The UFAL Medical Corpus is a collection of parallel (sentence-aligned bilingual) corpora for the medical domain, built specifically for machine translation of medical texts. It also served as training data for the WMT17 Biomedical Task, so it is a resource with a solid research track record behind it.

It was assembled by ÚFAL, the Institute of Formal and Applied Linguistics at the Faculty of Mathematics and Physics, Charles University (Prague, Czechia), during the EU-funded projects Khresmoi, KConnect, and HimL (“Health in my Language”), with support from EU Horizon 2020.

Eight language pairs are available: English-Czech, English-German, English-Spanish, English-French, English-Hungarian, English-Polish, English-Romanian, and English-Swedish.

This is a compilation of pre-existing parallel corpora that combine medical-domain and general-domain sources, including EU and international institutional documents (CESTA, EMEA / European Medicines Agency, ECDC, EUROPARL, JRC-Acquis, EU Bookshop, MultiUN, Cordis, Rapid), patent translations (PatTR), medical abstracts (MuchMore), subtitles (OpenSubtitles), a few medical web crawls, and bilingual dictionaries (DBpedia, Linguee).

The corpus (v. 1.0) was compiled between 2015 and 2017 from these existing sources.

How to use corpora in translation or writing:

• Check frequencies, usage, and collocations for terms and phrases
• Connect the corpus to your translation system
• Integrate the corpus with a machine translation engine for better quality

The corpus is free but requires registration. You fill in a short form (name, email, institution, country) and are sent a username to download the archive, which is about 25 GB. So this is a large dataset to work with locally, not an online search interface. The use is restricted to research and non-commercial purposes.

Click “Open resource” to go to the UFAL Medical Corpus page, register, and download the parallel data.