The speakers produced breathier vowels after voiced stops

University of Iowa vladimir-kulikov@uiowa.edu

Sundanese is usually described as a language with a laryngeal contrast between voiced and voiceless stops. However, phonetic properties of the laryngeal contrast in Sundanese have not been studied instrumentally. Evidence from neighboring related languages (Indonesian, Javanese, Madurese) suggests that the laryngeal contrast in Sundanese can be realized along the two major patterns: contrastive prevoicing with alternations of fundamental frequency (f0) as in Indonesian, or contrastive phonation of vowels following voiceless and voiced stops as in Javanese and Madurese. Two native speakers of Sundanese (male and female) pronounced syllables with voiceless and voiced stops (p, t, k, b, d, g) followed by seven vowels. Four phonetic properties of voicing were investigated: VOT, f0, phonation of a following vowel (H1-H2, H1-F1, H1-F2), and changes in F1/F2. The results show that while some properties of Sundanese stops (VOT, f0) are consistent with the voicing contrast in Indonesian, phonation of vowels that follow voiceless and voiced stops is more typical for languages with breathy voice (Javanese) or ‘register’ type languages (Madurese). The speakers produced breathier vowels after voiced stops. In addition, F1 and F2 of non-high vowels [ɛ, ǝ, ɔ, a] changed significantly after voiced stops. Voiced stops systematically caused raising and fronting of the following vowel. The results show that the laryngeal contrast in Sundanese has redundant phonetic properties, which may indicate an ongoing language change or influence from a neighboring language (Javanese).

1. Introduction

Descriptive works on Sundanese, a Western Indonesian language spoken by 30 million people in West Java, report that it has a voicing contrast between voiced and voiceless stops (Van Syoc 1959, Sudaryat 1985, Müller-Gotama 2001). However, this description is impressionistic and not based on acoustic measurements. Closely related languages – Indonesian, Javanese, and Madurese, – exhibit two patterns of voicing contrasts: 1) the contrast between fully voiced and voiceless unaspirated stops in Indonesian (Adisasmito-Smith 2004) and 2) the contrast involving alternation of vowel height after voiceless and voiced consonants in Javanese (Fagan 1988, Hayward 1993, Ladefoged and Maddieson 1996, Thurgood 2004) and Madurese (Cohn 1993a, Cohn and Lockwood 1994). This paper examines whether the voicing contrast in Sundanese follows the pattern with voice alternation on stops or the pattern with height alternation on the following vowel.

(1) Sundanese obstruents

Bilabial Alveolar Palatal Velar Glottal

Stops Voiceless p t c k ()

Voiced b d ɟ g

Fricatives s h

Sundanese has voiced and voiceless stops at four places of articulation: bilabial /p, b/, alveolar /t, d/, palatal /c, ɟ/, and velar /k, g/ (1). Sundanese has only voiceless alveolar /s/ and glottal /h/ fricatives. Kurniawan (2009) reports that word-initial voiced stops have robust prevoicing, and voiceless stops have a short-lag VOT, which is consistent with the patterns found in ‘true voice’ languages (Dutch, French, Indonesian, Russian).

(2) Sundanese vowels

Front Central Back

High i ɨ u

Mid ɛ ə ɔ

Low a

Sundanese has seven contrastive vowels: /i/, /ɛ/, /a/, /ɨ/, /ə/, /ɔ/, and /u/ (2). Acoustic observations suggest that non-high vowels /a/, /ə/, /ɛ/, /ɔ/ undergo raising and fronting after voiced stops while high vowels /i/, / ɨ /, and /u/ are not affected by voicing in the preceding consonant (3).

(3) /a/  [ɐ]~[] ombak ‘wave’

/ə/  [ɘ] gedé ‘big’

/ɛ/  [e] gedé ‘big’

/ɔ/  [o] botol ‘bottle’

This study presents a preliminary analysis of the acoustic properties of voiceless and voiced stops in Sundanese. It addresses the question about the place of the Sundanese contrast in the typology of voicing contrasts in Western Indonesian languages. To answer the question, we examined the acoustic properties of voicing on a consonant (VOT) and the acoustic properties on the vowel that follows voiced and voiceless stops. In addition to fundamental frequency (f0), which is often shown to be a correlate of the voicing contrast with lower values after voiced stops and higher values after voiceless stops, we investigated the voice quality of the following vowel. Differences in vowel phonation were examined for both spectral energy and formant frequencies.

The difference in amplitude between the first (H1) and second (H2) harmonics has been often used to distinguish between breathy and modal voicing. Phonation of breathy vowels is characterized by energy dominating at the fundamental frequency; thus, a greater H1-H2 indicates breathiness. The difference in amplitude between H1 and the peak harmonic forming the first (F1) and second (F2) formants indicates the abruptness of vocal fold closure. Greater H1-F1 and H1-F2 (a steeper spectral downward slope) characterize breathy phonation whereas smaller differences indicate modal or creaky voicing. The schematized spectra for different vowel phonation types, studied by Stevens (1999) are shown in Fig.1.

The results of the acoustic analysis were compared with the existing data on voicing in Indonesian, Madurese, and Javanese.

Figure 1. Schematized spectra for vowel phonation types, based on Stevens (1999, pp.86, 90), cited from Cho et al. (2002, p.202)

In the following section, I will review the studies of voicing in other Western Indonesian languages. The experimental setup is described in Section 3. The results are presented in Section 4. General discussion and conclusion are given in Section 5.

2. Previous studies

2.1. Voicing contrast in Indonesian

Indonesian has a two-way contrast between fully voiced and voiceless unaspirated stops.

Adisasmito-Smith (2004) studied voicing properties of vowel phonation after voiceless and voiced stops. Vowels after voiced stops were consistently pronounced with lower f0. No distinct pattern was found for spectral characteristics of vowels. Only the difference between energy values of H1 and F1 was found to be consistently lower after voiceless stops than after voiced stops for all speakers, which may indicate some breathiness of the vowel that follows voiced stops. No data is available or has been reported on height alternation in vowels as a function of voicing.

2.2. Voicing contrast in Javanese

In Javanese, the contrast between p, t, k and b, d, g, is described phonetically as ‘light’ vs.

‘heavy’ (Fagan 1988), or ‘stiff voice’ vs. ‘slack voice’ (Ladefoged & Maddieson 1996). Both definitions highlight the fact that neither voice nor aspiration is a sufficient acoustic property that distinguishes the two series of stops. According to Fagan (1988), both series have a short lag VOT. Mean VOT is 14.9 ms for ‘stiff’ stops and 19.1 ms for ‘slack’ stops. VOT differs significantly between the two series only in dental and velar stops. The two series do not employ voicing during closure as a distinctive property either, the difference being significant only in dental stops. Ladefoged and Maddieson (1996) claim that both types of Javanese stops are

produced with some activity in vocal folds, but it is different from what was reported for voiced and voiceless stops in other languages. ‘Light’ stops are articulated with vocal folds slightly approximated to each other, while ‘heavy’ stops involve vocal fold vibration which is less intensive than in ‘true’ voiced stops and includes breathiness. Therefore, this contrast is implemented in the quality of the following vowel rather than in the absence/presence of vocal fold vibration. Vowels after ‘stiff’ stops are pronounced with clear voice; vowels after ‘slack’

stops are pronounced with breathy voice, which affects spectral properties of vowels. According to Fagan (1988: 182), F1 lowering was one of the most important correlates of the voicing contrast. The vowel [a] was articulated with a significantly lower F1 after ‘voiced’ stops at all places of articulation. For one speaker, mean difference at the vowel onset was 159 Hz between [p] and [b], 152 Hz between [t] and [d], and 136 Hz between [k] and [g]. For the other speaker, the difference was 146 Hz, 201 Hz, and 166 Hz respectively. Fagan also found that F2 was higher after ‘voiced’ stops than after ‘voiceless’ stops. Fundamental frequency of the vowel (f0) was significantly higher only after bilabial and dental stops (16 Hz and 11 Hz for the first speaker, and 8 Hz and 10 Hz for the second speaker). This difference was minimal after velar stops for both speakers (3 Hz and 2 Hz).

Hayward (1993) also found that Javanese vowels are pronounced with lower F1 and f0 after /b/. F2 is higher after ‘voiced’ consonants. Similar, but not identical results of F1 lowering and F2 raising were obtained in Thurgood 2004, who studied production of Javanese vowels [a], [] and [u]. Thurgood showed that vowels after ‘voiced’ stops have less distinct formant structure, which is characteristic of breathiness. F0 was consistently lower after ‘voiced’ stops only for [a]. Other vowels displayed different properties. [] was produced with lower f0 after a velar, but with higher f0 after a bilabial stop. In [u], f0 was higher for both bilabial and velar stops. Lowering of F1 after voiced (slack) stops was not found in all vowels. F1 was higher for [u] after ‘voiced’ stops. F2 was higher after ‘voiced’ stops in all vowels.

This kind of vowel alternation, defined as ‘register’ difference, is found in other languages. Gregerson (1976) reports that Mon-Khmer has two sets of vowels: ‘head’, or more open vowels pronounced with higher pitch and tense clear voice, and ‘chest’, or close ‘breathy’

vowels pronounced with lower pitch, lax voice, and lowered larynx. The two registers in Mon- Khmer are also related to laryngeal specifications of preceding consonants. ‘Head’ register is found after voiceless stops, and ‘chest’ register occurs after voiced stops and sonorants.

2.3. Voicing contrast in Madurese

Madurese has a three-way contrast between voiceless unaspirated, voiceless aspirated, and voiced stops. According to Cohn (1993b, p. 114), mean VOT in voiced stops is 4 ms for /b/ and 2 ms for /d/; mean VOT in voiceless unaspirated stops is 11 ms for /p/, 14 ms for /t/ and 26 ms for /k/; mean VOT in voiceless aspirated stops is 50 ms for /p^h/ and 59 ms for /k^h/. Unfortunately, Cohn’s study does not have measurements of voicing during the closure. Spectrograms presented in Cohn (1993b) show that Madurese voiced stops are voiced in the intervocalic position.

However, it is not clear whether they are prevoiced word-initially; positive VOT values in voiced stops suggest that these stops in many cases must be realized without robust prevoicing.

These observations challenge the claim that Madurese has a series with true voiced stops.

Jessen (1998) argues that if a language has a contrastive feature [voice], it implies that voicing

must be present in the articulation of voiced stops and fricatives. Van Alphen and Smits (2004) report that prevoicing is the most important cue for voiced stops both in perception and production in Dutch. Lisker (2003) shows that prevoicing is the most important cue for voiced stops in Russian, another language with the contrastive feature [voice]. If ‘voiced’ stops in Madurese (as well as in Javanese) do not have robust prevoicing, then the contrast in this phonation type must be realized with features other than [voice].

Madurese vowels show height alternation after voiceless and voiced stops (Cohn 1993a, 1993b; Cohn and Lockwood 1994). Cohn and Lockwood (1994) argue that vowel alternation in Madurese is phonological. Vowels are divided into two distinct sets. ‘High’ vowels /i, ɨ, ɤ, u/

occur only after voiceless aspirated and voiced stops. ‘Non-high’ vowels /ɛ, ə, a, ɔ/ occur elsewhere: after voiceless unaspirated stops, fricatives, and sonorants (4).

(4) maca [maca] ‘read’ AV baca [bɤca] ‘read’ BF

The current phonological analysis of vowel alternation in Madurese is controversial.

Cohn (1993b) argues that high vowels are pronounced with lowered larynx. Therefore, the phonological process in Madurese can be explained as spreading of the feature [Lowered Larynx] ([LL]). Trigo (1991), in contrast, argues that the analysis of vowel alternation in Madurese must include two features: [Advanced Tongue Root] ([ATR]) for voiced stops and [LL] for voiceless aspirated stops. One difference between [ATR] and [LL] is in the vowel quality: [ATR] results in lowering of F1 and raising of F2, while [LL] results in lower values of both formants.

3. Experiment

The goal of the experiment was to obtain acoustic measurements for voiced and voiceless stops and the vowels that follow those stops.

3.1. Participants

Two native speakers of Sundanese participated in the experiment. The first speaker (EK) was a 27-year-old male born in Garut (West Java). Sundanese is his first language, although he also speaks Indonesian and English. The second speaker (MK) was a 27-year-old female born in Bandung (West Java). Sundanese is her first language, and she also speaks Indonesian and English. Both speakers maintain everyday communication in Sundanese in the family. They were not aware of the goal of the experiment.

3.2. Stimuli and recording

Seven Sundanese vowels /i, ɛ, ə, a, ɨ, ɔ, u/ were recorded in an anechoic chamber as pronounced by the speakers after voiced and voiceless stops /p/, /b/, /t/, /d/, /k/, /g/ in five types of syllables:

__V#, __Vr, __Vk, __Vŋ, and __Vh. The speakers read the list twice. The syllables were written in standard Sundanese orthography. The list also included distracters: disyllabic words and

syllables with fricatives. The stimuli were digitally recorded using a one-point condenser SONY ECM-MS907 microphone and a DELTA-66 soundcard at 22,100 Hz. The recordings were then marked and measured using the speech analysis software package PRAAT (Boersma &Weenink 2009).

3.3. Measurements

Positive VOT values for voiceless stops were taken from the point of the stop release to the onset of the second formant of the vowel. Negative VOT values for voiced stops were taken from the point at which low energy of vocal fold vibration started to the point of the stop release. F0 was taken at the onset of the vowel [a].

Energy values (dB) for the first (H1) and the second (H2) harmonics and the peak harmonics forming the first (F1) and the second (F2) formants were taken at the onset of the vowel [a] using FFT spectra with a 25 ms Hamming window (40Hz bandwidth). The frequencies of the first (F1) and the second (F2) formant were taken at the midpoint of a vowel where the formant structure was relatively stable.

4. Results

4.1. VOT

Table 1 shows that voiced and voiceless stops in Sundanese exhibit a pattern found for other true voice languages: voiceless stops have short-lag VOT and voiced stops are pronounced with robust prevoicing.

Table 1. Mean VOT values (ms) and standard deviations of initial voiced and voiceless Sundanese stops for the two speakers

EK (m) MK (f)

Voiceless Voiced Voiceless Voiced

Bilabial 23 (5.8) –66 (17.6) 10 (1.6) –47 (11.4) Alveolar 22 (3.5) –83 (24.9) 12 (3.2) –54 (18.5)

Velar 39 (6.5) –25 (34) 34 (7.8) –63 (40.1)

Total 28 (5.3) –58 (25.5) 18.7 (4.2) –54.7 (23.3)

The two speakers did not differ in production of voiced stops [F(1,68) = 0.123, p = 0.727]; however, speaker EK pronounces voiceless stops with significantly longer VOT than speaker MK [F(1,70) = 12.0, p < 0.01].

Place of articulation affects VOT values of voiceless stops [F(2,69) = 63.3, p <0.001].

Both speakers pronounced [p] and [t] with significantly shorter VOT than [k] (p<0.001). This pattern is consistent with other cross-linguistic findings (e.g. Lisker and Abramson 1964).

Voiced stops exhibit more variation. Speaker EK had longer prevoicing in bilabial and alveolar stops than in dorsal stops [F(2,33) = 17.236, p < 0.001]. For speaker MK, no significant difference in prevoicing between three places of articulation was found (p = 0.387).

a. [ba] b. [ga]

Figure 2. The waveforms and spectrograms of the bilabial (implosive) and velar (modal voice) word-initial stops preceding /a/ (EK)

Observation of waveforms of the voiced stops showed that they were pronounced in two distinct fashions. The difference is illustrated in Fig. 2. Some stops (most of [d]’s and [g]’s in the data) were produced with prevoicing in which amplitude of vocal fold vibration decreases toward the release point (Figure 4b). Other stops (predominantly [b] and occasionally [d]) are pronounced with increasing amplitude of vibration closer to the release point (Figure 4a). Lindau (1984) argues that such changes in amplitude are characteristic of implosive stops. Increased amplitude of vibration indicates that the lowering of the larynx was sufficient to counteract the pressure building in the oral cavity. Note that it is primarily bilabial stops and partly alveolar stops that are implosive in Sundanese. Most velar stops are pronounced with modal voice. This is consistent with the cross-linguistic pattern according to which languages with implosive stops tend to have them at bilabial and alveolar places of articulation (Ladefoged and Maddieson 1996).

4.2. Fundamental frequency

Results of a two-way ANOVA (Voice, Place) for speaker MK showed a significant effect of Voice [F(1,66)=61.59, p < 0.001] and interaction [F(2,66)=5.65, p < 0.01] but no effect of Place [F(2,66) < 1]. F0 was lower after voiced stops (M=198Hz, SD=9.2) than after voiceless stops (M=216Hz, SD=10.3), and the difference was the greatest for velars (27Hz). No significant effects were obtained for speaker EK [Voice: F(1,64)=1.74, p=0.193; Place: F(2,64)<1;

interaction: F(2,64)<1]. The results are summarized in Fig. 3.

Figure 3. Mean f0 values after voiceless and voiced stops for speakers EK and MK.

Significant differences between bars at α=0.05 are marked with an asterisk(*).

4.3. Spectral energy (H1-H2, H1-F1, H1-F2)

The results show (Fig.4) that both speakers exhibited differences in phonation of vowels after voiceless and voiced stops. Differences in two spectral characteristics (H1-H2 and H1-F2) reached the significance level for both speakers. A two-way ANOVA (Voice, Place) with H1-H2 as a dependent variable showed a significant main effect of Voice [EK: F(1,64)=22.93, p <

0.001; MK: F(1,64)=56.92, p < 0.001]. The effect of Place was significant for speaker MK [F(2,64)=24.81, p < 0.001], but only marginally significant for speaker EK [F(2,64)=2.97, p=0.058].

Figure 4. Energy values for H1-H2, H1-F1, H1-F2 at the onset of [a] for speakers EK and MK. Significant differences between bars at α=0.05 are marked with an asterisk (*).

Vowels exhibited smaller differences in amplitude in the lower range of the spectrum after voiced stops (-0.6 dB) than after voiceless stops (3.2 dB), suggesting that they were pronounced with a smaller open quotient after voiced stops. These differences were smaller in bilabials (3.6 dB) and in dentals (3.3 dB) than in velars (6.2 dB).

The effect of Voice was significant on H1-F2 for both speakers [EK: F(1,64)=5.99, p <

0.05; MK: F(1,64)=7.78, p < 0.01] as well as the effect of Place [EK: F(2,64)=9.69, p < 0.001;

MK: F(2,64)=11.43, p < 0.001]. Vowels had a steeper spectral slope in the higher range of the spectrum after voiced stops (16.2 dB) than after voiceless stops (13.6 dB), suggesting that vocal fold closure was more abrupt after voiced stops. Voicing in a consonant induced a breathier mode of vocal fold vibration in the following vowel. These differences were greater in bilabials (4.5 dB) and velars (2.6 dB) than in dentals (0.7 dB).

No distinct patter was found for H1-F1. The effect of Voice was not obtained for either speakers [EK: F(1,64)=2.79, p=0.099; MK: F(1,64)=3.76, p=0.057].

4.4. Formants (F1, F2)

The plotted F1 and F2 mean values for the seven vowels after voiceless and voiced stops are presented in Fig. 5 (speaker EK) and Fig. 6 (speaker MK).

Figure 5. Plotted F1 and F2 values for seven Sundanese vowels after voiceless and voiced stops (speaker EK). Arrows show significant raising and fronting of vowels after voiced

stops.

pa ka ta

t p

g

k

t p pE

kE tE

pu

tu ku

ki ti

pi

p t

k

da ba ga

b d

k d g b

dEbE gE

bu du

gu di bi

gi

d b

g

300

400

500

600

700

800 700 900

1100 1300

1500 1700

1900 2100

2300 2500

F2, Hz

Voiceless Voiced

Mean Voiceless Mean Voiced

F1, Hz

Speaker EK

To investigate whether changes in formant values at the midpoint of a vowel are determined by the environment, a multivariate (F1, F2) factorial 3 (Onset) x 5 (Coda) ANOVA was performed for each speaker. No main effect of Onset [EK: F1: F(2,195)=1.16, p=0.317; F2:

F(2,195)<1; MK: F1: F(2,195) < 1; MK: F2: F(2,195) < 1] or Coda [EK: F1: F(4,195) < 1; F2:

F(4,195)<1; MK: F1: F(4,195) <1; F2: F(4,195) <1] was obtained for either formant. Variation in formants was not affected by the place of articulation of the surrounded consonant.

Figure 6. Plotted F1 and F2 values for seven Sundanese vowels after voiceless and voiced stops (speaker MK). Arrows show significant raising and fronting of vowels after voiced

stops.

The effect of voicing of the preceding consonant on F1 and F2 was analyzed using a multivariate ANOVA (Voice, Vowel). Separate analyses were performed for each speaker. A multivariate factorial 2(voice) x 7(vowel) ANOVA found a significant effect of Voice [EK: F1:

F(1,196) = 99.97, p <0.001; F2: F(1,196) = 31.35, p < 0.001; MK: F1: F(1,196) = 113.33, p

<0.001; F2: F(1,196) = 42.26, p < 0.001] and Vowel [EK: F1: F(6,196) = 522.21, p <0.001; F2:

F(6,196) = 796.89, p < 0.001; MK: F1: F(6,196) = 484.59, p <0.001; F2: F(6,196) = 1224.94, p <

0.001]. The interaction between voice and vowel for F2 was not significant for speaker EK [F(6,196)=1.78, p=0.105], but it was significant for speaker MK [F(6,196)=3.92, p < 0.01]. The interaction between voice and vowel for F1 was significant for both speakers [EK: F(6,196) = 7.34, p < 0.001; MK: F(6,196) = 4.15, p < 0.01]. Speakers consistently pronounced vowels with

ta pa ka t p

g

k

t

p

pE kE tE

tu puku piti

ki

p t_

k_

da ga ba d b

k_

b d g

bE gE dE

du bu bi gu

di gi

b_ d

g_

300

400

500

600

700

800

900 800 1100

1400 1700

2000 2300

2600 2900

F2 (Hz)

Voiceless Voiced

Mean Voiceless Mean Voiced

F1, Hz

Speaker MK

lower F1 and higher F2 after voiced stops. Changes in F1 and F2 after voiced stops were greater in low vowels. To investigate this effect, separate ANOVAs were performed for each vowel.

Seven multivariate ANOVAs with F1 and F2 as dependent variables were performed to assess the effect of Voice for each vowel. The summary of the analysis is shown in Table 2. For each speaker, the main effect of Voice was significant on F1 for all vowels except [i]. The main effect of voice on F2 was significant in non-high vowels [ɛ, ɔ, a, ə], but it was not significant in high vowels [i, ɨ, u]. Thus, non-high vowels in Sundanese show a similar pattern of raising and fronting after voiced stops.

Table 2. Mean F1 and F2 differences (Hz) and after voiced and voiceless stops in seven Sundanese vowels for speakers EK and MK. St. errors are given in brackets. Significant

results for both F1 and F2 are given in bold.

F1 F2

EK Δ (st.error) F(1,28) p (sig.) Δ (st.error) F(1,28) p (sig.)

i -16.0 (11.6) 1.91 ns 4.1 (30.7) 0.18 ns

ɨ -9.1 (4.3) 4.39 0.05 34.4 (34.2) 1.01 ns

u -13.6 (6.0) 5.12 0.05 60.8 (32.7) 3.46 ns

ə -43.8 (13.0) 11.30 0.01 98.3 (48.8) 4.06 0.05 ɛ -49.6 (9.8) 25.45 0.0001 131.6 (30.9) 18.17 0.0001 a -80.3 (11.2) 50.96 0.0001 110.5 (28.6) 14.87 0.001 ɔ -34.5 (5.3) 42.06 0.0001 58.3 (23.9) 5.93 0.05 MK

i -20.5 (10.7) 3.72 ns 1.3 (26.3) 0.002 ns

ɨ -25.5 (6.1) 17.58 0.0001 58.4 (33.5) 3.03 ns u -29.5 (6.4) 21.42 0.0001 12.7 (24.3) 0.275 ns ə -68.4 (24.5) 7.82 0.01 136.3 (43.9) 9.65 0.01 ɛ -81.2 (11.6) 49.21 0.0001 157.1 (46.9) 11.23 0.01 a -75.7 (4.7) 261.9 0.0001 153.5 (30.2) 25.91 0.0001 ɔ -49.7 (11.8) 17.84 0.0001 55.5 (19.6) 8.01 0.01

The obtained values for /a/ are similar to Fagan’s results for Javanese, where at the steady state the mean difference in F1 for [a] was 86.3 Hz, and the mean difference in F2 was 84.7 Hz.

Smaller differences in high vowels are partly consistent with Thurgood’s study, in which Javanese /u/ was found to be lowered after voiced stops whereas /o/ and /a/ were raised.

4.5. Interim conclusion

The results of acoustic measurements of stops and vowels suggest that the laryngeal contrast in Sundanese has redundant phonetic properties. In addition to robust prevoicing in initial voiced stops, the contrast is enforced by changes in the vowel quality. The summary of these properties is given in Table 3.

Table 3. Properties of the voicing contrast in Sundanese

Voiceless Voiced

VOT short-lag negative

f0 higher lower

H1-H2 greater smaller

H1-F1 no pattern

H1-F2 smaller greater

F1 higher lower

F2 lower higher

Vowels after voiced stops show changes in spectral characteristics and height. Significant differences in H1-H2 suggest that phonation of such vowels has a relatively smaller open quotient, which indicates more ‘pressed’, or laryngealized voicing with vocal folds remaining closed for a longer time as compared with modal voicing. Two other findings about voicing in Sundanese stops should be noted here. First, laryngealized voicing in vowels was found to be more prominent after bilabial stops and less prominent after velars. Second, these differences in phonation correlate with implosion observed in waveforms of voiced stops. Recall that the majority of implosive stops were found among bilabials whereas velar stops were generally pronounced with modal voice.

To assess the relationship between implosion in voiced consonants and phonation of the following vowel, we performed a hierarchical regression analysis with H1-H2 as a dependent variable. A voiced bilabial stop [b] pronounced before [a] was marked as implosive or plosive based on inspection of a waveform. This variable was entered first. Other independent variables were VOT, f0, H1-F1, and H1-F2 entered in this order. The summary is given in Table 4.

Implosion was the only variable that explained a significant proportion of variance in H1-H2 (37.6%). Although the overall model was significant [F(5,11)=3.22, p < 0.05], entering VOT, f0, H1-F1, and H1-F2 did not significantly improve the model.

Table 4. A regression analysis examining H1-H2. The significant contributor to the model is marked in bold.

Effect R² R² change F change Sig.

1. Implosion 0.376 0.376 9.031 .009

2. VOT 0.449 0.073 1.862 .194

3. f0 0.462 0.013 0.306 .590

4. H1-F1 0.474 0.013 0.286 .603

5. H1-H2 0.594 0.120 3.250 .099

Significant differences in higher frequencies of the spectrum between vowels pronounced after voiceless and voiced stops suggest that voicing in consonants affected the mode of vocal

fold vibration. Vowels after voiced stops had more gradual adduction of the vocal folds, which is usually interpreted as characteristic of breathy voice (e.g. Stevens 1999, Cho et al. 2002). In addition, vowels after voiced stops were consistently pronounced with lower F1. The two properties are also found in other languages that have breathy phonation in vowels after voiced stops, e.g. Javanese.

The combination of acoustic properties found in voiced stops – implosion, F1 lowering, – strongly suggests that lowering of the larynx occurred systematically in production of voicing (Ladefoged and Maddieson 1996). We conclude that the voicing contrast in Sundanese has redundant phonetic properties. Two acoustic correlates of voicing – prevoicing and larynx lowering, – were found in voiced stops and the following vowel. In addition, some properties of breathy voice were also found in vowels after voiced stops.

5. Discussion and conclusions

The results show that the voicing contrast in Sundanese is different from other closely related languages, yet it shares phonetic properties of voicing found in these languages. The summary for the voicing contrasts in Sundanese, Indonesian, Madurese, and Javanese is shown in Table 5.

Table 5. Typology of voicing properties in Western Indonesian languages. ‘T’

represents a fortis (voiceless) stop; ‘D’ represents a lenis (voiced) stop. Empty cells indicate that no relevant data is available.

Sundanese Indonesian Madurese Javanese

/t/-/d/ /t/-/d/ /t/-/d/-/t^h/ /t/ stiff - /d/ slack

f0 T > D T > D T > D, T^h T ≥ D

H1-H2 T > D no pattern no pattern

H1-F1 no pattern T < D T < D

H1-F2 T < D no pattern T < D

F1 T > D T > D, T^h T > D

F2 T < D no pattern T < D

(non-high (all vowels

vowels) ‘register’)

According to the typology of phonological contrasts suggested by Iverson and Salmons (1995), Sundanese is closer to Indonesian, as both languages have a contrast between phonologically unspecified voiceless unaspirated stops and fully voiced stops specified with [voice]. Yet, alternation in vowel height makes it close to Madurese and Javanese, in which vowels are pronounced with lower F1 after voiced stops. Madurese, however, is different from Sundanese in terms of implementation of vowel raising. While only non-high vowels are raised and fronted after voiced stops in Sundanese, height alternation is phonological in Madurese.

High vowels occur after voiced and voiceless aspirated stops; low vowels occur after voiceless stops and other consonants. Vowel height alternation in Sundanese is very similar to what is found in Javanese; however, the laryngeal contrast in Javanese is between stiff and slack stops, which do not usually engage voicing word-initially.

The question why the voicing contrast in Sundanese has redundant phonetic properties is not easy to answer. Several scenarios can be explored here. Comparison between voicing in Sundanese and Madurese suggests an explanation of why voiced and voiceless aspirated stops in Madurese pattern together in vowel height alternation. Historically, Madurese voiceless aspirated stops derived from voiced stops (Stevens 1966). The Sundanese example suggests that voiced stops in proto-Madurese could have breathy phonation. The scenario in which modern Madurese voiceless aspirated stops lost voicing and inherited breathiness while voiced stops inherited voice and lost breathiness does not seem implausible. It is not clear, however, if modern Sundanese exhibits the initial stage of this process or the change in progress.

Redundancy in voicing properties in Sundanese can be a result of Javanese influence.

Adisasmito-Smith (2004) argues that bilingual Indonesian-Javanese speakers develop greater H1-F2 after voiced stops in Indonesian. Thus, greater H1-F2 in vowels after voiced stops in Sundanese may also be influenced by Javanese.

Although the reasons for phonological redundancy in Sundanese are not clear, the study provides evidence on consonant-vowel interaction in laryngeal contrasts. It is not unusual to find laryngeal contrasts that are enhanced¹ by other phonological features. Observations of cross- linguistic data reveal several possibilities. Vowel length, which is an independent contrastive phonological feature, can also enhance the contrast between voiced and voiceless stops (Cho 1976). In English, vowels before voiceless stops are about 50% shorter than before voiced stops (Peterson and Lehiste 1960). Voice is often found in languages which use other phonological features for a laryngeal contrast. For example, English and German, which use a contrastive phonological feature [spread glottis], have voiced stops in the intervocalic position or word- finally (in English). Thus, voice is used to enhance the contrast in positions where aspiration alone may be insufficient.

The analysis of the voicing contrast in Sundanese reveals patterns that are also found across languages. The lowering of F1 (vowel raising) and breathiness in vowels after voiced stops is found in Javanese and Mon-Khmer. The lowering of the larynx, which is a phonological feature that specifies implosive stops (e.g. in Degema), is also used as a phonetic feature that enhances and facilitates voicing in stops (Westbury 1983). In addition, implosion of voiced stops in Sundanese is more frequent in bilabials and dentals, which is consistent with cross-linguistic patterns (Ladefoged and Maddieson 1996).

This paper is a preliminary acoustic analysis of voicing in Sundanese, and it is far from being comprehensive. Any plausible phonological analysis of the Sundanese data should take into account other facts that were not examined in the paper, e.g. voicing in intervocalic and word-final stops.

References

Adisasmito-Smith, Niken. 2004. Phonetic and phonological influence of Javanese on Indonesian.

Doctoral dissertation, Cornell University, Ithaca, NY.

Cho, Taehong, Sun-Ah Jun, and Peter Ladefoged. 2002. Acoustic and aerodynamic correlates of Korean stops and fricatives. Journal of Phonetics 30:193–228.

1 Feature enhancement is discussed at length in Stevens and Keyser (1989).

Cohn, Abigail C. 1993a. Voicing and vowel height in Madurese: a preliminary report. Oceanic Linguistics Special Publications 24, Tonality in Austronesian Languages, 107–121.

Cohn, Abigail C. 1993b. Consonant-vowel interaction in Madurese: The feature Lowered Larynx. CLS, 29.

Cohn, Abigail C. and Catherine Lockwood. 1994. A phonetic description of Madurese and its phonological consequences. Working Papers of the Cornell Phonetics Laboratory 9:67–

92.

Fagan, Joel L. 1988. Javanese intervocalic stop phonemes: The light/heavy distinction. Studies in Austronesian Linguistics 76, ed. Richard McGinn, 173–202. Athens, Oh.

Gregerson, Kenneth J. 1976. Tongue-root and register in Mon-Khmer. Austroasiatic studies. Part 1, eds. P. N. Jenner, L. C. Thompson & S. Starosta, 323–369. University Press of Hawaii.

Hayward, Katrina. 1993. /p/ vs. /b/ in Javanese: Some preliminary data. SOAS Working Papers in Linguistics & Phonetics, vol. 3. 1–33.

Iverson, Gregory K. and Joseph C. Salmons. 1995. Aspiration and laryngeal representation in Germanic. Phonology 12:369–396.

Jessen, Michael. 1998. Phonetics and phonology of tense and lax obstruents in German. John Benjamins.

Kurniawan, Eri. 2009. Sundanese stops and typology of stop contrasts in other Javanic languages. Paper presented at the seminar on Laryngeal Phonology. Iowa City, Ia.

Ladefoged, Peter and Ian Maddieson, 1996. The sounds of the world’s languages. Blackwell Publishers.

Lisker, Leigh. and Arthur S. Abramson. 1964. A cross-language study of voicing in initial stops:

Acoustic measurements. Word 20:384-422.

Lindau, Mona. 1984. Phonetic differences in glottalic consonants. Journal of Phonetics 12:147–

155.

Müller-Gotama, Franz. 2001. Sundanese. München: Lincom Europa.

Peterson, Gordon E. and Ilse Lehiste, 1960. Duration of syllable nuclei in English. Journal of the Acoustical Society of America 32:693–703.

Stevens, Alan M. 1966. The Madurese reflexes of Proto-Malayopolynesian. Journal of the American Oriental Society 86:147-56.

Stevens, Kenneth N. 1999. Acoustic phonetics. Cambridge: MIT Press.

Stevens, Kenneth N. and Samuel Jay Keyser. 1989. Primary features and their enhancement in consonants. Language, 65:81-106.

Sudaryat, Yayat. 1985. Pedaran basa Sunda. [Sundanese grammar]. Bandung: Ekonomi.

Thurgood, Ela. 2004. Phonation types in Javanese. Oceanic Linguistics 43:277–295.

Trigo, Loren. 1991. On pharynx-larynx interactions. Phonology 8:113–136.

Van Syoc, Wayland Bryce. 1959. The phonology and morphology of Sundanese. Doctoral dissertation, University of Michigan, Ann Arbor, Mi.

Westbury, John R. 1983. Enlargement of the supraglottal cavity and its relationship to stop consonant voicing. Journal of Acoustic Society of America 73:1322–1336.