Implementation for Illustrative Sentences for English Multiword Expressions

Abstract

Recognizing multiword expressions is the fundamental task of the field. It could be potentially successfully approached in a supervised way, i.e. using a manually annotated training corpus to learn the characteristics (features) of multiword expressions as far as their structure and contextual environment is concerned. In succession, this knowledge would be used so as to locate multiword expressions that occur in another annotated text. The symbolic and statistical methods has been apparent in natural language processing (NLP) for some time. Multiword expressions are a key problem for the development of large-scale, linguistically sound natural language processing technology. We propose a method to search for illustrative sentences for English multiword expressions (MWEs) from a research paper database. We focus on syntactically flexible expressions such as “a lot of-work.” Traditionally, illustrative sentences that contain such expressions have been searched for by limiting the maximum number of words between the component words of the MWE. However, this method could not collect enough illustrative sentences in which clauses are inserted between component words of MWEs. We therefore devised a measure that calculates the distance between component words of an MWE in a parse tree, and use it for flexible expression search. We conducted experiments.