Neural Machine Translation

The Department of Computer Science Colloquium will be presented by Dr Alexandra Birch  of  University of Edinburgh, Scotland, with a talk entitled, "Neural Machine Translation".  

Recent interest in deep learning has lead to significant advances in the field of machine translation. Unlike the traditional statistical machine translation, neural machine translation builds a single neural network that can be trained to jointly align and translate. We present a number of experiments which address some of the limitations of standard NMT. Neural models typically allow for a limited vocabulary size and previous approaches use external dictionaries to address this. We introduce a simpler and more effective approach, making the NMT model capable of open-vocabulary translation by encoding rare and unknown words as sequences of subword units. We also present a method for incorporating the large amounts of monolingual data available in the NMT model by using back-translation. 

 Alexandra Birch is currently a senior research associate in Informatics at the University of Edinburgh. She completed her BSc and Honours degree at UCT and her PhD in Edinburgh. Alexandra has been working in the field of machine translation on many different sub-fields including reordering, evaluation, semantics, and spoken language translation. Her recent interests have focussed on neural machine translation where advances using sub-word units and monolingual data have beaten state-of-the-art baselines. She is currently the tutorial chair for ACL, she is helping to coordinate two EU translation projects HimL and TraMOOC, and she is the scientific project manager for the new EU project Scalable Understanding of Multilingual MediA.


Thu, 07 Apr 2016 - 13:00

Computer Science Lecture Theatre 302