Couple of approaches that I can think of from papers of which combining in some way would probably mean decent results. Matching left and right fragments of n l

Couple of approaches that I can think of from papers of which combining in some way would probably mean decent results. Matching left and right fragments of n length, adding POS tagging to the matching fragments and using wordnet relation closeness for terms that exist in wordnet.