摘要
通用大语言模型在古汉语语言信息处理任务上的效果往往不够理想,因此,我们从领域知识学习需求出发,针对古汉语信息处理任务“低资源”“富知识”的特点,从头构建了专门适用于古代汉语理解与生成的大语言模型“AI太炎”。通过合理的模型设计、数据处理、基座训练及微调,仅使用1.8B参数量即可取得较好效果。该模型具有较强的古典文献释读能力,支持句读标点、典故识别、词义解释及文白翻译等多种具有挑战性的文言文理解任务,兼容简体字和繁体字文本。实验显示,与大型通用模型和其他领域模型相比,“AI太炎”在多项古汉语信息处理关键任务上表现出明显优势,且达到了接近或超过人类基线的水平。此外,在辅助古籍整理、辞书编纂和语言研究等方面,该模型也表现出了很大的应用潜力。
Abstract
General-purpose large language models demonstrate notable capabilities in language comprehension and generation, achieving results comparable to or surpassing human performance in many language information processing tasks. Nevertheless, when general models are applied to ancient Chinese texts, their effectiveness is often unsatisfactory. Similarly, fine-tuning open-source foundational models encounter difficulties in adequately incorporating domain-specific knowledge. To address this challenge, this study introduces a large language model, AI Taiyan, specifically designed for understanding and generating ancient Chinese. Experimental results show that with an appropriately designed model, data processing, foundational training, and fine-tuning, satisfactory results can be achieved with merely 1.8 billion parameters. In key tasks related to ancient Chinese information processing such as punctuation, identification of allusions, explanation of word meanings, and translation between ancient and modern Chinese, this model exhibits a clear advantage over both general-purpose large models and domain-specific models, achieving or surpassing human baseline performance. This research provides a reference framework for the efficient construction of specialized domain-specific large language models. Furthermore, the paper discusses the application of this model in fields such as the collation of ancient texts, dictionary editing, and language research, supplemented by case studies.
关键词
大语言模型 /
古汉语信息处理 /
人工智能
Key words
large language model /
information processing of ancient Chinese /
artificial intelligence
李 绅,胡韧奋,王立军.
古汉语大语言模型的构建及应用研究[J]. 语言战略研究. 2024, 9(5): 22-33 https://doi.org/10.19689/j.cnki.cn10-1361/h.20240502
Li Shen, Hu Renfen and Wang Lijun.
Construction and Application of Ancient Chinese Large Language Model[J]. Chinese Journal of Language Policy and Planning. 2024, 9(5): 22-33 https://doi.org/10.19689/j.cnki.cn10-1361/h.20240502
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
基金
国家语委重大项目“古籍整理智能化关键技术研究”(ZDA145-9)、国家社科基金重大招标项目“基于历代训释资源库的中国阐释学理论体系构建与实践”(22&ZD257)。