Translation of verbal idioms

Verbal idioms constitute a challenge for machine translation systems: their meaning is not compositional, preventing a wordfor-word translation, and they can be discontinuous, preventing a match during tokenization. This paper presents the treatment of verbal idioms in our machine translation system, which addresses both challenges by deferring idiom matching until after the parse, and by allowing all parts of an idiom to be treated as a lexical unit, making alignment and transfer easier. Identification of idiom use is not a problem for us, because we learn idiom translation in context.