Academic Research · 2020

Understanding How We HearAcross Languages

An experimental study of how Chinese learners perceive and identify Korean vowels through the sound categories of their native language.

Master's Thesis

Perceptual Assimilation and Identification Patterns of Korean Vowels by Chinese Learners

Institution

Korea University

Department

Korean Language and Literature

Completed

February 2020

01 / Research Overview

The research question

How do Chinese learners perceive Korean vowels that do not correspond directly to the vowel categories of their native language? Does their perception differ between novice and advanced learners?

The study examined Korean monophthongs and diphthongs through two perception experiments. It explored how learners connected Korean vowels to Mandarin sounds and how accurately they could identify the Korean vowels.

The results were also used to examine the Perceptual Assimilation Model for L2 speech learning (PAM-L2).

Theoretical Framework

PAM-L2 & Phonological Overlap

The study adopted the Perceptual Assimilation Model for L2 (PAM-L2) and the phonological overlap approach to predict how learners would distinguish between Korean vowel contrasts. Predictions based on assimilation patterns were compared with the results of the identification experiment.

02 / Experiment Design

Designing the experiments

Two experiments were designed to investigate how Chinese learners perceive and identify Korean vowel sounds. The study examined both perceptual assimilation and vowel identification, using carefully prepared speech stimuli and a consistent experimental procedure.

Participants

38

Participants who completed both experiments

16

Novice learners

22

Advanced learners

Participants were recruited from northern regions of China where Mandarin varieties are similar to standard Mandarin. Novice learners had studied Korean for no more than six months, while advanced learners held TOPIK Level 5 or above.

Experiment 01

Assimilation

Participants listened to 51 Korean vowel stimuli and identified the Chinese sound that most closely resembled each stimulus. They also rated the perceived similarity using a 7-point scale.

Experiment 02

Identification

Participants listened to the same set of 51 Korean vowel stimuli and identified each sound by writing the corresponding Korean vowel in Hangul.

Experimental Procedure

01

Consent & Instructions

Participants received study instructions and provided informed consent.

02

Practice Tasks

Five practice questions were completed before each experiment.

03

Main Experiments

Participants completed 51 questions in each experiment.

04

Response Collection

Responses were recorded on paper and reviewed by the researcher.

Data collection took place between 15 June and 28 July 2019. Each participant completed both experiments in approximately 35–40 minutes on average.

03 / Stimulus Preparation

Building reliable speech stimuli

Careful stimulus preparation was essential to ensure consistency and reliability throughout the perception experiments. I designed the recording procedure, managed the recording sessions, screened the audio tokens, and prepared the final stimuli for experimental presentation.

01 / Design

Stimulus Design

Designed Korean vowel stimuli using a two-syllable V-da structure. The following consonant was selected to minimise potential effects on the preceding vowel and to account for phonotactic differences between Korean and Chinese.

02 / Recording

Recording Production

Organised recording sessions with three trained student broadcasters from Korea University’s campus broadcasting station in Seoul. Recordings were made in a studio using a TASCAM recorder at 44.1 kHz, 16-bit resolution, and mono format. An iPad-based presentation procedure was designed to minimise incidental noise during recording.

03 / Screening

Audio Screening & Selection

Reviewed 153 recorded tokens, covering 17 Korean vowels produced three times by each speaker. Evaluated pronunciation clarity and selected one representative token for each vowel from each speaker, resulting in 51 final stimuli.

04 / Validation

Acoustic Review & Validation

Examined acoustic properties using formant measurements (F1 and F2). Conducted an independent listening validation with two Korean phonetics graduate students. Both listeners correctly identified all 51 stimuli, achieving 100% accuracy.

05 / Preparation

Experimental Presentation

Prepared standardised audio presentation files using Praat and OpenSesame. Each stimulus followed a consistent playback sequence, with an attention signal, repeated audio presentation, and controlled silent intervals. Stimulus order was randomised to ensure consistent presentation across participants.

Recording & Validation Overview

153

Recorded tokens

51

Selected stimuli

3

Recording speakers

100%

Validation accuracy

Research Tools

PraatOpenSesamePowerPointTASCAM Recorder

04 / Data Analysis

From responses to evidence

Experimental responses were organised in Excel and processed using Python and Pandas. Matplotlib and R were used to visualise response patterns and analysis results, while R was also used to conduct statistical tests. The dataset included Mandarin sound mappings, seven-point similarity ratings, and Korean vowel identification responses.

Final Analytical Sample

38

Completed experiments

8

Excluded participants

15

Novice learners

15

Advanced learners

Eight participants were excluded based on the study's selection criteria, including consistency of experimental instructions and comparability between learner groups. The final dataset included 30 participants, equally divided between the two groups.

01 / Data Processing

Organising experimental responses

Experimental responses were organised in Excel and processed using Python and Pandas. Matplotlib and R were used to visualise response patterns and statistical results, while R was also used to conduct statistical tests and examine differences between learner groups.

Response Organisation

Structured participant responses and organised experimental data for analysis.

Data Processing

Processed response frequencies, percentages, similarity ratings, and identification accuracy.

Group Comparison

Examined response patterns and compared results between novice and advanced learners.

02 / Statistical Analysis

Analysing patterns and testing differences

Experiment 01

Perceptual Assimilation

Examined how learners associated Korean vowels with Mandarin sounds and how consistently particular sound mappings occurred.

Analysis Methods

  • • Response frequency and percentage analysis
  • • Seven-point mean similarity ratings
  • • 50% assimilation threshold
  • • One-sample t-tests
  • • Wilcoxon signed-rank tests
  • • PAM-L2 assimilation classification

Statistical tests were used to examine whether particular Mandarin sound mappings occurred more frequently than expected by chance. When the parametric test assumptions were not satisfied, the Wilcoxon signed-rank test was used as an alternative.

Experiment 02

Vowel Identification

Evaluated how accurately learners identified Korean vowels and examined differences in identification patterns between the two groups.

Analysis Methods

  • • Response frequency and accuracy analysis
  • • Vowel response cross-tabulation
  • • Shapiro–Wilk normality test
  • • Levene's test for homogeneity of variance
  • • Independent-samples t-test
  • • Wilcoxon rank-sum tests

Group-level and individual-vowel comparisons were used to examine identification accuracy and recurring error patterns.

03 / Key Results

Comparing vowel identification accuracy

The average identification accuracy across all Korean vowels was 87% for novice learners and 91% for advanced learners. However, the difference between the two groups was not statistically significant.

01 / Overall Vowel Identification Accuracy

Mean accuracy across all 17 Korean vowels.

Novice learners87%
Advanced learners91%

The four-percentage-point difference was not statistically significant (p > .05). The study therefore did not find sufficient evidence of a difference in overall identification accuracy between the two groups.

Source: Reconstructed from the overall results reported in the author's master's thesis (2020).

Consistently Identified Vowels

Nine vowels achieved identification accuracy of 90% or above in both learner groups: /아/, /이/, /우/, /에/, /야/, /예/, /와/, /웨/, and /의/.

Recurring Identification Difficulties

Some vowels showed lower identification accuracy and recurring confusion patterns, particularly /오/ and /어/ among novice learners.

Research Tools

PraatExcelPythonPandasMatplotlibR

05 / Key Findings

Perceptual assimilation patterns

The heatmaps show how each Korean vowel stimulus was mapped to Mandarin vowel responses by novice and advanced learners. Cell colour represents the response percentage; the smaller value underneath is the mean goodness-of-fit rating reported in the thesis. Blank cells indicate no response recorded in the table.

Table 22 · Thesis p. 44

Novice learners

17 Korean vowel stimuli

Stimulus
a
o
üe
ei
i
ao
ou
u
ai
e
ia
iao
iou
io
ü
ie
ua
uei
uo
uai
e.i
e.ai
ei.i
mao.er
o.i
ruai
rui
uei.i
uei/uai
uo.ao
u.ao
u.ei
u.o
u.ua
u.i
ia.e
ie.ai
i.ai
i.ei
iou.e
iou.er
ü.i
üe.i
/아/
100%6.80
/이/
100%6.69
/어/
29%6.23
58%5.88
13%5.33
/오/
7%5.00
4%4.50
4%5.00
73%5.91
2%4.00
2%4.00
4%6.00
2%5.00
/우/
100%6.16
/에/
27%5.83
71%5.69
2%4.00
/으/
100%5.40
/야/
98%6.34
2%5.00
/여/
84%5.79
11%5.80
2%7.00
2%5.00
/요/
9%5.50
84%5.45
4%7.00
2%5.00
/유/
89%5.15
4%6.50
2%4.00
4%5.00
/예/
87%5.90
2%6.00
9%5.50
2%3.00
/와/
98%6.39
2%4.00
/위/
4%4.00
9%4.25
69%4.77
4%4.00
2%6.00
2%3.00
7%6.00
2%6.00
/워/
76%5.91
2%4.00
7%6.00
13%4.67
2%6.00
/웨/
22%5.30
73%5.39
2%5.00
2%5.00
/의/
7%4.33
4%6.00
2%5.00
84%5.76
2%7.00
Lower response rate Higher response rateBold italic: 50% or aboveSmaller number: mean goodness-of-fit (7-point scale)

Table 33 · Thesis p. 54

Advanced learners

17 Korean vowel stimuli

Stimulus
a
i
o
io
ü
u
üe
ao
ai
e
ei
ia
iao
iou
ie
ua
uei
uo
uai
e.i
ou
o/u
re
uo.ao
uo.e
uo.o
u.ai
u.ao
u.e
u.i
ie.ai
i.i
iou/ü
ü.i
/아/
100%6.62
/이/
100%6.51
/어/
16%4.57
42%5.74
13%4.67
4%4.50
24%6.00
/오/
16%5.86
73%5.76
2%5.00
4%6.50
2%5.00
2%6.00
/우/
98%6.05
2%5.00
/에/
78%5.40
22%4.80
/으/
98%4.36
2%3.00
/야/
100%6.42
/여/
73%5.52
24%4.91
2%1.00
/요/
2%6.00
7%3.33
91%5.44
/유/
4%6.00
91%5.32
4%5.00
/예/
98%5.70
2%6.00
/와/
100%6.13
/위/
33%4.20
2%3.00
56%4.44
7%5.33
2%4.00
/워/
2%5.00
80%4.78
2%4.00
2%6.00
4%4.50
7%6.00
2%4.00
/웨/
2%3.00
20%5.11
73%4.64
2%3.00
2%2.00
/의/
44%2.60
16%4.43
38%5.65
2%2.00
Lower response rate Higher response rateBold italic: 50% or aboveSmaller number: mean goodness-of-fit (7-point scale)

Source: Reconstructed from Tables 22 and 33 in the author's master's thesis (2020). The response categories follow the original tables and are not merged across groups.

Interpreting the patterns

/오/ – /우/

Both vowels were commonly assimilated to the Mandarin sound /u/.

One of the most difficult contrasts to identify.

/요/ – /유/

Both vowels were commonly assimilated to the Mandarin sound /iou/.

One of the most difficult contrasts to identify.

/어/ – /오/

A notable confusion pattern appeared, particularly among novice learners.

The /어/ vowel was frequently misidentified as /오/ in the novice group.

Learning experience did not guarantee better identification

The overall difference in identification accuracy between novice and advanced learners was not statistically significant. The perceptual assimilation results nevertheless revealed distinct Mandarin sound mappings across Korean vowel stimuli and learner groups.

06 / Research Contribution

Connecting speech perception research with language education.

By examining the full Korean vowel system rather than limiting the analysis to selected vowel categories, the study identified assimilation patterns that may be overlooked by a more restricted approach. The findings also informed recommendations for explaining Korean vowel pronunciation to Chinese-speaking learners.

Academic research · Korean phonology · Speech perception01 / Master's Thesis