基于大语言模型的词典自动化编纂实验及思考  

An Experimental Study and Reflections on Automated Lexicographic Compilation Using Large Language Models

作　　者：柳长青 Liu Changqing

机构地区：[1]广东外语外贸大学外国语言学及应用语言学研究中心/词典学研究中心,广东广州510420

出　　处：《语言战略研究》2025年第2期61-72,共12页Chinese Journal of Language Policy and Planning

基　　金：教育部哲学社会科学研究重大攻关项目“百年来欧美外来词的全息语料库建设与‘汉语化’机制研究”(21JDA050);广东省高校科研基金重大培育项目“语言资源与外语教学研究”(2018WZDXM011);广东外语外贸大学引进人才科研启动项目。

摘　　要：大语言模型的出现对词典自动化编纂产生了重大影响。基于词向量和权重参数优化等基本原理,广东外语外贸大学词典学研究中心研发的“百年来欧美外来词语言资源平台”系统,引入大语言模型对外来词语料库等进行训练和调优,同时建立完善的人工审核机制,历时4000余小时生成3万余条外来词自动释义、溯源和例句。通过平台自有模型和ChatGPT对外来词释义准确率的对比实验,证明该编纂方法的有效性和可行性,发现持续更新数据库、加强文献研究、丰富专用语料库、强化人工检查,将对生成高质量释义产生较大影响。大语言模型较之传统手工编纂有碾压性优势,其应用将冲击现有电子词典形态,改变未来的词典查询行为。词典编纂专家与人工智能专家需要协同合作,对模型输出的内容进行质量控制。而版权问题也将转化为数据归属权问题,需要各方专家共同制定相关规则。大语言模型在汉语词典自动化编纂方面的应用潜力不应被低估,预计它还能被扩展至其他语言和专业领域的词典编纂中,进一步促进词典学领域的新突破。The emergence of large language models(LLMs)has produced a profound impact on the automation of lexicography.Based on fundamental principles such as word vectors and weight parameter optimization,the“Centurylong Western Loanword Language Resource Platform”developed by the Center for Lexicographical Studies at Guangdong University of Foreign Studies integrates LLMs to train and fine-tune a loanword corpus.With the support of a comprehensive manual review mechanism,the system generated over 30,000 automated definitions,etymologies,and example sentences for loanwords over a span of more than 4,000 hours.Comparative experiments on the accuracy of loanword definitions using the platform’s proprietary model and ChatGPT demonstrated the effectiveness and feasibility of this compilation method.Key findings suggest that continuously updating databases,strengthening literature research,enriching specialized corpora,and enhancing manual review processes signifi cantly contribute to generating high-quality defi nitions.LLMs exhibit overwhelming advantages over traditional manual compilation methods,and their application is poised to disrupt the current forms of electronic dictionaries,reshaping future dictionary lookup behaviors.Collaboration between lexicography and artificial intelligence experts is essential to ensure quality control over model outputs.Additionally,copyright issues will evolve into questions of data ownership,requiring experts from various fields to establish relevant regulations.Overall,the potential of LLMs in the automated compilation of Chinese dictionaries should not be underestimated,and their application is expected to extend to lexicography in other languages and specialized domains,fostering new breakthroughs in the field.

关 键 词：大语言模型 词典编纂 人工智能 自动化 外来词 

分 类 号：H002[语言文字—语言学]