On Wednesday October 14, Maryam will give a talk on “Source Language Word Embedding in Neural Machine Translation” in the lab meeting. Here is the abstract of the presentation: “Neural Machine Translation (NMT) is a new approach to machine translation that has shown promising results comparable to conventional Statistical Machine Translation (SMT) approaches in some cases. Unlike the conventional SMT, NMT systems aim to build a single neural network that can be jointly tuned to maximize the translation accuracy. A major drawback in NMT systems is the limitation on source and target vocabularies and therefore inability in handling large vocabularies. Recently some approaches have been proposed to improve the coverage of target vocabulary which directly affects the translation quality. In this work we focus on the source side vocabulary. We leverage monolingual data on the source side to improve the performance of NMT systems. We show that improving the word embedding on the source side effectively increases the translation quality without affecting the complexity of translation system.”
