Domain-specific Statistical Machine Translation

Our partners in industry require translation systems for specific application scenarios but typically possess only very little traning data. Therefore, we investigate how the use of domain-specific traninig data in statistical machine translation (SMT) can be optimized.

We have a small parallel corpus (5 million tokens) of Alpine texts available: the periodicals of the Swiss Alpine Club (SAC) digitalized in the project Text+Berg digital.

Project head:

• Martin Volk

Researchers:

• Rico Sennrich
• Magdalena Plamada (née Jitca)

The project is funded by the Swiss National Science Foundation and is running since 2010.

More information:

• Domain-specific Statistical Machine Translation