Evidence from the contemporary translation services market and many centuries of translation practice demonstrate that translation into a non-native language (L2 translation) can be performed effectively, despite the once-strong resistance to it on the grounds of it being perceived as unprofessional and inherently deficient. L2 translation is in fact unavoidable in the case of so-called languages of low diffusion, the command of which happens to be rather limited among native speakers of major languages. However, although the academic dispute about the validity of L2 translation seems decidedly milder now, there is still a lacuna within L2 translator training that needs to be addressed. This paper indicates that what usually betrays an L2 translation is its phraseological profile, often recognised as unnatural by native speakers of the target language. The aim of this paper is to propose a corpus-based data mining technique that may help L2 legal translator trainees become more observant with regards to phraseological patterning of foreign legal discourse, and more self-confident in taking well-informed translation decisions.