adjustments for Vowel Devoicing in Japanese: An Electromyographic
Study
Hajime Hirose*
Haskins Laboratories, New Haven
It is well known that high vowels between voiceless consonants are
often devoiced in many dialects of Japanese including Tokyo dialects
(Bloch, 1950; Han, 1962). Previous studies with a fiberscope revealed that
the glottis remained open for the devoiced vowel segments (Hirose, 1971a;
Sawashima, 1969, 1971). Based on an electromyographic study of the activity
of the vocalis muscle in articulation, the present author reported that de-
voicing of Japanese vowels appears to be a matter concerning the neural pro-
cess that determines the motor commands to the larynx (Hirose, 197la). In
the present study, electromyographic activities of selected intrinsic laryn-
geal muscles were examined with special reference to vowel devoicing in Jap-
anese in comparison with the production of voiceless consonants.
METHOD
A speaker of the Tokyo dialect served as the subject in the present
study and read randomized lists of test sentences sixteen times each. Each
sentence embedded a test word in a frame "soreo -- to ju:" (That we call ~~).
Table I lists the types of test words used in the experiment. They are all
meaningful Japanese words. No accent kernel is attached to those words ex-
cept for the last four pairs in the table, in which the position of the accent
kernel is indicated by the mark "1." Devoicing typically occurs for all
[1]'s between voiceless consonants as indicated in the table.
Electromyographic recordings were made using hooked-wire electrodes.
The wires used were insulated platinum-iridium alloy, the outer diameters of
which were approximately 50 microns. The electrodes were inserted perorally
using a curved probe into the posterior cricoarytenoid (PCA) and the inter-
arytenoid (INT) by indirect laryngoscopy, while percutaneous approach was
employed for insertion into the vocalis (VOC) and the cricothyroid (CT).
Further description of the insertion techniques may be found in previous re-
ports (Hirose, 1971a, b).
The electromyographic signals were recorded on a multichannel data re~
corder together with acoustic signals and automatic timing markers. The sig-
nals were reproduced, high-pass filtered, and fed into a computer after appro-
priate rectification and integration. The electromyographic signals were
averaged for more than fourteen selected utterances of each test sentence with
reference to a line-up point on the time axis representing a predetermined
‘Also Faculty of Medicine, University of Tokyo
157(1) Words with no accent kernel
[ sesse: [ sekdse: ]
sekke: U sekdke: J C sekite: J]
[ sette:
U sjdse: U sytze: J
C sydke: C sjige:
C sdte: ( side: J
C ssdne:
Ckirt ]
[ tenko: denko:]
initial [s] .... 7
initial [s,] ....
initial [2,] ....
medial [k] .... 4
One for each, otherwise
initial kgs ty dy
sis, sik, sit, sin
medial vere Seed, ty dh
kis, kit, klk,
ss, kk, tt,
(2) Words with accent kernel
Csert ] C ze eri J CT kesri J
Ctest i] Caeit i] [ pa'su )
Table I: List of test words used in the present study.
158speech event. In the present experiment, voice onset following [t] in the
frame " --- to yuu" in each sentence was taken for a line-up point. The data
recording and the computer-processing system employed in the present exper-
iment as described in more detail by Port (1971).
RESULTS
‘The laryngeal adjustments in terms of the opening and closing gestures
of the glottis for the voiced/voiceless distinction appeared to be executed
by reciprocal activities of the abductor and adductor muscle groups of the
larynx. In particular, the PCA consistently showed increasing activity for
the voiceless portions of test utterances, while its activity was suppressed
for the production of voiced segments. Conversely, INT activity appeared to
be suppressed for voiceless portions and increased for voiced ones, thus pre~
senting a sort of inversion of the pattern of PCA activity throughout the
utterance.
Figure 1 illustrates an example of the averaged EMG curves of, from
bottom to top, the PCA, the INT, and the VOC, for the utterance of [soreo
zyike: to ju:] and of [soreo s,ige: to ju:], thus comparing the patterns
of the muscle activity in respect to the [z,] vs. [s,] and [k] vs. [g] con-
trasts. It is clearly demonstrated in both cases that there is a reciprocal
pattern of activity between the PCA and the INT.
In the case of [soreo s,ige: to ju:], for example, the PCA shows in
creasing activity for the production of voiceless [s,] and [t] and remains
suppressed for the rest of the test utterance. On the contrary, the INT shows
a rapid decrease in activity for [s,] and [t], while it stays at high level
for the rest. The timing of the peak PCA activity approximately coincides
with that of the maximum suppression of INT activity. There is a shallow dip
in the INT curve, apparently corresponding to [g] production.
For the utterance of [soreo z,ike: to Ju:], the PCA shows increasing
activity for [k] and [t] and suppressed activity for the rest. The INT shows
a gradual decrease in activity for the sequence [zi], folloved by further
suppression corresponding to increasing PCA activity.
The activity of the VOC generally stays at a high level for the vowel
portion of the utterance, while it becomes low for consonant segments regard-
less of the voiced/voiceless distinction, although the activity is usually,
but with some exceptions, somewhat higher for a voiced consonant than for a
voiceless consonant if we compare the averaged EMG values for a given set of
voiced/voiceless consonant pairs.
Figure 2 compares the averaged EMG curves for the sentences embedding
[sette:] vs. [sekjte:], where the interconsonantal [i] is devoiced in the
latter.
It is shown that PCA activity increases for the sequence [kjt] as well
as for the geminate [tt] and initial [s], while the INT is markedly suppressed
for these sequences.
Jin the examples in Figure 1, VOC activity is higher for [g] than for [k] but
lower for [z,] than for [8].
159