Towards High-Reliability Speech Translation in the Medical Domain

Graham Neubig, Sakriani Sakti, Tomoki Toda, Satoshi Nakamura, Yuji Matsumoto, Ryosuke Isotani, Yukichi Ikeda

Abstract: In this paper, we describe the overall design for a speech translation system that aims to reduce the problems caused by language barriers in medical situations. As first steps to building a system according to this design, we describe a collection of a medical corpus, and some translation experiments performed on this corpus. As a result of the experiments, we find that the best of three modern translation systems is able to translate 33%-81% of the sentences in a way such that the main content is understandable.