The Hunglish bitext query system

Modern methods of machine translation and automated dictionary- and thesaurus building take parallel texts as their input, which contain the same material in two languages. Sentence alignment is a task that is key to preparing the data for machine translation and related tasks. We developed the Hunglish bitext query system. The system consists of two components. One of them is for searching in the entire database, while the other takes a pair of documents as input, extracts a set of parallel sentence pairs from them to expand the database. Users can also vote on the quality of sentence pairs, thus improving the parallel corpus. The service is currently provided for Hungarian-English, but the software is entirely language(pair)-independent.