汉日语境中大语言模型的礼貌知识生成能力探析  

A Study of Politeness Knowledge Generation Capabilities of Large Language Models in Chinese-Japanese Contexts

作　　者：李瑶 于富喜[2] 毋育新[3] Li Yao;Yu Fuxi;Wu Yuxin(Northwest Normal University;Xi'an International Studies University)

机构地区：[1]西北师范大学外国语学院 [2]西安外国语大学商学院 [3]西安外国语大学日本文化经济学院

出　　处：《日语学习与研究》2024年第5期113-127,共15页Journal of Japanese Language Study and Research

基　　金：2021年度国家社会科学基金西部项目“面向人机对话系统的日语语用距离调节机制研究”(批准号:21XYY009)的阶段研究成果。项目主持人:毋育新。

摘　　要：迄今为止,礼貌是语用学领域经久不衰的话题。在人工智能时代,提升大语言模型在语言研究与语言教学中的重要地位离不开大语言模型生成准确且到位的礼貌知识体系。为此,本研究聚焦“礼貌研究的发展历程以及各阶段的代表性理论”和“礼貌与敬语的关系以及敬语的分类”两大主题,以文心一言、ChatGPT、Hugging face三种大语言模型在汉日语境中生成的相关内容为切入点,使用质性分析法探讨其生成能力。研究发现,在汉日语境中,三种大语言模型均具备良好的对话能力和强大的知识涌现能力,但是较之于人类围绕上述两大主题总结的研究成果,其生成的内容存在不准确、不完全、不稳定的特征,介绍的代表性理论具有明显的滞后性;对自身生成礼貌知识的准确度评价不到位;尚难在语言研究与语言教学中扮演礼貌知识提供者与评估者的角色。语言学与计算机科学等学科的交叉融合迫在眉睫。Politeness is a hot topic in the field of pragmatics.In the age of artificial intelligence,enhancing the role of large language models in language research and language teaching is inseparable from the accurate generative capabilities of politeness.This study employs qualitative analysis to examine large language models'generative capabilities of politeness in Chinese and Japanese contexts,focusing on two main themes:"the elaboration of politeness research and its classical theories"and"the relationship between politeness and honorifics,and the categorization of honorifics."It was found that in both Chinese and Japanese contexts,large language models exhibit good conversational ability and strong knowledge generation.However,the contents they produce are often incomplete,inaccurate,unstable,and the related theories show a significant lag.Large language models don't effectively evaluate their accuracy in the generation of politeness,and struggle to serve as a content provider and evaluator in language research and language teaching.Cross-disciplinary collaboration between linguistics,computer science and other fields is urgently needed to address these challenges.

关 键 词：人工智能 大语言模型 礼貌 敬语 汉日对比研究 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] H36[自动化与计算机技术—控制科学与工程] H15[语言文字—日语]