Wong, Sheung PingSheung PingWongDr. CHAN Kin Wing2026-06-262026-06-262025Wong, S. P., & Chan, K. W. (9 Jun 2025). The SYU Cantonese segmentation chatbot (SYU-CSC): Innovations in Cantonese natural language processing. Academia Sinica Linguistics Forum 5, 2nd Conference Room (3/F), Humanities and Social Sciences Building, Academia Sinica, Taipei, Taiwen.http://hdl.handle.net/20.500.11861/27772Cantonese segmentation poses significant challenges for natural language processing (NLP) due to the lack of word boundaries in the Chinese script, as well as the “linguistic duality” (文白相混) often occurring in speech, where both vernacular spoken Cantonese and standard written Chinese (SWC) lexical forms may be produced in utterances (Xiang et al. 2022, Jiang et al. 2025). Whereas Cantonese segmentation tools have emerged in recent years (Shen et al. 2021, Lee et al. 2022), they remain reliant on corpora and dictionaries for parsing, which limits their ability to handle idiomatic expressions, colloquialisms, and slang. Overall, Cantonese remains a low-resource language, given that the performance of large language models (LLMs) for Cantonese continues to lag behind those for SWC/Mandarin and English (Jiang et al. 2025). <br> The present paper will introduce the capabilities of the SYU Cantonese Segmentation Chatbot (SYU-CSC), a Cantonese-specific segmentation tool developed by the Language Centre at Hong Kong Shue Yan University (HKSYU), representing a novel research method of using artificial intelligence (AI) to conduct Cantonese segmentation. Preliminary testing (Chan and Chan in press) has already found that SYU-CSC achieves greater accuracy in segmentation, part-of-speech tagging, and Cantonese/SWC lexical classification compared to other LLMs such as GPT-4o and ERNIE Bot (文心一言). This paper will demonstrate how the functionalities of SYU-CSC can be deployed for basic textual analysis tasks to facilitate multilingual translation, corpus creation, and further linguistic analysis. This demonstration is supported by quantitative experimental results from an expanded training dataset. By further validating the performance of SYU-CSC, this study hopes to contribute to a potential breakthrough in NLP for Cantonese through originating an effective multifunctional tool capable of addressing the intricacies present in the Cantonese language.enCantonese SegmentationArtificial Intelligence (AI) ChatbotsVernacular Spoken CantoneseStandard Written Chinese (SWC)Part-Of-Speech TaggingSYU Cantonese Segmentation Chatbot (SYU-CSC)Conference Paper