Climbing towards the NLU of the universal reading of shei ‘who’

Abstract The wh -expressions shei ‘who’ and shenme ‘what’ in Mandarin Chinese not only convey an interrogative meaning but also exhibit existential and universal readings in specific contexts ( Huang 1982 , Cheng 1991 , 1995 , Li 1992 , Tsai 1994 , Lin 1996 , 1998 ). Focusing on the universal interpretation of shei , this paper has three objectives. First, we demonstrate that current state-of-the-art large language models (LLMs) such as ChatGPT, lack reliability in distinguishing these three distinct readings of shei . Second, we develop a specialized natural language processing and understanding (NLP/NLU) system capable of processing and interpreting shei across diverse contexts with greater accuracy, transparency, and consistency. Unlike current LLMs, our system is built upon Wang et al.’s ( 2019a , 2019b ) generative linguistics-based NLP/NLU software tools, Articut and Loki, enabling it to require significantly less training data to interpret the universal reading of shei . Third, we compare our model’s performance with that of ChatGPT-4o, demonstrating its superior accuracy and robustness in interpreting the universal reading of shei .