预训练语言模型在日文文本难易度自动分类中的应用

Application of Pre-Trained Language Models in Automatic Difficulty Classification of Japanese Texts

摘要: 在日语教学过程中，选择难易度合适的日文文本作为教学材料有利于提高日语学习者的学习兴趣及效率。日语具有词汇量大、语法复杂等特征，对文本难易度分类提出了挑战。本文尝试采用多种基于神经网络的日语预训练语言模型，通过收集历年日本语能力测试真题以及模拟题作为数据集以训练日文文本难易度自动分类模型。实验结果表明，预训练语言模型在日文文本难易度自动分类任务上能够表现出较好的性能。基于预训练语言模型的日文文本难易度自动分类方法将为计算机辅助日语学习系统以及电子化教材开发等提供有力的技术保障。

Abstract: In the process of Japanese teaching and learning, it is important to select Japanese texts with appropriate difficulty levels as educational materials. This can help to enhance Japanese language learners' interest and efficiency. Japanese language is characterized by its large vocabulary and complex grammar, which makes it challenging to automatically classify the difficulty levels of Japanese texts. This paper attempts to apply several neural network-based Japanese pre-trained language models in training automatic difficulty classification models of Japanese texts, by collecting historical Japanese Language Proficiency Test questions and simulated questions as datasets. Experimental results show that pre-trained language models have good performance in automatically classifying the difficulty levels of Japanese texts. The pre-trained language models-based approach for automatic difficulty classification of Japanese texts will offer significant technical support for the development of computer-assisted Japanese learning systems and electronic textbooks.