A cross-linguistic acoustic comparison of unreleased word-final stops: Korean and Thai

Kimiko TsukadaInterspeech 2004

TL;DR

This paper investigates the acoustic properties of unreleased word-final stops in Korean and Thai, two languages where these sounds are phonemically contrastive. The study reveals that while the acoustic cues for these stops are similar across both languages, subtle differences in formant trajectories might explain why native speakers can distinguish them more easily than English speakers.

What Problem Does This Paper Address?

Word-final stops in English can be either released or unreleased, and the presence of a release burst significantly impacts intelligibility. However, in languages like Korean and Thai, word-final stops are typically unreleased, yet native speakers can still reliably distinguish them. This raises a crucial question: what acoustic cues do these languages utilize to signal the place of articulation for unreleased stops, and why are these cues less apparent or utilized by speakers of languages like English? This paper aims to characterize these acoustic features to understand the underlying perceptual mechanisms.

Methods

The study involved recording monosyllabic words produced by native Korean and Thai speakers, specifically focusing on vowels (/i/, /a/, /u/) followed by unreleased stops ([p], [t], [k]). The researchers analyzed the formant frequencies (F1 and F2) at the offset of these vowels. They employed a three-way ANOVA to examine the effects of language, stop place, and vowel type on these acoustic measures. Formant trajectories were also visually inspected and normalized to allow for direct comparison across different vowel and stop combinations.

Datasets

The study utilized speech materials from:

  • Two native Korean speakers recorded at the University of Alabama at Birmingham, USA.
  • Four native Thai speakers recorded at MARCS Auditory Laboratories, University of Western Sydney, Australia.

The data consisted of monosyllabic words produced in the speakers' native languages (L1), with vowels /i/, /a/, /u/ followed by word-final stops [p], [t], [k]. A total of 396 tokens (180 for Korean, 216 for Thai) were included in the acoustic analysis.

Key Insights and Ideas

  • Cross-linguistic Acoustic Similarity: The acoustic cues for unreleased word-final stops in Korean and Thai are remarkably similar, suggesting a common set of phonetic strategies for these languages.
  • Vowel-Dependent Stop Cues: The place of articulation of the final stop significantly influences the formant trajectories of preceding vowels, particularly F2. This effect is not uniform across all vowels, with front vowels (/i/) and non-front vowels (/a/, /u/) showing different patterns of F2 change.
  • The Role of F2 Offset: The F2 offset (the frequency of the second formant at the end of the vowel) appears to be a critical acoustic cue. For instance, F2 offset is often higher before [t] for non-front vowels, and higher before [k] for the front vowel /i/.
  • Subtle Differences Matter: While overall acoustic patterns are similar, visual inspection of formant trajectories suggests potential subtle differences in the extent of formant movement between Korean and Thai, particularly for the vowel /u/ before [p] and [k]. These subtle differences might be crucial for native listeners' perception.
  • English Listener Challenges: The paper alludes to previous findings suggesting that English speakers struggle more than native speakers to identify the place of articulation of unreleased stops, implying that the acoustic cues utilized by Korean and Thai speakers are not as readily perceived or interpreted by English speakers.

Key Results

  • Significant Stop Place Effect: The place of articulation of the final stop had a significant effect on F1 and F2 offset values in most instances for both Korean and Thai.
  • Vowel Interaction: The effect of stop place interacted with vowel type, meaning the way the stop influenced the vowel's formants depended on whether the vowel was front (/i/) or non-front (/a/, /u/).
  • Specific F2 Patterns:
    • For /i/, F2 offset was generally higher before [k] than before [p] and [t].
  • For /a/, F2 offset was highest before [t], lowest before [p], and intermediate before [k].
  • For /u/, F2 offset was higher before [t] than before [p] and [k].
  • Language Effect Not Significant (for formants): The main effect of Language (Korean vs. Thai) was not statistically significant for the formant offset values analyzed, suggesting comparable acoustic realization strategies.

Limitations and Future Work

  • Limited Number of Speakers: The study used a small number of speakers (two Korean, four Thai), which might limit the generalizability of the findings.
  • Focus on Acoustic Measures: While acoustic analysis is crucial, the paper acknowledges that articulatory gestures might provide further insights into how these cues are produced and perceived.
  • Perception Experiments: The authors suggest future research should include perception experiments with native Thai and English speakers to directly test how well they can identify unreleased stops in both languages. This would help confirm whether the subtle acoustic differences observed have perceptual consequences.
  • Comparison with English: A direct comparison of acoustic features of unreleased stops in English with those in Korean and Thai is proposed as a future research direction to understand why English speakers might have more difficulty.
  • Investigating Other Cues: While formants are analyzed, other potential acoustic cues (e.g., duration, spectral characteristics) are not explicitly detailed in this section and could be explored further.

Summary updated

BibTeX
@inproceedings{tsukada2004cross,
  author = {Kimiko Tsukada},
  title = {{A cross-linguistic acoustic comparison of unreleased word-final stops: Korean and Thai}},
  booktitle = {Interspeech},
  year = {2004},
}
Figure