Identifying phrasemes via interlingual association measures - A data-driven approach on dependency-parsed and word-aligned parallel corpora

Abstract

In corpus linguistics, statistical association measures play a major role in identifying collocations such as ‘play’ and ‘role’ in ‘play a role’. Those two words that appear considerably more often in the same context than one would expect from a random distribution are collocates. They typically constitute meaning beyond the bare combination of both words’ semantics.
We employ the same association measures on interlingual word co-occurrences based on statistical word alignment and combine them with intralingual association measures on syntactical dependency relations in order to identify phrasemes. Support verb constructions exemplify our approach. They are characterized by the respective verb contributing little to the semantics of the whole construction, which we can determine with the aid of our intralingual association measures.