VOT trumps other measures in predicting Korean children’s early mastery of tense stops Eun Jong Kong Mary E. Beckman Jan Edwards LSA2010 January 7 th.

VOT trumps other measures in predicting Korean children’s

early mastery of tense stops

Eun Jong Kong Mary E. Beckman

Jan Edwards

LSA2010 January 7th ~ 10th

Introduction

Three-way laryngeal contrast in Korean stops• lax /p, t, k/ vs. tense /p’, t’, k’/ vs. aspirated /ph, th, kh/

e.g., /tal/ moon vs. /t’al/ daughter vs. /thal/ mask

Mastery pattern of Korean stops:• Transcription-based studies (Kim, Y, 1996; Kim &

Pae, 2005; Kim, M, 2008) described: • All (lax, tense and asp.) were mastered before three

years. • Tense stops appeared first in youngest children’s

productions. (before 2;6)

• Early mastery of tense stops:• Is this predictable based on the acoustic property of

Korean stops? • Cross-lingusitically exceptional or not?

Burst Voicing OnsetTime (s)

121.2 121.30

5000

Fre

quen

cy (

Hz)

Short lag VOT

Time (s)15.13 15.380

5000

Fre

quen

cy (

Hz)

Lead VOT

BurstVoicing Onset Burst Voicing Onset

[do'mata] tomato toastdonut

VOT = Burst – Voicing Onset

• Voice Onset Time (VOT: Lisker and Abramson 1964) and stop voicing/ aspiration contrast across languages

Early mastery of tense stops: Is this predictable?

Long lag VOT

Time (s)0 0.2006

0

9000

Fre

quen

cy (

Hz)

Language Lead Short lag Long lag

English (Macken & Barton 1980a) voiced voiced voiceless

Spanish (Macken & Barton 1980b) voiced voiceless

Cantonese (Clumeck et al 1981 ) unaspirated aspirated

Thai (Gandour et al 1986) voiced unaspirated aspirated

Long lag

Lead[t] [th] [d]

Short lag

Universal order of mastery and VOT

The category with short lag VOT is mastered first: • the least precise articulation (Kewley-Port and Preston 1974).

is achieved by the glottis opening at any time during the oral occlusion: easy to produce!

Short lag VOT of Korean tense stops

Lead

Short lag

Long lag

tense lax, aspirated

short lag

Long lag

is achieved by the laryngeal muscle tenseness that delays voicing onset until the oral release.

Korean tense stops: • short lag VOT• creaky voice quality (H1-H2)• high fundamental frequency (F0):

associated with H-tone accentual phrase in Seoul Korean (Cho, Jun and Ladefoged, 2002; Kagaya, 1974; Hardcastle 1974; Jun 1993).

When we take a closer look at Korean tense stops …

• VOT values of Korean tense stops predicts the early mastery.

At first glance…

Multiple acoustic cues of Korean stops

Multiplicity of acoustic cues suggests:

• VOT cannot be the only predictor of Korean stop mastery order.

• The relative role of other cues must also be considered.

• We need a more sensitive quantitative model of the role of VOT in relation to F0 and H1-H2 in explaining the observed order in the transcription-based study:

Goal of the study

• This study explores the relationship between native speaker transcriptions of children’s stop productions and the three acoustic characteristics (VOT, F0, H1-H2).

• Mixed effects logistic regression models– Analysis (1): children’s productions

transcribed accuracy as a function of age– Analysis (2): adults’ stop productions

target category as a function of VOT, F0 and H1-H2– Analysis (3): children’s stop productions

transcribed category as a function of VOT, F0 and H1-H2

Database

• Materials- word-initial coronal and dorsal stops embedded in real words - following vowel contexts varied among /i, a, u/.

• Participants- 70 Korean-speaking children (aged 24 to 72 months)- 20 adults (10 males and 10 females: aged 18-30 years)- recruited and recorded at Seoul, Korea.

• TaskA picture-prompted auditory word-repetition task

9/t’al.gi/ strawberry

10/ta.ram.dʒwi/ squirrel

11/th a.dʒo/ ostrich

• Accuracy judgment measures: – native speaker transcriptions of ‘correct’ or

‘incorrect’ in children’s stop productions.

• Acoustic measures: – All plosive productions also were acoustically analyzed to

measure VOT, F0 and H1-H2 (voice quality measure).

Analysis measures

burst voice onset

VOT (sec.)

Time (s)141.9 142.1

-0.06935

0.08031

0

141.871398 142.072843torn4_20000

freq

uen

cy (

Hz)

H1-H2 (dB)

first harmonic (H1)

second harmonic (H2)

frequency (Hz)

am

pli

tud

e (

dB

)

20ms

F0 = 1/interval

Results- Analysis(1): transcribed accuracy ~ age

• a trend of more tense substitutions in the younger children’s productions.

• the highest transcribed accuracy for the tense stops in the youngest.

•VOT: short lag VOT for tense stops

Results – Analysis 2: Acoustic characteristicsadults’ stops:

short lag

short lag

VOT F0 H1-H2

•F0: lower F0 for lax stops•H1-H2: lower H1-H2 for tense stops

• β1, β2, and β3 (coefficients)

: the absolute value of coefficient is the relative contribution of each predictor in determining the dependent variable.

• Two sub-sets of logistic regression model were made– Adult-tense model predicts tense vs. non-tense (i.e., lax and

asp.)– Adult- lax model predicts lax vs. asp.

Results: Analysis (2) adults’ stops

Mixed effects logistic regression model:target category (target or non-target) as a function of

VOT, H1-H2 and F0

€

logtarget consonant

1− target consonant

⎛

⎝ ⎜

⎞

⎠ ⎟= β 0 + β1VOT + β 2F0 + β 3H1H2

Results: Analysis (2) adults’ stops

Steeper slope of VOT than H1-H2

1. Adult-tense model: tense (vs. non-tense) ~ VOT + F0 + H1-H2

Results: Analysis (2) adults’ stops

2. Adult-lax model: : lax (vs. asp) ~ VOT + F0 + H1-H2

Steeper slope of F0 than VOT

Children’s stops

2 year olds

4 year olds

F0 H1-H2VOT

short lag

short lag

•VOT: overlap at a short lag VOT range at a younger age•F0: relatively lower F0 for lax stops•H1-H2: lower H1-H2 for tense stops at an older age.

Results – Analysis 3: Acoustic characteristics

• β1, β2, and β3 (coefficients)

: the absolute value of coefficient is the relative contribution of each predictor in determining the dependent variable.

• Two sub-sets of logistic regression model were made– Child-tense model predicts transcribed tense vs. non-

tense (i.e., lax and asp.)– Child-lax model predicts transcribed lax vs. asp.

Results: Analysis (3) children’ stops

Mixed effects logistic regression model:transcribed category (transc. target or not) as a function of

VOT, H1-H2 and F0

€

logtranscribed consonant

1− transcribed consonant

⎛

⎝ ⎜

⎞

⎠ ⎟= β 0 + β1VOT + β 2F0 + β 3H1H2

Results: Analysis (3) children’s stops

Child-tense model: tense (vs. non-tense) ~ VOT + H1-H2 + F0

steepest slope of VOT

H1-H2

VOT F0

not sig.

Child-lax model: lax (vs. asp.) ~ VOT + H1-H2 + F0

Almost equally steep slope of VOT and F0

H1-H2

VOT F0

Results: Analysis (3) children’s stops

Summary and Conclusion

• Transcription accuracy: early mastery of tense stops

• Acoustic characteristics: Adults’ stops

– VOT: short lag VOT (tense), long lag VOT (lax and asp.)– F0: lower F0 (lax), higher (tense and asp.)– H1-H2: negative H1-H2 (tense), positive H1-H2 (lax and asp.) Children’s stops:

– overlap at the three acoustic dimensions. – VOT: three categories at a short lag VOT range in a younger

age.

• Quantitative model relating transcription categories to acoustic parameters showed that:The transcriber identified the children’s productions as tense primarily based on the VOT values of the productions.The transcribed was less influenced by F0 and H1-H2.

• This dominant role of VOT in the transcription judgment of tense stops can explain the early mastery of tense stops– children produce a short lag VOT value before

mastering a long lag VOT or a lead VOT cross-linguistically.

– having a short lag VOT suffices for a transcriber to identify it as a correct production of a tense stop in Korean.

• Given the role of VOT in the transcription analysis, the early mastery of the Korean tense stop is not such an exceptional pattern after all.

Summary and Conclusion

Acknowledgement

• This study was supported by NIDCD Grant 02932.

• Thanks to

The children who participated in the study and the parents who gave their consent.

Thank you!!!

Ευχαριστώ πολύ

감사합니다24

25/taŋ.gɨn/ carrot

26/tal.phɛŋ.i/ snail

Summary of findings• Transcription accuracy: early mastery of tense stops

• Acoustic characteristics Adults’ stops:

– VOT: short lag VOT (tense), long lag VOT (lax and asp.)– F0: lower F0 (lax), higher (tense and asp.)– H1-H2: negative H1-H2 (tense), positive H1-H2 (lax and asp.) Children’s stops:

– overlap at the three acoustic dimensions. – VOT: three categories at a short lag VOT range in a younger

age.

• Mixed effects logistic regression models– transcribed tense vs. non-tense was primarily predicted by

VOT and less by H1-H2. – transcribed lax vs. aspirated stops were explained by the F0

and VOT.

Korean tense stop: • short lag VOT • relatively high F0, creaky voice quality

Transcriptions and acoustic parameters:The quantitative model of the role of VOT showed

that:• The transcriber identified the children’s productions

of the tense target dominantly by the VOT values of the tokens, less affected by other acoustic aspects such as F0 or H1-H2.

Discussion and Conclusion