How Machine Translation Works. A whistle-stop tour from the past to the present and future

Over the last few years machine translation has been making tremendous progress, so much so that for certain language pairs and genres human parity (and even super-human performance) has been claimed (and contested). In this talk van Genabith retraces the story of machine translation from rule-based, statistical to neural approaches (RBMT, SMT and NMT). The objective is to present the main ideas underpinning the approaches in accessible terms and, following this, provide a glimpse of what is currently in the labs.

近年来，机器翻译（machine translation）领域取得了长足进步，其发展成效之显著，使得部分语言对及特定文本类型下的机器翻译系统已被宣称达到人类等效水平（甚至实现超人类表现），不过这一论断也存在诸多争议。在本次演讲中，范·吉纳比思（van Genabith）将回溯机器翻译从基于规则、统计到神经的技术演进历程，涵盖基于规则的机器翻译（Rule-based Machine Translation, RBMT）、统计机器翻译（Statistical Machine Translation, SMT）以及神经机器翻译（Neural Machine Translation, NMT）三大技术路径。本次演讲旨在以通俗易懂的语言阐释各技术路径背后的核心思想，并在此基础上，简要展现当前实验室中正在探索的前沿研究方向。