A rapid method to extract multiword expressions with statistic measures and linguistic rules

Abstract

Multiword Expressions (MWEs) have been the bottleneck in NLP. Particularly, the resource of fixed MWEs can improve the performance of tasks and implications of NLP. Due to complex characters of MWEs, it is hard to make difference between fixed MWEs and unfixed MWEs. This paper puts forwards an approach to extract fixed MWEs rapidly. First the definition of fixed MWEs is given. Features contributing to determinate fixed MWEs are considered both in statistic measures and in linguistic information. We extract fixed MWEs in the frame of multi-features and do manual evaluation. Experiment shows that the approach is effective. Our job can provide a desired list of fixed MWEs for NLP implication. © 2011 Springer-Verlag.