Acoustic Representation and Realization of Weak Elements Subcategories: In the Case of Tianjin Mandarin

Zhijie Li, Hui FengInterspeech 2025

TL;DR

This study investigates how "weak elements" (unstressed syllables) in Tianjin Mandarin are phonetically produced and realized, focusing on acoustic differences across various subcategories. The research reveals that while vowel quality changes are consistent across weak elements, their tonal realization follows two distinct patterns: target-setting and tone spreading, depending on the specific subcategory.

What Problem Does This Paper Address?

In linguistics, "weak elements" or unstressed syllables are common across languages and often exhibit reduced duration and altered vowel quality. However, in tonal languages like Mandarin, their realization is more complex due to the interplay of tones. While previous research has explored weak elements in Mandarin, there's no consensus on the specific acoustic mechanisms governing their realization, especially concerning their tonal contours. This paper aims to fill this gap by examining the acoustic representation and realization of different subcategories of weak elements in Tianjin Mandarin, a dialect with a unique tonal system, to understand if distinct phonetic and tonal realization strategies exist.

Methods

The study employed a combination of acoustic analysis and statistical modeling. Six native Tianjin Mandarin speakers were recorded reading sentences containing disyllabic words with weak elements.

  • Acoustic Measurements: The researchers analyzed the first two formants (F1, F2) and fundamental frequency (F0) of the weak elements. For diphthongs, formants were sampled across ten equal segments, while for monophthongs, average F1 and F2 values were calculated.
  • Normalization: Formant data was normalized using z-scores to account for individual vocal tract differences.
  • Vowel Quality Analysis: Vowel Formant Dispersion (VFD) was calculated to assess vowel quality and distribution.
  • Tone Analysis: Fundamental frequency (F0) was converted to semitones and normalized. Linear mixed-effects models were used to analyze pitch contours and their relationship with subcategories and preceding tones.

Datasets

The study utilized a corpus of sentences specifically designed to elicit weak elements from various subcategories within disyllabic word groups. The corpus ensured that each of the four full tones in Tianjin Mandarin (T1-T4) preceded the weak elements across all subcategories. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Consistent Vowel Quality, Diverse Tone Realization: A striking finding is that while the acoustic changes in vowel quality (F1 reduction, F2 centralization) are largely consistent across all weak element subcategories, their tonal realization patterns diverge significantly.
  • Two Distinct Tonal Mechanisms: The study identified two primary strategies for weak element tone realization in Tianjin Mandarin:
  • Target-Setting: Certain subcategories (affixes, locative prepositions, directional verbs, reduplicated words, and optional weak elements) appear to aim for a fixed mid-low pitch target, irrespective of the preceding tone.
  • Tone Spreading: Other subcategories (structural particles, aspectual particles, habitual weak elements, and functional weak elements) are more context-dependent, with their tonal realization being influenced by the tones of adjacent syllables.
  • Dialectal Specificity: The unique tonal system of Tianjin Mandarin provides a crucial context for observing these distinct realization patterns, suggesting that dialectal variations can lead to different phonological strategies for weak elements.

Key Results

  • Vowel Distribution: Across monophthongs, weak elements showed a general reduction in F1 and centralization of F2. Back vowels ([o], [u]) exhibited a slight fronting (increase in F2), while front vowels ([a], [i], [1]) and [r] showed centralization (reduction in F2). Diphthongs displayed similar patterns of reduction and centralization.
  • Vowel Formant Dispersion (VFD): VFD was generally greater for non-weak vowels than for weak vowels, indicating that weak element vowels are positioned closer to the vowel space centroid, suggesting reduced dispersion.
  • Tone Realization:
    • Subcategories like structural particles, aspectual particles, habitual weak elements, and functional weak elements showed significant differences in pitch targets based on preceding tones.
  • Affixes, locative prepositions, directional verbs, reduplicated words, and optional weak elements consistently targeted a mid-low pitch, suggesting a target-setting mechanism.
  • Linear mixed-effects models confirmed that both subcategories and preceding tones significantly influenced the intercept (overall pitch height) and linear trend (directional movement) of the pitch contour. Subcategories also significantly affected the quadratic trend (curvature).

Limitations and Future Work

  • Participant Pool: The study acknowledges that the participant pool was relatively small (six speakers), and expanding this to capture broader variability is a future direction.
  • Conversational Context: The data was collected from elicited speech. Investigating weak elements in natural, conversational contexts would provide further insights into their real-world realization.
  • Underlying Mechanisms: While the study identified target-setting and tone spreading as realization strategies, the underlying phonological or cognitive mechanisms driving these patterns remain unclear. Future research could employ perceptual experiments or neuroimaging techniques to explore how listeners process weak elements and how they are represented mentally.

Summary updated

BibTeX
@inproceedings{li2025acoustic,
  author = {Zhijie Li and Hui Feng},
  title = {{Acoustic Representation and Realization of Weak Elements Subcategories: In the Case of Tianjin Mandarin}},
  booktitle = {Interspeech},
  year = {2025},
}
Figure