“Terminator” Integrating Terminology in Neural Machine Translation
For years, human translators have leveraged terminology in ensuring translation consistency and reducing ambiguity. Terminology is arguably the most important external translational resource in automated translation workflows (TW). Term translation is a well-known problem in machine translation (MT) research. The term translation becomes a more challenging task for MT systems while encountering morphologically rich and complex languages. To the best of my knowledge, a viable solution to integrate terminology into MT is as yet unavailable, either in academic or industry. A suitable solution to integrate terminology into MT would certainly be a breakthrough in MT research, and a blessing for the translation industry. The conventional phrase-based statistical MT decoder Moses enables a solution to term translation with its XML-markup approach. The term translation problem has reached new heights in NMT due to its architecture. In neural MT (NMT) research, the terminology integration is still a less examined area and regarded as an unsolved problem.In this application, I identify research questions, set my research goals, and propose viable solutions with the intent of terminology integration in NMT in industrial scenarios.