Syllable perception depends on tone perception

Iris Chuoying Ouyang, Khalil IskarousInterspeech 2012

TL;DR

This study reveals that in Mandarin, a tone language, the perception of syllable boundaries is not solely determined by acoustic cues like formant patterns. Instead, pitch contours (tones) significantly influence how listeners segment speech, even overriding clear segmental information. This suggests a complex, language-specific interaction between tonal and segmental cues in speech perception.

What Problem Does This Paper Address?

Traditionally, syllable perception has been understood as primarily driven by segmental information (the sounds that make up speech) and the permissible syllable structures within a language. Pitch, which carries lexical meaning in tone languages like Mandarin, has generally been considered independent of syllable perception. However, this paper questions this assumption, investigating whether the crucial role of pitch in tone languages might lead to pitch contours influencing how listeners segment speech into syllables. Specifically, the authors want to know if pitch changes can interfere with or even override spectral (segmental) information in determining syllable count, and if this effect persists across different speech rates.

Methods

The researchers designed a perception study using synthesized Mandarin words. They manipulated two key factors:

  • Pitch Contours (F0): They generated synthetic F0 contours that varied along a continuum from clearly bitonal (two distinct tones) to monotonal (a single, smooth contour). This was achieved using the qTA model, which simulates tonal coarticulation. The idea is that a bitonal contour might suggest two syllables, while a monotonal contour might suggest one.
  • Speech Rate: Target words were embedded in carrier sentences recorded at three different speech rates (fast, medium, slow) and also presented in isolation. This allowed them to examine the effect of F0 changes across varying temporal contexts.

Participants were presented with a target word (either naturally produced or synthesized with manipulated F0) and had to choose between two pictures, one representing a bisyllabic word and the other a monosyllabic word. This forced-choice identification task measured how often listeners perceived a word as monosyllabic, which would indicate that the pitch contour influenced their perception away from the inherent bisyllabic segmental information.

Datasets

The study used synthesized target words based on naturally produced tokens. The specific words and their tonal/vocalic properties are detailed in Table 1 and Table 2 of the paper. No new benchmark datasets were introduced; rather, existing Mandarin words were manipulated for experimental purposes.

Key Insights and Ideas

  • Pitch is a powerful syllable cue in tone languages: The most striking finding is that manipulated pitch contours significantly altered listeners' perception of syllable number, even when the underlying segmental information clearly indicated a bisyllabic word.
  • Tonal influence can override segmental cues: Listeners sometimes perceived a word as monosyllabic when the pitch contour suggested a single tone, despite the acoustic properties of the speech segments pointing towards two syllables. This suggests that the auditory system actively integrates tonal information into syllable segmentation.
  • Language-specific perception: The results support the idea that speech perception mechanisms are shaped by the specific linguistic properties of a language. What constitutes a cue for syllable boundaries can differ across languages, with pitch being a prime example in tone languages.
  • Interaction of cues: The findings highlight that segmental and tonal cues are not processed in isolation but interact dynamically during speech perception.

Key Results

  • F0 contours significantly influenced syllable perception: As F0 contours became more monotonal (suggesting a single tone), participants were more likely to perceive the word as monosyllabic. Specifically, the most monotonal contours induced 20% more monosyllabic choices compared to the most bitonal contours.
  • Speech rate also played a role: Slower carrier sentences led to a higher percentage of monosyllabic choices (15% more than fast carrier sentences), suggesting that temporal factors interact with tonal perception.
  • Vocalic structure matters, but less than tone: While different vocalic sequences ([u.u], [Vu.u], [u.uV]) did affect response tendencies (with [u.u] sequences leading to more monosyllabic perceptions), the impact of F0 contours was more pronounced.
  • No interaction between F0 and speech rate: Interestingly, the study found no significant interaction between the effects of F0 contours and speech rate.

Limitations and Future Work

  • Limited segmental variation: The study controlled for vocalic sequences, but the segmental material was primarily based on naturally produced tokens. Future work could explore a wider range of segmental variations and their interaction with tones.
  • Focus on Mandarin: While the findings are specific to Mandarin, future research could investigate similar interactions in other tone languages with different tonal inventories and structures.
  • Further investigation of vocalic structures: The paper notes that the differences observed between vocalic structures warrant further investigation, particularly regarding the perceptual merging of vowels and glides in diphthongs.
  • Integration of cues: The authors suggest that syllable perception in tone languages involves the integration of information from both pitch and segments, a complex process that could be explored further with more sophisticated modeling.

Summary updated

BibTeX
@inproceedings{ouyang2012syllable,
  author = {Iris Chuoying Ouyang and Khalil Iskarous},
  title = {{Syllable perception depends on tone perception}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure