Event

Unsupervised Learning of Language Models for MT Speaker

Timo Honkela

Helsinki Univerisity of Technology

Tuesday, November 18, 2008
12:30

Development of high quality machine translation has approved to be very challenging task. At best, the translation results can be reasonably good, for example, when the domain is small. Another important challenging factor is that the development of a machine translation system requires a lot of human effort. This is true especially in the case of systems that are based on human-specified rules for morphology, syntax, lexical selection, semantic analysis and generation. However, MT research has progressed in recent years thanks to statistical machine learning methods, sufficient computational resources, open source tools and increasing availability of bilingual parallel text resources. In this presentation, the focus is on unsupervised learning methods that can be used in creating language models for MT. The main emphasis is on how to deal with written language but also an experiment on speech-to-speech translation is described.