神经网络机器翻译:进展与挑战

Neural machine translation: Advancements and challenges

随着全球化进程的加速和国际交流的日益密切,传统人工翻译方式已无法满足迅猛增长的翻译需求。机器翻译以方便快捷的优势逐渐深入人们的生活,迎来了新的发展阶段。作为机器翻译的最新范式,神经网络机器翻译极大地提升了翻译的质量,有专家声称机器翻译已经能够达到“接近人工译文”或“等同于人工译文”的效果,从而引发了学界关于机器翻译是否要取代人工翻译的探讨。神经网络机器翻译究竟在哪些方面取得了进步?还面临着哪些亟待解决的问题与挑战?又该从哪些方面提升其翻译质量?围绕这些问题,本文简要回顾神经网络机器翻译的质量提升成果,分析其面临的问题与挑战,多层面探讨其质量提升路径,以期为神经网络翻译系统的研发提供参考借鉴。

With the rapid progress of globalization and increasing international exchanges, traditional ways of translating cannot satisfy the growing demand of translation activities anymore. Meanwhile, machine translation gradually integrates into people’s life with the advantage of convenience and speed, and ushers in a new stage of development. As a new paradigm of machine translation, neural machine translation has achieved state-of-the-art results. Some experts even claim that it has achieved “near-human parity” or “ human parity”, and it triggered a discussion in the academic field on whether machine translation will replace human translation. What progress has been made in neural network machine translation? What are the problems and challenges that need to be solved? And in what ways should the quality of neural machine translation be improved? This paper briefly reviews the advancements of the neural machine translation, analyzes the existing problems and challenges, and discusses multiple ways of improving neural machine translation performance,so as to shed new light on the development of neural machine translation systems.