Design and Collection of an L2 English Corpus with a Suprasegmental Focus for Chinese Learners of English

Mingxing Li, Shuang Zhang, Kun Li, Alissa M. Harrison, Wai-Kit Lo, Helen MengICPhS 2011

Introduction

Presented at ICPhS 2011, this research details the development of a specialized speech corpus aimed at understanding how Chinese learners of English acquire suprasegmental features—the rhythm, stress, and intonation that constitute the "music" of a language. While individual sounds (segmentals) are often the focus of language instruction, prosody plays a critical role in perceived fluency and communicative effectiveness.

The Prosodic Gap: Chinese vs. English

Chinese and English represent distinct prosodic typologies. Chinese is a tonal, syllable-timed language, whereas English is non-tonal and stress-timed. This fundamental difference often leads to "prosodic transfer," where learners apply the rhythmic and melodic patterns of their native Mandarin or Cantonese to English speech. Such interference can result in a "staccato" rhythm that hinders intelligibility. Prior to this study, there was a significant lack of large-scale, systematically annotated datasets to analyze these specific L2 deviations.

Methodology and Corpus Design

The researchers constructed a controlled recording environment to capture five key prosodic dimensions: lexical stress, utterance-level stress (narrow focus), intonation, function word reduction, and prosodic disambiguation. The resulting corpus includes over 25,000 utterances from 217 speakers (109 Mandarin, 108 Cantonese), providing a robust dataset for analysis.

Mapping of corpus components to specific annotation types

Annotation Framework

The annotation process combined automated and manual techniques. Initial phonemic segmentation was performed using a custom speech recognizer for forced alignment, followed by manual refinement in Praat by trained linguists. The researchers utilized two primary schemes:

  • ToBI (Tones and Break Indices): A standard for detailed intonation and phrasing.
  • RULF (Rising, Upper, Lower, Falling): A simplified "coarse" labeling scheme designed to facilitate easier evaluation and integration into automated tools.
Multi-layered annotation of a prosodic disambiguation sentence showing ToBI and RULF tiers

Key Insights

  • The "Staccato" Effect: Qualitative analysis confirmed that Chinese learners tend to produce pitch accents on nearly every syllable, reflecting the syllable-timed nature of their native languages. This creates a rhythmic transfer effect that differs significantly from the stress-timed "gallop" of native English.
  • Intonation Deviations: A striking 92% of participants used a low-rising contour (L-H%) for yes-no questions, rather than the high-rising contour (H-H%) typical of native English speakers.
  • Selective Stress Errors: Errors in word stress were found to be item-specific rather than systemic. For example, while many words were handled correctly, 63% of speakers struggled with "misunderstand," and 31% with "refrigerator."

Conclusion and Future Directions

This corpus provides a "gold standard" for studying L2 prosody. By combining detailed ToBI labels with the more accessible RULF scheme, the authors have created a resource that is valuable for both linguistic research and the development of Computer-Aided Language Learning (CALL) technologies. Future work will focus on using this data to build automated feedback systems that help learners master the rhythm and melody of English speech.

Summary updated

BibTeX
@inproceedings{li2011design,
  author = {Mingxing Li and Shuang Zhang and Kun Li and Alissa M. Harrison and Wai-Kit Lo and Helen Meng},
  title = {{Design and Collection of an L2 English Corpus with a Suprasegmental Focus for Chinese Learners of English}},
  booktitle = {ICPhS},
  year = {2011},
}
Figure