大语言模型新兴语法解读能力评估:以完成体“有”为例

Evaluation of the Syntax Interpretation Ability of Large Language Models:Taking the Perfective Aspect“You”as an Example

文章评估ChatGPT 4.o、文心一言3.5、Monica以及Claude Instant这4类大语言模型在解读语法标记时的表现,特别是它们对汉语新兴完成体标记“有”的解读能力,同时辅以人类解读能力作为评估基线。结果表明,尽管新一代大语言模型在语法解读方面已有显著提升,但与人类解读能力相较,在处理完成体标记“有”时仍表现不佳,表明各类大语言模型对新兴功能语素的推断能力有限。

The article evaluates the performance of four Large Language Models——ChatGPT 4.o,Wenxin Yiyan 3.5,Monica,and Claude Instant——in interpreting grammatical markers,particularly their ability to interpret the emerging perfective marker“you(have).”This article also uses human interpretation as a baseline for assessment.The results show that although the new generation of Large Language Models has made significant improvements in grammatical interpretation,they still perform poorly when dealing with the perfective marker“you(have),”indicating limited inference ability for emerging functional morphemes,compared to human interpretation ability.