Speech-Translation: From Domain-Limited to Domain-Unlimited Translation Tasks

Stephan Vogel

Carnegie Mellon University, Pittsburgh, USA

Speech translation research started more than 15 years ago.
Early projects and demonstrations, like Verbmobil and work
done by the C-Star consortium where restricted to domain-limeted
applications, esp. travel domain.  We see continued interest
in this kind of projects, and also a strong tendency to commercialize
speech translation technology.  In recent years the goal have
become more challenging:  domain-unlimited speech translation.
Projects like TC-Star in Europe and GALE in the United States
aim at translating parliamentary speeches, broadcast news
and even broadcast conversations.  Ongoing research focuses
on questions like tighter coupling between speech recognition
and translation, segmenting speech recognition output into
sentence-like units, and how this affects translation quality,
detecting and removing disfluencies, and end-to-end optimization.
In this paper we will review some of recent work done in both
domain-limited and domain-unlimited speech translation.  We
will show where progress has been made and highlight areas,
where initial expectations have not been met so far.