Justice in China: Showing the way with forensic voice comparison

Project summary

Legal justice hinges on the quality of forensic evidence evaluation. This is particularly true for forensic voice comparison (FVC) in China, where the scientific bases of methods and procedures are limited. A recent survey revealed a number of serious issues, notably 1) the lack of empirical validation of features analysed by FVC practitioners; 2) potential overestimation of the strength of voice evidence when using the current conclusion framework; and 3) the lack of specific strategies for mitigating cognitive bias induced by exposure to task-irrelevant information. This project aims to improve the quality of analysis in FVC cases in China by 1) empirically validating a number of speech features commonly used in FVC casework in Mandarin; and 2) exploring Chinese FVC practitioners’ perspectives on issues related to conclusion frameworks and cognitive bias.

Our project findings will have a direct impact on FVC practitioners’ evaluation of Mandarin speech evidence in the future. Specifically, our results will assist them in making informed decisions about which speech sounds to analyse and how best to analyze them forensically. The findings will also help practitioners avoid cognitive bias and express their conclusions using an appropriate framework. Theoretically, the findings will contribute to a comprehensive model explaining where speaker-specific information is encoded in the speech signal and why individual voices differ from one another. Ultimately, the anticipated improvement in forensic analysis of Mandarin speech evidence will result in a decrease in miscarriages of justice in China and Chinese communities worldwide.

Collaborators: Dr. Bruce X. Wang, Prof. Paul Foulkes, Prof Cuiling Zhang

Funding

• General Research Fund, Hong Kong Research Grants Council. (PI: Ricky Chan, HKD794,197)

• Seed Fund for Basic Research, University of Hong Kong. (PI: Ricky Chan, HKD68,733)

Forensic voice comparison: acoustic-phonetic and automatic features

Using a forensically-oriented speech corpus, this project compared the speaker discriminatory performance of acoustic-phonetic features and MFCCs under the Bayesian likelihood ratio framework. Based on predictions from phonetic theories, we also explore how acoustic-phonetic features and MFCCs may be combined for optimal speaker-discriminatory performance, and the effects of speech style mismatch and non-contemporaneous recordings. The findings will have direct implications for forensic voice comparison casework.

Collaborator: Dr. Bruce X. Wang

Interim output

• Chan, R. & Wang, B. (2024). Do long-term acoustic-phonetic features and mel-frequency cepstral coefficients provide complementary speaker-specific information for forensic voice comparison? Forensic Science International, 112199.

Bilingual language coactivation and implicit learning

Project summary

Abundant research has shown that bilinguals cannot completely switch off their non-current language during real-time L2 lexical processing. However, previous studies have predominantly focused on how the covert activation of the non-current language influences trial-by-trial processing in the current language; whether such covert activation affects the development of new implicit linguistic knowledge during L2 processing remains an empirical question. This project aims to determine if language activation can be manipulated to affect implicit learning of form-meaning connections, thereby bridging research in the fields of bilingual language coactivation and implicit language learning. The expected findings will also inform educationalists and L2 learners on how L2 implicit learning outcomes can be optimized through the manipulation of learners’ language activation.

Collaborators: Dr. Dave Cayado, Dr. Bronson Hui

Interim outputs

• Chan, R. Cayado, D. & Hui, B. (resubmitted after minor revision). Covert activation of bilinguals' non-current language and implicit learning (registered report). Language Learning.

• Cayado, D., Chan, R. (2023) The influence of prior linguistic knowledge on L2 semantic implicit learning: Evidence from Cantonese-English bilinguals. Language Learning, 73(3), 657-682.
   

Why are lexical tones difficult to learn? Insights from incidental learning of tone-segment connections

Project summary

Tone languages (e.g. Cantonese, Mandarin, Thai) employ contrastive pitch patterns (i.e. lexical tones) to distinguish word meaning. Why L2 lexical tones are difficult to acquire, especially for learners with non-tone L1, has been a long-standing issue for researchers, language teachers and learners alike. While L2 sounds present different kinds of challenges to learners at the phonetic, phonological, and lexical levels, previous studies on L2 tone learning mostly focused on the phonetic level (e.g.  L2 tone discrimination/identification), and and a few on the lexical level. However, the ability to perceive or discriminate different L2 tones explicitly does not entail the ability to encode pitch patterns as abstract tone categories at the syllable level, which is the pre-requisite of using abstract tone categories as lexical cues.

The project mainly investigates the role of 1) L1 prosodic system; 2) musical training; and 3) proficiency in a tone language on the forming novel abstract syllable-level tone categories (phonological level), via the incidental learning (i.e. without intention) of tone-segment mappings.

Funding

• Faculty of Arts and Social Sciences Research Fund, Lancaster University. (PI: Ricky Chan, GBP£2200)

• Small Project Funding, HKU. (PI: Janny Leung; Co-I: Ricky Chan, HKD 72,800)

Interim outputs

• Chan, R. & Leung, J. (2020). Why are lexical tones difficult to learn? Insights from the incidental learning of tone-segment connections. Studies in Second Language Acquisition, 42(1), 33-59.

• Chan, R. (2019). Difficulty in learning L2 tones: Insights from the incidental learning of tone-segment mappings. Proceedings of the 17^th International Congress of Phonetic Sciences. Melbourne, Australia.

Tonal parameters for forensic speaker comparison

Project summary

This project examines speaker variability in the production of lexical tones and its implications for forensic voice comparison. Based on spectral and statistical analyses of controlled production data, it has been demonstrated that the dynamic contours of Cantonese and Mandarin citation tones and co-articulated tones exhibit significantly greater between-speaker variation than within-speaker variation across different speaking rates, tonal contexts, and voice levels. Further analysis reveals that tone inventory density has a clear effect on the degree of variation in tone coarticulation, potentially attributable to speakers’ effort to maintain perceptual contrast. The findings demonstrate that lexical tones may provide useful parameters for forensic voice comparison.

A further study explores the speaker-discriminatory performance of various approaches to tonal f0 parameterisation. Findings suggest that modelling surface tone f0 generally yields better results than modelling the underlying articulatory mechanism of tone production.

Funding

• Doris Zimmern HKU-Cambridge Hughes Hall Scholarships. (Ricky Chan, GBP 89,237)

• Cambridge Overseas Trust, University of Cambridge. (Ricky Chan, GBP 17,000)

• Fieldwork Funding, University of Cambridge. (Ricky Chan, GBP 5,700)

• The International Association for Forensic Phonetics and Acoustics (IAFPA) Research Grants. (Ricky Chan, GBP1,250)

Interim outputs

• Chan, R. (accepted). Tone languages. In F. Nolan, K. McDougall & T. Hudson (Eds), Oxford Handbook of Forensic Phonetics. Oxford University Press.

• Chan, R., Wang, B. (2024). Modelling lexical tones for speaker discrimination. Language and Speech. Advance online publication.

• Chan, R. (2020). Speaker discrimination: citation tones vs. coarticulated tones. Speech Communication, 117, 38-50.

• Chan, R. (2016). Speaker variability in the realization of lexical tones. International Journal of Speech, Language and the Law, 23(2), 195-214.

• Chan, R. (2015). Speaker variability in the production of coarticulated Tones. Proceedings of the 16th International Congress of Phonetic Sciences. Glasgow, UK.

Implicit learning of form-meaning connections: Evidence from eye tracking data

Project summary

Whether form-meaning connections can be learnt implicitly has important implications for theories of language processing and learning, but most relevant studies focused on participants’ awareness of their linguistic knowledge; the lack of awareness during the learning process has not yet been convincingly demonstrated. This project aims to determine whether real-time language learning can take place implicitly.

Funding

• Seed Fund for Basic Research, University of Hong Kong. (PI: Ricky Chan, HKD 150,000)

• Faculty of Arts Start-up Grant, University of Hong Kong. (PI: Ricky Chan, HKD  180,000)

Effects of syllable structure on tonal coarticulation

Project summary

This project investigates how syllable structure may affect tonal alignment, tone trajectory, and tonal coarticulation in Cantonese.

Cross-linguistic study of prominence perception

Project summary

This international collaborative project uses the same procedure and stimuli to test prominence perception for speakers of typologically distinct and related varieties. In varying the prosodic properties of the languages tested, this study aims to tease apart the universal and the language-specific aspects of perceptual prominence.

Interim output

• Leemann, A., Kolly, M. J., Li, Y., Chan, R., Kwek, G. & Jesperson, A. (2016). Towards a typology of prominence perception: the role of duration. Proceedings of the 8th International Conference on Speech Prosody (SP2016). Boston, MA.