This paper describes our project to support translation of streaming texts on social networks, in particular Twitter. Since machine translation of this type of content is still almost unusable, we rely on volunteers to provide and score the translations. The translations will serve as a testbed and development data for our MT systems tuned for this domain. The project thus serves multiple purposes: From the users’ point of view, we would like to provide a smooth access to timely information in foreign languages. From our translators’ point of view, we want to provide them with interesting content and material to improve their language skills. Finally, we admit that our project is still primarily a research exercise: As MT researchers, we are interested in learning to handle the specific challenges that this type of content brings. We hope to acquire an interesting collection of data for MT development and to gradually improve our MT processing pipeline for this type of text.
展开▼