Corpora: Idiomaticity in NLP (query)

[Ivo Sanchez]

Dear list members:

I am looking for references on quantitative results about the differences
between employing pure distributional approaches to NLP applications and
analyses that take into account any dimension of idiomaticity (mostly
co-lexicalization). So for example I'd like to find some paper that compares
the performance of an Automatic Translation system that translates
word-by-word "make a difference" into "hace una diferencia"   vs. a system
that translates words according to their collocates "make a difference" into
"HAY una differencia" or a syntactic engine that analyzes strings such as in
front of as a PP instead of as simply a P.
Any other thoughts on this issue will also be appreciated.

Thank you in advance,
Ivo Sanchez