Resolving Chinese Anaphora with ChatGPT
Shuangshuang Chen · 2024
This study evaluates ChatGPT's ability to resolve Chinese anaphora using a newly developed Chinese collection of Winograd schemas, achieving an impressive accuracy rate of 91.5%. The results illustrate ChatGPT's effectiveness across multiple languages, placing it favorably alongside other prominent language models and showcasing its capabilities in addressing complex anaphora resolution challenges. Employing a refined methodological approach, the research enhanced schema collections to better capture Chinese linguistic idiosyncrasies and integrated detailed reasoning analyses for model evaluations. The findings highlight the considerable promise of transformer-based models like ChatGPT in advancing Natural Language Processing (NLP) applications beyond English, contributing to global linguistic inclusivity. Furthermore, the study emphasizes the necessity for ongoing research aimed at expanding language model coverage to include other languages and incorporating more extensive contextual and commonsense knowledge into NLP technologies.