基于内容词融合的神经机器翻译方法  

A content word fusion method for neural machine translation

作　　者：赵忠超 唐忠 谢京天 李付学[2] 闫红[2] ZHAO Zhongchao;TANG Zhong;XIE Jingtian;LI Fuxue;YAN Hong(Institute for Computer Sciences and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;College of Electrical Engineering,Yingkou Institute of Technology,Yingkou 115014,Liaoning,China)

机构地区：[1]沈阳化工大学计算机科学与技术学院,沈阳110142 [2]营口理工学院电气工程学院,辽宁营口115014

出　　处：《智能计算机与应用》2024年第12期157-162,共6页Intelligent Computer and Applications

基　　金：辽宁省自然科学基金(2021-YKLH-12,2022-YKLH-18)。

摘　　要：基于Transformer的神经机器翻译模型是目前主流的机器翻译范式,达到了最先进的性能水平。然而,这种模型无法捕捉单词在句子语义中的重要性,如内容词(实词)比功能词(虚词)表达的更重要,进而会导致翻译错误或歧义。为了解决这个问题,提出了一种基于内容词融合的方法来改进模型。首先,根据内容词识别算法将句子中的单词分为内容词和功能词;然后,利用不同的融合策略将源语言句子中的内容词嵌入到模型中,指导模型翻译过程。在多个翻译任务上的实验证明,基于内容词融合的方法优于基线模型,提升了模型的翻译性能。The Transformer-based neural machine translation model is currently the prevailing paradigm in the research field,achieving state-of-the-art performance.However,these models fails to capture the importance of words in the semantic context of a sentence,such as content words being more crucial than function words.This limitation can lead to translation errors or ambiguities.To alleviate this issue,the paper proposes an approach called content-word fusion to enhance the model's performance.Firstly,the paper utilizes a content word identification algorithm to categorize words in a sentence into content words and function words.Next,the paper employs various fusion strategies to incorporate the content words from the source language sentence into the model,thereby providing guidance for the translation process.Experimental results from multiple translation tasks demonstrate that the content-word fusion method outperforms the baseline models,significantly improving translation performance.

关 键 词：神经机器翻译 词嵌入 内容词 TRANSFORMER 融合策略 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]