基于Mem2Seq网络和多级记忆网络的翻译对话系统设计  

Design of a translation dialogue system based on Mem2Seq network and multi-level memory network

作　　者：陆春兰 LU Chunlan(Xi’an Urban Architectural College,Xi’an 710114,China)

机构地区：[1]西安城市建设职业学院,西安710114

出　　处：《自动化与仪器仪表》2025年第2期228-232,共5页Automation & Instrumentation

基　　金：校级《以就业为导向的高职计算机专业大学英语课程的教学实践研究》(JKY20230109)。

摘　　要：英语翻译近年来随着全球化的发展,起着越来越重要的沟通桥梁作用。但是现有的翻译对话系统数据库常常出现解码错误,导致翻译的不准确性。因此,研究提出结合端到端可微分网络和多级记忆网络模型来优化翻译对话系统。结果显示,研究所提模型在实际英语翻译时,准确度在K值为1、3时最为稳定。研究设计的模型可以最大程度上避免输入指令梯度消失,进而处理更复杂的指令与长文本。英语翻译的对话系统的智能化将会得到进一步提升,其也将实现自主注意力的表达能力,丰富单词表征,理解语句的更深层次含义。这为机翻行业提供了极具价值的实验科研方向,能够在一定程度上助推翻译行业经济发展,增加使用者的生活便捷度与满足感。In recent years,English translation has played an increasingly important role as a communication bridge with the de-velopment of globalization.However,existing translation dialogue system databases often suffer from decoding errors,leading to trans-lation inaccuracies.Therefore,the study proposes a combination of end-to-end differentiable networks and multi-level memory net-work models to optimize translation dialogue systems.The results show that the model proposed by the research institute has the most stable accuracy in actual English translation at K values of 1 and 3.The research designed model can minimize the disappearance of input instruction gradients,thereby processing more complex instructions and long texts.The intelligence of the dialogue system in English translation will be further improved,which will also enable autonomous attention expression ability,enrich word representa-tion,and understand deeper meanings of sentences.This provides a highly meaningful research direction for the machine translation industry,which can to some extent promote future economic development,enhance people’s convenience and satisfaction in life.

关 键 词：英语翻译 Mem2Seq 多级记忆网络 指针网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP39[自动化与计算机技术—控制科学与工程]