Practical large-vocabulary speech recognition in a multilingual environment

Abstract This paper is concerned with the use of a commercial large-vocabulary speech recognition system by a team of mainstream users in the course of their everyday work. In particular, it describes use by translators working in four languages in a multilingual environment at the European Commission. The paper begins by describing some of the differences between the point-of-view of a typical researcher in speech recognition and that of a typical mainstream user. It points out some of the psychological barriers that must be overcome if speech recognition is to gain really widespread acceptance, and it concludes that such acceptance will depend at least as much on the sharing of experience between users as on technical advances. The overall results of the trials at the European Commission were encouragingly positive, but several unexpected problems were encountered, many of them related to the multilingual environment. The paper describes how most of these problems are being addressed.