We only have DE->EN and EN->DE currently.
Normally we would need matching corpora, but investigate Unsupervised Machine Translation Using Monolingual Corpora Only.
Since we have an international team, it might be best to include some of our multilingual speakers in this. I sadly only really know English! (and a little Spanish vocab)