Theme:

The Consortium for Speech Translation Advanced Research (C-STAR) projects has attempted to solve issues in speech translation. Speech translation systems are designed as a system combined machine translation (MT) system with automatic speech recognition (ASR) system. Researches on text translation with large vocabulary such as newspaper text and HTML documents by using statistical approaches are being actively investigated. Text translation is still difficult and speech translation is more complicated than text translation. The difficulty in speech translation in terms of verbal information is caused by processing spoken language and ASR output. To accomplish speech translation, we have to solve problems in MT for spoken language and handle ASR output. In this workshop, we will focus on handling spontaneous aspects in spoken language and speech recognition output including multiple recognition hypotheses.

Speech recognition results are not always perfect. Speech translation performance is degraded by recognition errors. When we translate speech so that maintains original information in source speech as much as possible by handling recognition errors, we have concrete questions that we hope to see addressed include, but are not limited to the following issues.

- How can ASR output be translated more accurately even if recognition errors exist?

- How much MT performance could be enhanced by considering multiple hypotheses?

- Which hypothesis can contribute for MT performance?

- How to select best hypothesis in recognition results which can be translated more accurately?