Corpus-based evidence for approximating semantic transparency of complex verbs

lexical concepts. Furthermore, more specific sets of the modifier relation (MO) and its subclasses may model the verbs in a more meaningful way. At last, future studies will need to address the problem of polysemy of the verbs both in corpus-based evidence and human association scores. To summarize, using dependency parses allows us to exploit verbs with separated verb particles. This enlarges the empirical basis and boosts the frequency counts for verbs and their dependents that are lost in other approaches. Our results suggest that the meaning similarity between verbs can be well modeled by the distributional similarity based on modifiers in addition to that of arguments.