Natural Language Understanding for Simultaneous Conference Interpretation

Abstract

Conference interpretation is an active area of linguistics with growing challenges in technological integration. Despite advancement in information technology, simultaneous interpreters have not yet been provided with adequate tools to bring down the stress level that accompanies their profession. The booth setting and the way they perform have not been changed a lot over the years. Although a number of computer approaches have been presented to make the task of conference interpreters less challenging, most of them fail to meet their actual needs. Some of those approaches add to the pressure that interpreters are already under as they require human input, while others are restricted to certain languages. This paper proposes a new approach that makes use of automatic speech recognition (ASR) combined with a cloud-based machine translation (MT) that transcribes spoken words and provides in-depth translation in a contextual manner through the use of a compiled glossary. The proposed approach provides for the first time an instantaneous transcription of a speech, a domain detection through a part-of-speech tagger, and an adequate translation of the terminology used. Our approach has been tested in terms of transcription accuracy, domain extraction, and terminology identification and retrieval using English and Arabic speeches that cover different domains.