Exploring the mechanism of tonal contraction in taiwan Mandarin

Chierh Cheng, Yi Xu, Michele GubianInterspeech 2010

TL;DR

This paper investigates how tones in Mandarin Chinese are compressed when words are spoken faster, leading to syllable contraction. They found that speakers still attempt to produce the original tones, even when squeezed into a shorter time, suggesting that the underlying tonal information is preserved rather than deleted. This challenges existing phonological models and points towards an articulatory rather than purely phonological explanation for tone contraction.

What Problem Does This Paper Address?

When people speak quickly, they often shorten words, a phenomenon known as syllable contraction or merger. In tonal languages like Mandarin, this contraction can affect the pitch contours (tones) of syllables. The specific problem this paper tackles is understanding how these tones are compressed during rapid speech in Taiwan Mandarin. Existing phonological theories, like the "Edge-in model," propose that only certain tones (the "edge" tones) are preserved, while others are deleted. However, experimental evidence has been mixed, with some studies suggesting that contracted syllables often have flat or simple pitch contours, making it hard to discern the original tones. This paper aims to resolve this by investigating whether the underlying tonal information is truly lost or simply distorted and compressed due to articulatory constraints.

Methods

The researchers employed a combination of experimental design and advanced statistical analysis.

  • Stimuli Design: They created disyllabic (/ma/+/ma/) nonsense sequences with all 16 possible tone combinations in Taiwan Mandarin. These sequences were embedded in carrier sentences designed to induce different speaking rates.
  • Data Collection: Six native Taiwan Mandarin speakers were recorded producing these stimuli at three different speech rates: slow, natural, and fast.
  • Segmentation: Tokens were classified as "contracted" or "non-contracted" based on the presence or absence of a clear second nasal consonant.
  • Acoustic Measurement: Fundamental frequency (Fo) contours and Fo velocity were extracted for the target syllables.
  • Functional Data Analysis (FDA): This is the core analytical technique. FDA is used to analyze data that are functions (like continuous Fo contours) rather than single numbers.
  • Normalization: Fo contours were normalized to semitones and their average removed to focus on shape.
  • Time Registration: Curves with different durations were "registered" (stretched or compressed) in time to a common duration, using syllable boundaries as landmarks. This allows for direct comparison of curves of different lengths.
  • Functional t-test: This statistical test was applied to compare the average Fo contours and velocities of contracted versus non-contracted syllables, identifying time intervals where differences were statistically significant.

Datasets

The study used a newly created dataset of recorded speech.

  • Custom Stimuli: 16 disyllabic /ma/+/ma/ nonsense sequences with all possible tone combinations were created.
  • Carrier Sentences: These sequences were embedded in carrier sentences designed to manipulate speaking rate.
  • Recordings: Speech from six native Taiwan Mandarin speakers was recorded at three different speaking rates.
  • Total Tokens: The experiment generated a large dataset of 5184 tokens for analysis, allowing for systematic comparisons across different conditions.

Key Insights and Ideas

  • Tonal Targets Persist: Despite extreme compression in contracted syllables, speakers still seem to aim for the original tonal targets. This is evidenced by the presence of similar gestures (rises and falls) in the same order, even if their magnitude is reduced.
  • Articulatory vs. Phonological: The findings suggest that tone contraction is more of an articulatory phenomenon (a consequence of squeezing movements into less time) rather than a purely phonological rule of deleting tones.
  • Edge-in Model Challenged: The results directly contradict the predictions of the "Edge-in model," which posits deletion of non-edge tones. The study shows that even tones not at the "edges" of a disyllabic sequence appear to leave traces.
  • Velocity as a Key Indicator: Analyzing Fo velocity was crucial. It revealed that while the final pitch contour might be flattened in contracted syllables, the underlying movement towards the target tone is still detectable in the velocity profiles.

Key Results

  • HR Tone Contraction: In the comparison of contracted vs. non-contracted HR (High-Rising) tones, Fo contours differed significantly almost everywhere. However, the differences were mainly due to amplitude, indicating that the underlying shape was similar. Velocity profiles also showed differences, particularly in the second half, where the R target was achieved in non-contracted but only attempted in contracted syllables.
  • HR vs. Single H: When comparing contracted HR with a single H tone (as predicted by the Edge-in model), significant differences were found. These differences suggest that the contracted HR still retains some characteristics of the original R tone, rather than being reduced to a simple H.
  • Functional t-tests: These tests confirmed significant differences between contracted and non-contracted curves in many time intervals, but the nature of these differences (amplitude vs. complete absence) was key to the interpretation.

Limitations and Future Work

  • Limited Stimuli: The study focused on disyllabic /ma/+/ma/ sequences. While this allowed for controlled comparisons, it might not generalize to all types of syllable contractions or different phonemic sequences.
  • Binary Classification: The distinction between "contracted" and "non-contracted" was binary, excluding intermediate cases. Future work could explore these intermediate stages.
  • Generalization: The authors mention that other tone combinations were analyzed and yielded similar results, but a full analysis is available online. Further work could explore a wider range of tonal contexts and languages.
  • Future Directions: The authors suggest that future research should focus on the articulatory mechanisms underlying these contractions and explore how these findings might apply to other tonal languages or different types of phonetic reduction. The large dataset collected is also intended for long-term investigation.

Summary updated

BibTeX
@inproceedings{cheng2010exploring,
  author = {Chierh Cheng and Yi Xu and Michele Gubian},
  title = {{Exploring the mechanism of tonal contraction in taiwan Mandarin}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure