the phonology and phonetics of epenthetic vowels … phonology and phonetics of epenthetic vowels in...

The Phonology and Phonetics of Epenthetic Vowels

in Korean Loanwords Hyun-ju Kim

Stony Brook University

1. Introduction

Vowel epenthesis to repair illicit syllable structures is common in Korean loanword

adaptation. Since tautosyllabic consonant clusters are prohibited in Korean, consonant

clusters must be broken by inserting a default vowel //1 (e.g. milkh ‘milk’, riphth ‘lift’).

Due to the restriction that coda consonants in Korean must not be released, consonants

with release are realized as an unreleased consonant in a coda position in native

phonology. For example, fricatives or affricates such as /s, c, ch, h / in the underlying

form are realized as an unreleased stop [t] in a coda position (e.g. /kkoch/ [kkot]

‘flower’). However, in loanwords, release of coda consonants in source words is

preserved by parsing them as onsets via vowel epenthesis (e.g. kas ‘gas’, buch ‘boots’).

Epenthetic vowels tend to resist accent in Kyungsang Korean (KK) loanwords

(Kenstowicz and Sohn 2001; Broselow, to appear; Rhee and Kim 2003). Kyungsang

Korean, which is spoken in southeastern Korea, is a pitch accent language where the

locus of a high pitch accent is determined lexically, as in Japanese. However, words

whose accent is not specified in the lexicon follow a default accent pattern, which is

penultimate accent (Kim 1997). Therefore, there are two groups in the KK native lexicon

with respect to accent assignment: one is a free accent group, whose accent is lexically

specified (e.g. kamani ‘rice sack’, satari ‘ladder’) and the other is a default accent group

with penultimate accent (e.g. apci ‘father’, kosmtochi ‘porcupine’) 2. Accentuation of

1 // will be used for a high back vowel in Korean throughout the paper although it is pronounced most close to the value of [] based on IPA (Lee 1996). 2 At least in North Kyungsang Korean, a default penultimate accent shows up clearly when it comes to words which consist of more than three syllables. Words longer than trisyllabic get accented on the penult consistently (e.g. kosmto chi ‘porcupine’). In addition, evidence of the default accent comes from

Hyun-ju Kim ________________________________________________________________________

KK loanwords is predictable in words consisting of only light syllables3, following a

default accent pattern where the penultimate syllable is accented:

(1) Penultimate accent in loanwords

ikha ko ‘Chicago’

kai t4 ‘guide’

rati o ‘radio’

ameri kha ‘America’

However, accent does not fall on the penult if the vowel in the penult is epenthetic (2).

(2) Non-penultimate accent with epenthetic vowels

thosth ‘toast’

pe sth ‘best’

sthro ‘straw’

methro ‘metro’

This failure to accent the penultimate vowel in (2) cannot be attributed to the presence of

the vowel // in the penultimate syllable because the lexical high back vowel in native

words may be accented (3).

alternative accent patterns: words whose initial syllable is accented in a free accent group allow the penult to be accented alternatively as shown in the following examples (Kim 1997). i) kpuki ≈ kpuki ‘turtle’ kamani ≈ kamani ‘straw rice-bag’ kamulchi ≈ kamu lchi ‘mullet’ to ksuri ≈ toksuri ‘eagle’ 3 When words contain closed syllables, penultimate accent may be disturbed due to weight sensitive accent assignment. However, accent patterns in loanwords are still predictable. (For more details, see Kenstowicz and Sohn 2001) 4 The vowel sequences a.i in ka.i.t. and i.o in ra.ti.o. are not diphthongs in Korean and each vowel is the nucleus of a syllable.

2

The Phonology and Phonetics of Epenthetic Vowels in Korean Loanwords

________________________________________________________________________ (3) Penultimate accent with lexical vowels

snim ‘monk’

ssro ‘for oneself’

hrta ‘to flow’

aphta ‘to be sick’

Why do epenthetic vowels behave differently from lexical vowels? Is it because they are

different phonetically from lexical vowels or because the phonological grammar forces

them not to be accented?

In order to answer the questions raised above, I investigate the acoustic quality of

epenthetic vowels in KK loanwords in this study. The rest of the paper is organized as

follows: Section 2 presents hypotheses concerning the acoustic quality of epenthetic

vowels. Section 3 presents an acoustic study of epenthetic vowels in KK. Section 4

discusses the findings of the accent-epenthesis interaction. Section 5 is the conclusion.

2. Hypotheses and implications

Two hypotheses guided an acoustic study of epenthetic vowels in KK as follows:

(4) Hypothesis I:

The epenthetic vowel is phonetically different from the corresponding lexical

vowel: the epenthetic vowel is intrinsically shorter and less prominent.

Hypothesis II:

The epenthetic vowel is phonetically the same as the corresponding lexical

vowel.

Under Hypothesis I that epenthetic vowels are not phonetically identical to lexical

ones, epenthetic vowels may be distinguishable from lexical vowels in the acoustic signal.

Several previous studies actually reported that epenthetic vowels in other languages were

3

Hyun-ju Kim ________________________________________________________________________

phonetically not the same as lexical vowels in that they were less prominent than lexical

vowels. Gouskova and Hall (to appear) found that epenthetic and lexical vowels [i] in

Lebanese Arabic were acoustically distinct for some speakers: epenthetic vowels were

either shorter in duration or backer (lower in F2) or both. Davidson (2006) suggests that

inserted schwa in production of non-native sequences by English speakers is not an

epenthetic vowel but a transitional vowel, which is not a real vowel but a “transitional

vocoid” appearing on the acoustic signal due to inadequate gesture coordination. She

reports that the acoustic properties of transitional schwa are a shorter duration and a

lower F1 (higher in height), suggesting that speakers are not repairing phonotactically

illegal sequences with an epenthetic vowel, but rather that they are unable to adequately

coordinate the two consonant gestures. Hall (2006) proposes diagnostics to distinguish

transitional vowels (or intrusive vowels) from epenthetic vowels. According to Hall,

transitional vowels are non-syllabic, invisible phonologically, schwa-like or a copy of a

nearby vowel, occur in heterorganic clusters, highly variable in duration or disappear at

fast speech rates, and they do not repair illicit structures. On the other hand, epenthetic

vowels repair a structure that is marked, the vowel quality may be fixed or copied (need

not be schwa), and they do not disappear in faster speech rates.

In KK, if epenthetic vowels, high back vowels, are shorter in duration and more

centralized (more fronted, higher F2, or lower, higher F1) than lexical vowels, it is

plausible to assume that epenthetic vowels are transitional rather than epenthetic. Here is

a possible scenario: KK first adapters simply produced loanwords like thosth5

‘toast’ with

a transitional vowel. Subsequent listeners were also sensitive to the acoustic differences

between lexical vowels and transitional vowels and preserved the differences. If so,

epenthetic vowels are unaccentable because they are not real vowels, so they are invisible

to phonological processes.

On the other hand, following Hypothesis II, if epenthetic vowels are phonetically

identical to lexical vowels, epenthetic vowels should be part of the phonological

representation. Then, the accent-epenthesis interaction must be explained by some

abstract formal account. Positing abstract underlying representations for epenthetic

vowels is necessary to account for the accent-epenthesis interaction since epenthetic 5 Subscript ‘’ stands for a transitional vowel, .

4


________________________________________________________________________ vowels are not distinguishable from lexical vowels in the surface form. For example,

Alderete (1999) proposes a formal account for the accent-epenthesis interaction by using

the constraint HEADDEP, which does not allow epenthetic vowels to be in the head

position of a foot.

(5) HEADDEP (Alderete 1999)

Nonlexical vowels are not allowed in prosodic heads.

(epenthetic vowels cannot be the head of a foot)

The constraint HEADDEP, ranked higher than other accent constraints, forces epenthetic

vowels not to be accented, so it derives non-penultimate accent patterns when

penultimate vowels are epenthetic as in thosth ‘toast’ (cf. penultimate accent in words

without epenthetic vowels: aphta ‘to be sick’). The constraint HEADDEP presupposes that

speakers access the underlying status of epenthetic vowels in order not to put accent on

epenthetic vowels. However, this raises the question of how learners know that the

vowels are epenthetic when there is no cue to the contrast between epenthetic vowels and

lexical vowels in the acoustic signal.

3. An acoustic study of epenthetic vowels in Korean loanwords

An acoustic study of epenthetic vowels in Korean loanwords was designed in order to

test the two hypotheses outlined in the previous section: i) The epenthetic vowel is

intrinsically shorter and less prominent than the corresponding lexical vowel; ii) The

epenthetic vowel is phonetically the same as the corresponding lexical vowel. The study

incorporates online adaptation of nonce English words. By studying how Korean

speakers produce illegal consonant clusters contained in pseudo-words written in English,

I examined whether vowels inserted in online adaptation are real vowels, which are

phonologically inserted, or transitional, which appear due to failure of coordinating the

two consonant gestures.

5

Hyun-ju Kim ________________________________________________________________________

3.1 Experimental Design

3.1.1 Participants

I recruited three male and three female speakers of Kyungsang Korean, all

graduate/undergraduate students at Stony Brook University in New York, who came from

Busan, a city located in the southern part of Kyungsang province. Their ages ranged from

23 to 31 (mean 28). The length of stay in the U.S. ranged from 4 months to 5 years. In

order to make sure that the KK speakers shared the same accent patterns, test words

including both native words such as slphta ‘to be sad’ or kophta ‘to be hungry’ and

loanwords such as kiphth ‘gift’ or ri phth ‘lift’ were elicited. They all showed the same

accent patterns.

3.1.2 Materials

The word list consisted of 80 words, including 9 test loan words, 16 test native words, 25

nonce words and 30 fillers. Most test words were CVCVCV, and the second vowel was a

target vowel // to be measured (epenthetic vs. lexical). Nonce words were included in

order to see how vowel epenthesis was realized phonetically in online adaptation. The

quality of inserted vowels in the nonce words was compared to the corresponding lexical

vowels in native words and to epenthetic vowels in lexicalized loanwords. The compared

pairs were matched for the quality of the preceding and the following vowels in order to

avoid any influence from vowel coarticulation. Also, the consonant following the target

vowel was the same in matched pairs to avoid effect of the quality of the consonant on

the preceding vowel's quality or duration. Since perfect minimal pairs of loanwords and

native words are extremely rare in the Korean lexicon, a test of nonce words provides a

way to control immediate segmental context for comparison between epenthetic vowels

and lexical vowels. The nonce forms were only perfect matches in consonants

surrounding the target vowel and vowels on either side of those consonants; there were

6


________________________________________________________________________

some differences in word initial consonants. The spelled and expected phonetic form of

nonce words is given below.

(6) Test word sets6:

a. Nonce words7 Native words

cafda [kaphda] aphda 'to be sick'

bofgo [pophgo ] kopho 'hungry and '

cafsin [kaphsi n] aphsin ‘sick-infix(honorific)’

tagda [thagda] tamgda 'to soak'

lusni [lusni] wu sni ’lauh-suffix’

b. Nonce words Loanwords

vift [piphth ] kiphth ’ift’

boft [pophth] so phth ’soft’

fost [pho sth ] ko sth ’host’

bogma [pogma] togma ’doma’

3.1.3 Procedure

Participants were recorded in a sound-treated room using a Shure SM 48 microphone and

a Marantz PMD 660 recorder at 44.1 kHz sampling rate. They read a randomized list of

words given in a Korean carrier sentence, “nann __________ puth malhætta” (I said

from the word of ______ ) with 5 repetitions. Only three repetitions were analyzed; the

first and the last were excluded. Nonce words were introduced as newly imported product

names and asked subjects to read them in a way that they considered to be comfortable in

the native language, although nonce forms were written in English. Each speaker looked

6 All test words used in the experiment are given in the appendix I. There were some differences in accent between the nonce words set and the native words set. However, I found that there was no accent effect on vowel duration in my other study. For more details, see the appendix II. 7 Fricatives and voiced obstruents such as f, s, g were used as stimuli for vowel insertion since fricatives are not possible coda consonants in Korean and voiced obstruents also trigger vowel epenthesis in order to preserve the feature [voice] (cf. Kang 2003).

7

Hyun-ju Kim ________________________________________________________________________

through the word list in order to familiarize himself/herself with all the words before

recording. Speakers were asked to read in their own dialectal accent.

3.2 Measures taken

A total of 900 tokens (50 test words × 6 subjects × 3 reps) were measured. All

measurements were taken using Praat (Boersma and Weenink 2005). Three duration

measurements were taken: the target vowel (the second vowel in each test word) in the

token; the consonant preceding the target vowel; aspiration of the consonant. Vowels

were segmented manually by examining the spectrograms and waveforms. The vowel

boundary was determined by the presence of clear formant structure and a sharp change

of waveform. In general, vowel duration was measured from the onset of the second

formant to the offset of the second formant.

However, the target vowel was sometimes devoiced because some environment

of the target vowel coincided with the context where vowel devoicing occurs in Korean

(cf. Jun and Beckman 1993)8. The target vowel was devoiced in 235 tokens out of 900

tokens (26%), so those 235 tokens (32 of 288 native tokens, 11%; 41 of 288 nonce match

tokens to native words, 14%; 90 of 162 loan tokens, 55%; 72 of 162 nonce match tokens

to loanwords, 44%) were excluded from analysis.

Average values of the first two formants (F1-F2) were automatically taken at the

midpoint of the vowel by Praat’s Burg algorithm9. Formant frequency values were then

verified manually. Extreme formant frequency values in 27 out of 665 tokens (4%),

which fall more than two standard deviations away from the mean value, were excluded.

Therefore, in total, 638 tokens (247 native tokens; 243 nonce match tokens to native

words; 67 loan tokens; 81 nonce match tokens to loanwords) were analyzed for the study.

8 Previous studies (e.g. Jun and Beckman 1993) report that vowel devoicing of high vowels may occur between voiceless consonants, and aspirated stops trigger the most vowel devoicing on the following high vowels i,u, in Korean. The environment of vowel epenthesis such as voiceless consonant clusters coincides with the context where vowel devoicing often occurs in Korean. 9 The formant measurements were taken based on the settings of a maximum frequency of 5500 for female speakers and 5000 for male speakers and a window length of 0.025 seconds.

8


________________________________________________________________________ 3.3 Results

3.3.1 Epenthetic vowels in nonce words vs. lexical vowels in native words

Mean duration and mean formant values of the target vowel between the two word

groups, nonce words vs. native words, were compared using a repeated measures analysis

of variance (ANOVA). The independent variables were underlying status (epenthetic vs.

lexical) and subject. The dependent variables were measurements of duration, F1 and F2.

The hypothesis that epenthetic vowels are transitional or less prominent than lexical

vowels predicts that epenthetic vowels will be shorter in duration than lexical vowels. If

transitional vowels lack a target gesture, they should be pronounced more like schwa, and

therefore transitional vowels should be realized in lower or more fronted (centralized)

position since the corresponding lexical vowels are unrounded high back vowels. Thus, if

epenthetic vowels are transitional, they should be higher in F1 (lower in height) and

higher in F2 (more centralized) than lexical vowels.

Contrary to the predictions of Hypothesis I, results showed that, in general, there

was no significant difference in acoustic quality between epenthetic vowels and lexical

vowels for F1, F2, and duration. The F1 and F2 of the epenthetic vowel were slightly

lower than those of the lexical vowel. The duration of the epenthetic vowel was

somewhat longer than that of the lexical vowel. However, none of the differences was

significant (p > 0.05). The ANOVA results are given in Table 1.

9

Hyun-ju Kim ________________________________________________________________________

Table 1: Combined ANOVA results for all subjects10

epenthetic Vs in

nonce words lexical Vs in native words

mean s.d mean s.d F(1,4) p F1(Hz) 398.50 45.61 409.06 49.33 1.70 0.262 F2(Hz) 1492.49 127.42 1532.24 121.75 0.92 0.391 V duration(ms) 41.21 5.67 37.67 4.92 4.46 0.102

ep:N=243;lex: N=247, Computed using alpha = .05

Overall, the results turned out to be opposite to the predictions of the transitional vowel

hypothesis that the epenthetic vowel would be shorter and more centralized. This might

be due to hyper-articulation since speakers would be more careful in pronouncing nonce

words than in reading familiar native words. Epenthetic vowels in nonce words would be

higher and backer than lexical vowels since they showed lower mean F1 and F2 values.

This might be caused by longer duration, providing enough time to reach the target

gesture of high back vowel, .

However, it was not clear that the lower F1 and F2 values of epenthetic vowels

were really related to longer vowel duration when the patterns of each individual were

examined. Individual results showed that speakers with longer epenthetic vowels (W1,

W3, M1, M2) were not always those with lower F1 and F2 values as shown in Figures 1,

2, 3.

10 The ANOVA results analyzed by speaker gender showed that the patterns in each gender group were consistent with overall results, slightly lower F1 and F2 and a little longer duration in epenthetic vowels than in lexical vowels (but not significantly, p>.05). No significant interaction effects of gender and defendant variables (F1, F2, duration) were found. Therefore, the data split by gender group is not reported here.

10


________________________________________________________________________

0

10

20

30

40

50

60

W1 W2 W3 M1 M2 M3

dura

tion

in m V in Native

V in Nonce(N)

Native: aphda, kopho

aphsin, ..

Nonce(N):

cafda[kaphda]

bofgo[pophgo]

cafsin[kaphsin]

…

Figure 1: Individual results for duration: W: Women, M: Men, Nonce(N): Nonce matches for

native words, V in Native: lexical vowels, V in Nonce (N): epenthetic vowels

As for F1 values, two subjects (W1, W3) showed higher F1 values in epenthetic vowels,

but four other subjects (W2, M1, M2, M3) showed very little difference in F1 between

the two vowels as in Figure 2.

0

100

200

300

400

500

600

W1 W2 W3 M1 M2 M3

F1(H

z V in NativeV in Nonce(N)


aphsin, ..

Nonce(N):

cafda[kaphda]

bofgo[pophgo]

cafsin[kaphsin]

…

Figure 2: Individual results for F1: W: Women, M: Men, Nonce(N): Nonce matches for native

words, V in Native: lexical vowels, V in Nonce (N): epenthetic vowels

11

Hyun-ju Kim ________________________________________________________________________

In terms of F2 values, two subjects (W2, M1) showed a lower F2 in epenthetic vowels

than in lexical vowels, but W3 showed a higher F2 in epenthetic vowels. W1, M2, and

M3 did not have much difference in F2 between the two vowels. The results are

illustrated in Figure 3.

0

500

1000

1500

2000

2500

W1 W2 W3 M1 M2 M3

F2(H

z V in Native

V in Nonce(N)


aphsin, ..

Nonce(N):

cafda[kaphda]

bofgo[pophgo]

cafsin[kaphsin]

…

Figure 3: Individual results for F2: W: Women, M: Men, Nonce(N): Nonce matches for native

words, V in Native: lexical vowels, V in Nonce (N): epenthetic vowels

As described so far, subjects who produced epenthetic vowels as longer than lexical ones

did not show a consistent direction in formant values: among subjects (W1,W3, M1, M2)

with longer epenthetic vowels, W1 had higher F1 and lower F2 in epenthetic vowels, W3

had higher F1 but higher F2, M1 had lower F2, and M2 did not exhibit a difference in

both formant values.

Pearson correlation test was performed for each subject in order to see whether

formant values were correlated with duration. The results showed that there was

significant correlation between formant values and duration for some speakers as shown

in Table 2. However, it is difficult to conclude that duration and formant values were

correlated in general because of the presence of individual differences.

12


________________________________________________________________________ Table 2: Pearson Correlation

duration vs. F1 duration vs. F2 Subjects correlation significance(2-tailed) correlation significance(2-tailed)

M1 (N=45) 0.156 0.305 -0.168 0.269 M2(N=78) 0.341** 0.002 -0.318** 0.005 M3(N=92) -0.095 0.367 -0.211* 0.043 W1(N=92) 0.238* 0.022 -0.167 0.111 W2(N=93) 0.138 0.188 -0.206* 0.047 W3(N=94) 0.169 0.103 0.079 0.447

*. Correlation is significant at the 0.05 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). M2, M3, W2 showed significant negative correlation between duration and F2, which

suggests that longer vowels tended to be produced backer since F2 became lower as

duration became longer. This negative correlation between F2 and duration seems to be

in accordance with hyper-articulation account. However, W1 showed no significant

correlation between F2 and duration but significant correlation between F1 and duration.

The others, M1 and W3, showed no significant correlation between both formants and

duration. Furthermore, all subjects (M1, M2, W1, W2, W3) except M3 showed positive

correlation between F1 and duration and M2 and W1 reached significance: longer vowels

were higher in F1(more fronted). This is not consistent with the prediction that hyper-

articulation would rather lower F1 than make it higher. Therefore, the hyper-articulation

hypothesis cannot clearly account for epenthetic vowels in nonce words, which were

produced in a longer duration and lower F1 and F2, and needs further investigation.

To sum up, lower mean values of F1 and F2 in epenthetic vowels did not

necessarily suggest that epenthetic vowels were hyper-articulated. In addition, the

differences in duration, F1 and F2 for the two vowels do not reach significance.

Therefore, I conclude that the quality of the epenthetic vowel and the lexical vowel is not

really different.

13

Hyun-ju Kim ________________________________________________________________________

3.3.2 Epenthetic vowels in Nonce words vs. Epenthetic vowels in Loanwords

Although epenthetic vowels in English nonce words were not phonetically different from

lexical vowels in native words, it is conceivable that epenthetic vowels in nonce words

might have been produced differently from epenthetic vowels in lexicalized loanwords,

e.g. longer in duration due to unfamiliarity. If epenthetic vowels in loanwords are similar

to those in nonce words, we remove the possibility that unfamiliarity caused epenthetic

vowels in nonce words to be produced differently from normal epenthetic vowels in

loanwords. On the other hand, the nonce word effect will be confirmed if inserted vowels

in nonce words are significantly longer than those in loanwords. In order to control

adjacent consonant/vowel type effect, the neighboring consonants and vowels were the

same in matched pairs as in (7).

(7) Nonce words Loanwords

vift [pi phth ] kiphth ‘ift’

boft [po phth] so phth ‘soft’

fost [pho sth ] ko sth ‘host’

bogma [pogma] togma ‘doma’

The combined results showed that speakers did not have a significant difference

in the phonetic quality of epenthetic vowels between loanwords and nonce words (p

>0.05), as illustrated in Table 3.

Table 3. Combined ANOVA results of production of epenthetic vowels

Vs in nonce words Vs in loanwords mean s.d mean s.d F(1,4) p F1(Hz) 380.77 52.62 369.69 52.23 0.17 0.70 F2(Hz) 1349.38 221.58 1308.1 168.17 0.43 0.55 V duration(ms) 34.18 7.58 33.77 10.71 0.06 0.82

nonce: N=81; loan: N=67, Computed using alpha = .05

14


________________________________________________________________________ The overall mean duration of epenthetic vowels in loanwords was almost same as that of

epenthetic vowels of nonce words. Average formant values of the vowels, however, were

slightly higher in nonce words both in F1 and F2, which suggests that they may be

produced slightly lower and more fronted than in loanwords. Somewhat centralized

epenthetic vowels in nonce words were not what would be predicted from hyper-

articulation with a longer duration. The higher F1 and F2 values rather suggest that the

vowels did not reach (or undershot) the target gesture of the conventional vowel //,

which is unrounded, high and back. Nevertheless, the difference in the formants was not

significant (F1, p=0.70; F2, p=0.55).

On the other hand, when the averages were separated by gender, the female group

showed a significant difference in F2 (p<.05) as illustrated in Table 4.

Table 4. ANOVA results by gender

V in nonce words V in loanwords mean s.d mean s.d F(1,2) p Men F1(Hz) 392.47 69.94 366.56 21.99 1.50 0.35 F2(Hz) 1177.99 102.14 1213.26 186.47 0.74 0.48 V duration(ms) 35.40 11.54 38.96 14.10 1.87 0.30

Women F1(Hz) 369.06 40.27 372.82 79.41 1.80 0.31 F2(Hz) 1520.78 155.52 1402.93 94.6 63.38 *0.02 V duration(ms) 32.97 2.46 28.57 2.60 2.63 0.25

Men: nonce:N=34, loan:N=30; Women: nonce: N=47, loan: N=37 A star indicates that the differences are significant at α=.05

Female group produced epenthetic vowels of nonce words with higher F2 than in

loanwords, which may suggest that epenthetic vowels tended to be more fronted in online

adaptation. Since their duration was longer than or similar to the corresponding vowels

in loanwords, higher F2 does not necessarily mean that they had a lexical target but did

not meet it because of gestural undershoot, which generally occurs due to decreased

duration.

15

Hyun-ju Kim ________________________________________________________________________

The individual results showed that subjects W2, W3, M2, and M3 had consistent

patterns with lower or similar mean F1 values and higher F2 values in online adaptation.

Subjects W1, M1 showed distinct patterns such as higher F1 or lower F2 for epenthetic

vowels of nonce words. The individual results for the formant values are given in Figures

4 and 5.

0

100

200

300

400

500

600

W1 W2 W3 M1 M2 M3

F1(H

z V in Loan

V in Nonce(L)

Loan:

kiphth

kosth

togma, ..

Nonce(L): vift [piphth]

fost [pho sth]

bogma[pogma]…

Figure 4. Individual results for F1: W: Women, M: Men, Nonce(L): Nonce matches for loanwords

0

200

400

600

800

1000

1200

1400

1600

1800

2000

W1 W2 W3 M1 M2 M3

F2(H

z V in Loan

V in Nonce(L)

Loan:

kiphth

kosth

togma, ..


fost [pho sth]

bogma[pogma]…

Figure 5. Individual results for F2: W: Women, M: Men, Nonce(L): Nonce matches for loanwords

16


________________________________________________________________________ Female subjects tended to produce epenthetic vowels somewhat longer or with the same

duration, but male subjects, with the exception of M1, tended to make epenthetic vowels

shorter for nonce words. The individual results for duration are given in Figure 6.

0

10

20

30

40

50

60

70

80

W1 W2 W3 M1 M2 M3

dura

tion(

m

V in Loan

V in Nonce(L)

Loan:

kiphth

kosth

togma, ..


fost [pho sth]

bogma[pogma]…

Figure 6. Individual results for duration: W: Women, M: Men, Nonce(L): Nonce matches for loanwords

However, none of the differences between the vowels in loanwords and the vowels in

nonce words were significant11.

It remains unexplained what caused the F2 values in the target vowels to be

significantly higher in nonce words than in loanwords for female subjects. Stimuli for

nonce words were given in forms without inserted vowels, written in English (e.g. vift),

while comparable loanwords were written with final vowels in Korean (e.g. kiphth ‘gift’).

Since word final epenthetic vowels were not examined in the current study, it is not clear

whether the word final epenthetic vowels in nonce words were the same quality as in

loanwords. If the word final epenthetic vowels in nonce words were produced differently

from those in loanwords, the higher F2 values of word-medial epenthetic vowels could

have been attributed to co-articulation with the following vowel. See also Appendix IV. I

will leave this issue for future investigation.

11 Extended analyses of the results, regarding adjacent consonant effect on vowel duration and adjacent vowel effect on vowel formants, are presented in appendices III, IV.

17

Hyun-ju Kim ________________________________________________________________________

In sum, the overall results demonstrated that epenthetic vowels in both lexicalized

loanwords and on-line adaptations were not different phonetically from lexical vowels,

which suggests that the unaccentablity of epenthetic vowels cannot be ascribed to

acoustic properties.

4. Discussion

4.1 Articulation of Epenthetic vowels

The evidence that Kyungsang Korean speakers produced epenthetic vowels which were

acoustically similar to lexical vowel // suggests that KK speakers do not distinguish

epenthetic vowels phonetically from corresponding lexical vowels. Also, this result

suggests that epenthetic vowels in KK are true vowels rather than transitional.

The results of completely neutralized epenthetic vowels in KK are different from

what Gouskova and Hall (to appear) found in their study of Lebanese Arabic. Lebanese

Arabic speakers optionally produced epenthetic vowels which were distinctively either

shorter or backer than lexical vowels. Davidson (2006) also demonstrates that English

speakers produce inserted schwas significantly differently from lexical schwas. A

question naturally arises: why were KK speakers different from English speakers or

Lebanese speakers in the production of epenthetic vowels? Considering cross-linguistic

phonotactics, certain types of complex codas are permissible both in Lebanese Arabic

and in English, subject to language-specific restrictions. However, complex codas are

never possible in Korean phonology. Because tauto-syllabic consonant clusters must be

repaired via consonant deletion or vowel epenthesis in Korean, Korean speakers do not

have the option to produce consonant clusters without inserting a vowel or even with a

transitional vowel. Therefore, it is plausible that language-specific phonetics driven by

Korean phonology forces Korean speakers to produce a full realization of epenthetic

vowels.

18


________________________________________________________________________ 4.2 Failure to Accent Epenthetic vowels

Gouskova and Hall (to appear) suggest that incomplete neutralization helps learners to

access underlying representations of epenthetic vowels based on the acoustic differences

between epenthetic vowels and lexical vowels. However, epenthetic vowels in Korean

loanwords are completely neutralized and phonetic information about epenthetic vowels

is not available to speakers in the acoustic signal of the surface form. Therefore, it

remains to be explained why epenthetic vowels tend to be unaccented in positions where

a lexical vowel would attract accent. Furthermore, if epenthetic vowels sound like lexical

vowels, how do learners know not to put accent on them?

In KK, accent is assigned on the penult if a word consists of only light syllables

following the default accent assignment constraints: Align-R, Ft-Binarity, Trochee (Kim

1997; Kenstowicz and Sohn 2001; Broselow, to appear). Otherwise, when accent is

lexically specified in native lexicon, it is realized faithfully (e.g. kamani ‘rice sack’, satari

‘ladder’), so FaithAccent is ranked at the top. But the distinction between the two accent

groups, a free accent group and a default accent group, is not clear because the default

accent group is a proper subset of the free accent group, which obscures the default

accent pattern. On the other hand, loanwords are assumed to have no accent in URs, and

they are accented on the penult when a word consists of open syllables, following the

default accent assignment. However, when loanwords contain epenthetic vowels in the

penultimate position, the words carry non-penultimate accent.

In terms of learning a phonological grammar, the markedness-over-faithfulness

(M>>F) bias has been suggested as the initial state by many researchers (e.g.

Gnanadesikan 2004; Hayes 2004). The initial state is postulated to have the ranking of

Accent Constraints >> FaithAccent (markedness constraints >> faithfulness constraints)

in learning of the accent-epenthesis interaction. As learners are exposed to exceptional

cases, they revise their grammar by promoting a relevant faithfulness constraint,

FaithAccent, above the markedness constraints, Accent Constraints. Despite predictable

penultimate accent with loanwords, learners may over-generalize free accent to

loanwords because there are exceptional cases to have non-penultimate accent in

19

Hyun-ju Kim ________________________________________________________________________

loanwords due to accent-epenthesis interaction. Overgeneralized learning of the target

grammar is schematized as in (8).

(8) Overgeneralized learning of the accent-epenthesis interaction

Native words Loanwords

Initial G

(M>>F

bias)

Accent Constraints

>> FaithAccent (penultimate accent,

default accent group):

e.g.)

ssro ‘by oneself’,

apci ‘father’

aphta ‘hurt’

Accent Constraints >>

FaithAccent (default accent)

e.g.)

ikhako ‘Chicago’

trama ‘drama’

Target G

(promotion

of F)

FaithAccent >>

Accent Constraints (lexical accent group): e.g.)

kamani ‘rice sack’

satari ‘ladder’

FaithAccent>>Accent

Constraints (accent-epenthesis

interaction, overgeneralized free

accent): e.g.)

risth (*risth) ‘list’,

stha (*stha)‘star’,

sthro (*sthro) ‘straw’

It is not surprising that loanwords with non-penultimate accent would have been

lexicalized with non-penultimate accent since KK has a native lexical group whose

accent is unpredictable (e.g. kamani ‘rice sack’, satari ‘ladder’). Under the assumption

that the first adapter could hear the contrast between lexical vowels and no vowels and

did not put accent on an epenthetic vowel in the penult, it is plausible that subsequent

listeners would store words with the non-penultimate accent faithfully.

Nevertheless, KK speakers still seem to have a strong tendency to carry default

penultimate accent to loanwords containing only light syllables unless the penult is

epenthetic. Lee (2003) reports that 91.8% (370 out of 403 tokens) of trisyllabic

20


________________________________________________________________________

loanwords with only light syllables carry penultimate accent consistently and 26 items

out of the remaining 33 trisyllabic words, which have antepenultimate accent, contain an

epenthetic vowel on the penult. This demonstrates that existence of counterexamples with

non-penultimate accent does not undermine the default accent pattern. Then, how do KK

learners conclude that the words with non-penultimate accent contain epenthetic vowels

rather than that the accent is lexically specified?

Distinct combinations of segment sequences in loanwords may give learners a

basis for analogical generalization for epenthetic vowels. First, aspirated obstruents such

as th, ph, ch are used relatively more frequently in loanwords than in Korean native words.

This is because there is no voicing contrast in Korean and the voicing contrast in a source

language is preserved as an aspiration contrast in loanwords: voiceless obstruents are

replaced by aspirated correlates and voiced obstruents are replaced by lax or sometimes

tense ones. Second, due to the maximal syllable template CVC (No complex clusters),

vowel epenthesis prevails in loanword adaptation. Initial and final clusters of voiceless

consonants are common in English words (straw, skew, best, gift). And the sequences of

consonants are repaired by inserting a default vowel //. As a result, the sequences of

aspirated consonants(Ch) or sCh clusters with an epenthetic vowel (Ch Ch or sCh) are

very common in loanwords (e.g. ri phth ’lift’, te skh ‘desk’, sthro ‘straw’) although

they are marked or rare combinations in native words12. Third, because Korean coda

consonants are never released, all consonants with release or aspiration are neutralized in

coda position in native phonology. However, in loanwords, release of coda consonants of

source words is preserved by inserting a vowel (e.g. ps ps ‘bus’, nok nokh

‘knock’). Kang (2003) suggests that vowel epenthesis to preserve features of coda

consonants such as release is motivated by perceptual similarity between the English

input and the Korean output. Therefore, sequences of an aspirated or released consonant

and an epenthetic vowel become very common combinations in Korean loanwords from

English, since the main phonological environments of vowel epenthesis are after released

coda consonants or between the initial/final voiceless consonant clusters. Based on these

12 A reference for the phonotactic probabilities in loanwords versus in native words is not available since this fact is based on the author’s insight and a short corpus study. Further statistical study is required in order to confirm this generalization.

21

Hyun-ju Kim ________________________________________________________________________

differences between loanwords and native words, KK speakers have a basis for assuming

that a vowel // which appears after released/aspirated consonants is epenthetic rather than

lexical.

The segment distribution information may also help KK speakers to construct

separate lexical strata in the Korean lexicon. The vowel // is always epenthetic and never

appears as non-epenthetic in loanwords13. And it is usually not accented in loanwords

from English. Rhee and Kim (2003) report that accented epenthetic vowels are only 9.5%

(244/2562) regardless of position based on accent patterns made by 13 NKK native

speakers’ judgments over 200 loanwords (2600 tokens (13x200)), which contrasts with

90% of accented lexical vowels in the final closed syllable. On the other hand, the lexical

vowel // contained in native words, which has the same phonetic quality as epenthetic

vowels, does not have restrictions in accent assignment: namely, it does not resist accent

in KK (e.g. snim ‘monk’, ssro ‘by oneself’). Assuming that the Korean lexicon is

divided into at least two lexical strata, NATIVE and FOREIGN, based on the lexicon

stratification hypothesis originally proposed by Ito and Mester (1999) for the Japanese

lexicon, Korean speakers take the vowel // to be epenthetic in the FOREIGN stratum.

The two distinct lexical strata are supported by discrepancy in repair strategies

between native phonology and loanword phonology in Korean. Illicit consonant clusters

are repaired in the native phonology by deleting a consonant (e.g., /talk/ -> tak 'chicken),

whereas in loanword phonology a vowel is inserted (e.g. milkh ‘milk’).

To recapitulate, the finding that epenthetic vowels are identical phonetically to

lexical vowels raises the question of how KK speakers learn the covert accent-epenthesis

interaction, because the status of epenthetic vowels is not accessible from the surface

form alone. The problem may be solved by assuming that KK speakers can access the

underlying status of epenthetic vowels, relying on other cues such as the probabilistic or

analogical information of distinct phonotactic distribution of the foreign lexicon.

13 High front vowel /i/ is sometimes used as epenthetic vowels in loanwords (e.g. penti ‘bench’ vs. jts ‘shirts’) motivated by preservation of perceptual cues (Kang 2001), and /i/ may be used as a non-epenthetic vowel (e.g. iti ‘city’).

22


________________________________________________________________________ 5. Conclusion

The present phonetic study of epenthetic vowels in KK was performed in order to

test the hypothesis that information concerning the status of epenthetic vowels is

available in the acoustic signal: phonetic properties of epenthetic vowels were

hypothesized to be distinct from those of the lexical vowels. Epenthetic vowels may have

shorter duration or distinct formant frequencies, or both, which would suggest that

epenthetic vowels are not real vowels but transitional vowels, so they are invisible in

phonological processes. Therefore, they are unaccented. However, the results showed that

epenthetic vowels were not different phonetically from lexical vowels in the surface form,

which suggests that the unaccentablity of epenthetic vowels cannot be ascribed to

acoustic properties. Rather, learners must get information concerning the underlying

status of epenthetic vowels somewhere beyond the phonetic realization of these vowels.

Given the fact that epenthetic vowels are completely neutralized with lexical

vowels, learning the covert accent-epenthesis interaction involves complicated

learnability issues. However, KK speakers still show evidence of a default accent pattern

in loanwords even when there are words that violate the pattern. While these words

contain historically epenthetic vowels, I have shown that these vowels are now

phonetically indistinguishable from lexical vowels. However, if KK speakers identify

foreign words based on probabilistic or analogical information of overall word structure

such as distinct phonotactic distribution, they can assume that foreign words with non-

penultimate accent contain epenthetic vowels.

To conclude, accent-epenthesis interaction must be a phonological process

independent of phonetics. A default accent pattern with accent-epenthesis interaction,

which is manifested in KK loanwords, demonstrates that there is a default accent pattern,

which can be violated under certain circumstances.

23

Hyun-ju Kim ________________________________________________________________________

Appendix I. Stimuli

I. Native words ~Nonce words Nonce words [expected form] Native words

cafda [khaphda] aphda 'to be sick'

bofda [pophda] kophda 'to be hungry'

/phVd/

sufda [suphda] slphda 'to be sad'

cafgo [khaphgo ] aphgo ‘sick and’

bofgo [popho] kopho 'hungry and '

/phVg/

sufgo [supho] slpho 'sad and'

cafsin [khaphsin] aphsin ‘sick-infix(honorific)’

bofsin [pophsin] kophsin 'hungry-infix'

/phVs/

sufsin [suphsin] slphsin 'sad-infix' /gVd/ tagda [thagda] tamgda 'to soak' /gVg/ tagko [thaggo ] tamggo 'soak and' /gVs/ tagsin [thagsi n] tamgsi n 'soak-infix'

bosme [posme] psmjn ’take off-if’ /sVm/ lusme [lusme] wu smjn ’lauh-if’ bosni [posni] psni ’take off-suffix’ /sVn/ lusni [lusni] wu sni ’lauh-suffix’

II. Loanwords ~ Nonce words

Nonce words [expected form] Loanwords

teft [thephth] rephth ’left’ boft [pophth] so phth ’soft’

/phVth/

vift [piphth] kiphth ’ift’ fost [pho sth] ko sth ’host’ /sVth/

kest [kesth] testh ’test’ dethku [teskhu] teskh ’desk’ /sVkh/

bithkee [piskhi] wiskhi ’whiskey’ bogma [togma] togma ’doma’ /gVm/,

/gVkh/ tugkup [tugkhp] mgkhp ’mu cup’

24


________________________________________________________________________ Appendix II. No accent effect on vowel duration

There were some differences in accent between the native words set and the nonce words

set (in Appendix I). The target lexical vowel in native words was sometimes accented but

sometimes unaccented. However, the target epenthetic vowel in nonce word matches for

native words was consistently unaccented, namely post-accented. One might think that

this difference would cause durational differences between lexical vowels and epenthetic

vowels. However, I found that there was no accent effect on vowel duration in my other

study, which compared duration of accented lexical high vowels with that of unaccented

lexical high vowels in KK native words, produced by the same subjects who participated

in this study. The following words in (9) were stimuli and four repetitions produced by 6

speakers were analyzed.

(9) Unaccented high vowel

nophita ‘heighten’

i phita ‘dress’

nuphita ‘lay down’

cophita ‘narrow’

Accented high vowel

phita ‘be carried on’

caphi ta ‘be captured’

cphi ta ‘be folded’

palphi ta ‘be stepped on’

As presented in the following table, duration of accented vowels is slightly longer than

that of unaccented vowels, but the difference is just a few milliseconds and not

significant. Therefore, I assume that positional difference of the lexical vowel, accented

vs. unaccented, would not be a problem.

25

Hyun-ju Kim ________________________________________________________________________

(10) Mean duration of the target vowel

unaccented V

(16x6 tokens)

accented V

(16x6tokens)

Mean duration (ms.) 44.68 48.47

s.d 11.98 10.66

Appendix III. Adjacent consonant effect on vowel duration

Vowel durations seem to vary depending on the context: namely, epenthetic vowels in

nonce words versus native words were longer than in nonce words versus loanwords. In

other words, epenthetic vowels in loanwords were shorter than the corresponding vowels

in native words as shown in Figure 7.

Native: aphda,

kopho,

aphsin

Loan:

kiphth,

kosth,

teskh

0

10

20

30

40

50

60

W1 W2 W3 M1 M3 M2

V in Native

V in Loan

Figure 7. Mean vowel duration by word type: W: Women, M: Men, V in Native: lexical

vowels, V in Loan: epenthetic vowels

This may be because the neighboring consonants in test loanwords differ from those in

test native words. Epenthetic vowels in test loanwords are placed before aspirated

consonants or between aspirated consonants (e.g. thos th ‘toast’, ki phth ‘gift’) while

lexical vowels in test native words are before non-aspirated consonants (e.g. kopho

26


________________________________________________________________________ ‘hungry and’). Therefore, it appears that the target vowels were realized consistently

shorter before aspirated stops.

Figure 8 illustrates the effect of adjacent consonants on vowel duration more

clearly. Both lexical vowels and epenthetic vowels displayed similar patterns for the

surrounding consonants, which suggests that there was no interaction between vowel

duration and vowel type. Therefore, epenthetic vowels are not different from lexical

vowels in the effect of adjacent consonants on vowel duration.

20

25

30

35

40

45

50

ph/gVd ph/gVg ph/gVs sVm sVn

dura

tion(

m

V in Native

V in Nonce(N)

Native: aphda, kopho, aphsin, psmjn, psni, .. Nonce(N): kaphda, pophgo kaphsin, posme, posni, ..

Figure 8. Variation of mean vowel duration by surrounding consonants: Nonce(N):Nonce matches for native words, V in Native: lexical vowels, V in Nonce(N): epenthetic vowels

Appendix IV. Adjacent vowel effect on formant frequencies

Formant frequencies also varied depending on the surrounding context of the target

vowel: namely, they change according to the qualities of vowels in the following syllable.

As illustrated in Figure 10, F1 became lower when the following vowel was a high vowel.

This tendency suggests that the properties of high back vowel // can be flexible, easily

influenced by the neighboring vowels.

27

Hyun-ju Kim ________________________________________________________________________

200

250

300

350

400

450

Vda Vgo Vsi Vmi Vni

F1(H

z V in Native

V in Nonce(N)


Figure 10. Mean F1 values by context: Nonce(N):Nonce matches for native words, V in

Native: lexical vowels, V in Nonce(N): epenthetic vowels

Mean F2 frequencies showed clearer changes depending on the frontness of the

following vowels: a following vowel /o/ pulled down F2 values of epenthetic vowels

significantly while a following front vowel /i/ raised the F2 values. Combined with F1

changes, these facts demonstrate that the unrounded high back vowel undergoes vowel

assimilation with the following vowel.

0

200

400

600

800

1000

1200

1400

1600

1800

2000

Vda Vgo Vsi Vmi Vni

F2(H

z V in Native

V in Nonce(N)


Figure 11. Mean F2 values by context: Nonce(N):Nonce matches for native words, V in

Native: lexical vowels, V in Nonce(N): epenthetic vowels

As shown in Figures 10 and 11, the formant values of a vowel // are highly influenced by

the qualities of the following vowel. Tendencies such as a lower F1 and a higher F2

28


________________________________________________________________________ before a front high vowel /i/ demonstrate that the gesture of the vowel // has already been

adjusted toward the following vowel. However, despite the main neighboring vowel

effect on vowel formants, the figures illustrate that there was no interaction between

formant values and vowel types. Therefore, these results confirm that lexical vowels and

epenthetic vowels were not phonetically different.

References

Alderete, J. (1999). Head dependence in stress-epenthesis interaction. In Ben

Hermans and Marc van Oostendorp (eds.), The Derivational Residue in

Optimality Theory, 29-50. Amsterdam: John Benjamins.

Alderete, J. and Tesar, B. (2002). Learning covert phonological interaction: an analysis of

the problem posed by the interaction of stress and epenthesis. Report no. RuCCS-

TR-72, Piscataway, NJ: Rutgers Center for Cognitive Science. Available as ROA-

543.

Boersma, P. & Weenink, D. (2003). Praat: Doing phonetics by computer (version 4.3.17).

[Computer program]. Retrieved in 2003 from <http://www.praat.org/>

Broselow, E. (to appear). Stress-epenthesis interactions in constraint-based phonology. In

Bert Vaux & Andrew Nevins (eds.), Rules and Constraints, and phonological

phenomena. Oxford: Oxford University Press.

Davidson, L. (2006). Phonology, phonetics, or frequency: Influences on the production of

non-native sequences. Journal of Phonetics 34:104-137.

Gnanadesikan, A. (2004). Markedness and faithfulness in child phonology. In Rene

Kager, Joe Pater and Wim Zonneveld (eds.), Constraints in Phonological

Acquisition, 73-108. Cambridge: Cambridge University Press.

Gouskova, M. & Hall, N. (to appear). Acoustics of epenthetic vowels in Lebanese

Arabic. In Parker, Steve (eds.), Phonological Argumentation. Essays on evidence

and motivation. London: Equinox.

29

Hyun-ju Kim ________________________________________________________________________

Gouskova, M. (2006). The phonology and phonetics of unstressable vowels in

Lebanese Arabic. Handout presented at colloquium series of the Department of

Linguistics, Stony Brook University, Spring 2006.

Hall, N. (2004). Implications of vowel intrusion for a gestural grammar. Ms.

http://research.haifa.ac.il/~nancyh/

Hall, N. (2006). Cross-linguistic patterns of vowel intrusion. Phonology 23: 387-429.

Hayes, B. (2004). Phonological acquisition in Optimality Theory. In Rene Kager, Joe

Pater and Wim Zonneveld (eds.), Constraints in Phonological Acquisition, 158-

203. Cambridge: Cambridge University Press.

Ito, J. and Mester, A. 1999. The phonological lexicon. In N. Tsujimura (eds.), The

Handbook of Japanese Linguistics, 62-100. Cambridge, MA: Blackwell.

Jun, S. & Beckman, M. 1993. A gestural-overlap analysis of vowel devoicing in

Japanese and in Korean. Paper presented at the 67th Annual Meeting of the

Linguistic Society of America. Los Angeles, CA.

Kang, Y. (2003). Perceptual similarity in loanword adaptation: English postvocalic

word-final stops in Korean. Phonology 20: 219-273.

Kenstowicz, M. & Sohn, H. (2001). Accentual adaptation in North Kyungsang Korean.

In M. Kenstowicz (eds.), Ken Hale: a life in language, 239-270. Cambridge, MA:

MIT Press.

Kim, N. (1997). Tone, Segments, and their Interaction in North-Kyungsang Korean: A

Correspondence Theoretic Account, Doctoral dissertation, Ohio State University.

Lee, D. (2003). The tone patterns of loan words in South Kyungsang Korean. Eoneo

28-3: 451-486.

Lee, H. (1996). Kukeo Eumseonghak (Korean Phonetics). Seoul: Taehaksa.

Prince, A. & Tesar, B. (2004). Learning phonotactic distributions. In Rene Kager, Joe

Pater and Wim Zonneveld (eds.), Constraints in Phonological Acquisition,245-

291. Cambridge: Cambridge University Press.

Rhee, S. & Kim, J. (2003). Daegu pangeon chayongeoeyseouy sapipmoumkwa koceo

aekseynthu (Epenthetic vowels and pitch accent in the Loanword Phonology of

North Kyungsang Korean). Eoneohag 37: 223-252.

30

the phonology and phonetics of epenthetic vowels … phonology and phonetics of epenthetic vowels in...

Documents