合作原则视角下对大模型“会话含义”的识别缺陷分析——以ChatGPT模拟医患对话为例

Identifying Deficiencies Analysis in Large Language Models’ Recognition of Conversational Implicature from the Perspective of Cooperative Principle—A Case Study of ChatGPT-Simulated Doctor-Patient Dialogue

摘要: 本文从合作原则视角，分析ChatGPT在医患对话中“会话含义”识别缺陷。研究发现，其在量、质、关系及方式准则上存在信息不足或冗余、输出错误信息、回复偏离核心及表达晦涩等问题，根源在于训练数据局限、算法语义与情感分析不足及医学领域特殊性。建议通过优化数据、改进算法及融合医学知识提升其识别能力，以推动大模型在医疗领域安全有效应用，平衡技术与人文关怀。

Abstract: This paper analyzes the deficiencies of ChatGPT in recognizing conversational implicature in doctor-patient dialogues from the perspective of the Cooperative Principle. The study finds that ChatGPT exhibits problems such as insufficient or redundant information, output of false information, responses that deviate from the core issue, and obscure expressions, which correspond to violations of the maxims of Quantity, Quality, Relation, and Manner respectively. The underlying causes include limitations in training data, insufficient semantic and sentiment analysis in the algorithm, and the unique characteristics of the medical domain. Suggestions are made to improve ChatGPT’s recognition capability by optimizing training data, refining algorithms, and integrating medical knowledge, in order to promote the safe and effective application of large language models in the medical field while balancing technological advancement with humanistic care.