jMWE: A Java Toolkit for Detecting Multi-Word Expressions
Nidhi Kulkarni, Mark Alan Finlayson · 2011
jMWE is a Java library for implementing and testing algorithms that detect Multi-Word Expression (MWE) tokens in text. It provides (1) a detector API, including implementations of several detectors, (2) facilities for constructing indices of MWE types that may be used by the detectors, and (3) a testing framework for measuring the performance of a MWE detector. The software is available for free download. jMWE is a Java library for constructing and testing Multi-Word Expression (MWE) token detectors. The original goal of the library was to detect tokens (instances) of MWE types in a token stream, given a list of types such as those that can be extracted from an electronic dictionary such as WordNet (Fellbaum, 1998). The purpose of the library is not to discover new MWE types, but rather find instances of a set of given types in a given text. The library also supports MWE detectors that are not list-based. The functionality of the library is basic, but it is a necessary foundation for any system that wishes to use MWEs in later stages of language processing. It is a natural complement to software for discovering MWE types, such as mwetoolkit (Ramisch et al., 2010) or the NSP package (Banerjee and Pedersen, 2003). jMWE is available online for free download