Data Mining Meets Collocations Discovery

In this paper we discuss the problem of discovering interesting word sequences in the light of two traditions: sequential pattern mining (from data mining) and collocations discovery (from computational linguistics). Smadja (1993) defines a collocation as “a recurrent combination of words that cooccur more often than chance and that correspond to arbitrary word usages.” The notion of arbitrariness underlines the fact that if one word of a collocation is substituted by a synonym, the resulting phrase may become peculiar or even incorrect. For instance, “strong tea” cannot be replaced with “powerful tea”. Acquisition of collocations, a.k.a multi-word units, are crucial for many fields, like lexicography, machine translation, foreign language learning, and information retrieval. We attempt to describe the collocations discovery problem as a general problem of discovering interesting sequences in text. Moreover, we give a survey of common approaches from both collocations discovery and data mining and propose new avenues for fruitful combination of these approaches.