基于预训练语言模型的汉语字形谜解谜模型

A Chinese Character Riddle Solving Model Based on Pre-trained Language Models

摘要: 该文提出了一种基于预训练语言模型的汉语字形谜自动解答系统。汉语字形谜是一类利用汉字结构和形态构造的谜语。为了实现字形谜的自动解答,该文整合了结构预测、关键词提取和谜底映射等多个模块。通过构建综合性的汉字数据集,并利用BGE、BART等Transformer架构的预训练语言模型进行训练和评估,验证了该方法在汉字谜语解答任务中的有效性。实验结果表明,该系统在解析汉字谜语和生成准确谜底方面达到了96.90%的准确率,有效提高了谜语解答的自动化水平。该文的研究成果有助于推动汉字处理技术的发展,并为智能化语言教学和文化传承应用提供了新的技术支持和思路。

Abstract: This paper presents an automatic solution system for Chinese character riddles based on pre-trained language models. Chinese character riddles leverage the structure and morphology of Chinese characters. Accordingly, the proposed system integrates multiple modules, including structural prediction, keyword extraction and answer mapping. The pre-trained language models such as BGE and BART are employed and a comprehensive Chinese character dataset is constructed. Experimental results reveal an accuracy rate of 96.90% in parsing Chinese character riddles and generating accurate answers.