基于语义信息共享Transformer的古文机器翻译方法  

Machine Translation of Ancient Chinese Text Based on Transformer of Semantic Information Sharing

作　　者：周成彬 刘忠宝[1,2,3] ZHOU Chengbin;LIU Zhongbao(School of Software,North University of China,Taiyuan 030051,China;Institute of Language Intelligence,Beijing Language and Culture University,Beijing 100083,China;School of Software,Quanzhou University of Information Engineering,Quanzhou,Fujian 362000,China)

机构地区：[1]中北大学软件学院,太原030051 [2]北京语言大学语言智能研究院,北京100083 [3]泉州信息工程学院软件学院,泉州362000

出　　处：《情报工程》2022年第6期114-127,共14页Technology Intelligence Engineering

基　　金：教育部哲学社会科学研究后期资助项目“大数据环境下数字人文理论、方法与应用研究”(21JHQ081);福建省社会科学基金项目“大数据驱动的古籍故事化表达与情景化再现研究”(FJ2022A018)。

摘　　要：[目的/意义]中国古籍浩如烟海,承载了古人的精神和智慧,对古文进行翻译有利于继承和发扬中华传统文化。随着人工智能技术的发展,利用计算机实现古文的自动翻译具有重要意义。[方法/过程]然而,目前对古文进行机器翻译的研究还比较少。因此,本文根据古文与现代文属于同一语言的特点,通过共享词表与嵌入层参数的方法,提出了基于语义信息共享的Transformer模型实现古文到现代文的自动翻译,使用BLEU作为评价指标。[结果/结论]实验表明,该模型的BLEU值达到了31.43,相比传统的基于GRU和LSTM的seq2seq模型分别提高了26.94和15.19个BLEU值,比基准Transformer模型提高了13.41个BLEU值,证明了模型的有效性。[Objective/Significance] Chinese ancient books are as vast as a vast ocean, carrying the spirit and wisdom of the ancients, and translating ancient texts is conducive to inheriting and carrying forward traditional Chinese culture. With the development of artificial intelligence technology, it is important to use computers to realize automatic translation of ancient texts. [Methods/Processes] However, there are still relatively few studies on machine translation of these ancient texts. Therefore, according to the characteristics that ancient texts and modern texts belong to the same language, this paper proposes a Transformer model based on semantic information sharing to realize automatic translation from ancient texts to modern texts by sharing vocabulary and embedding layer parameters, using BLEU as an evaluation index. [Results/Conclusions] Experiments show that the model achieves a BLEU value of 31.43, which is 26.94 and 15.19 BLEU values higher than the traditional GRUbased and LSTM-based seq2seq models, respectively, and 13.41 BLEU values higher than the benchmark Transformer model, so the model is effective.

关 键 词：机器翻译 Transformer模型 BLEU 古文翻译 

分 类 号：G35[文化科学—情报学] TP391[自动化与计算机技术—计算机应用技术]