Seminar from Frédéric Blain, lecturer at University of Wolverhampton
Localization: IC2, Boardroom and online
Speaker: Frédéric Blain
Quality Estimation: Application to Machine Translation.
What is a good translation? This is a simple question, yet quite a difficult one to answer, in particular when the translation has been generated automatically.
Machine Translation (MT) has been a very active field of research for several decades now, yet the question of how “good” state-of-the-art MT systems are still divides the Research Community. MT is used daily, almost everywhere by almost everyone, ranging from professionals to average end-users. However, in most use cases (e.g. online platforms) clues as to whether and to which extent these automatically generated texts are reliable, are non-existent.
This talk aims to introduce the concept of Quality Estimation and its application to Machine Translation. In other words, how to predict the “quality” of MT output when no reference is available (i.e. Evaluation cannot be done). State-of-the-art architectures will be presented and exploratory research directions will be discussed. Time permitting, the talk would finish with some pointers to open-source resources one can use to get started.