The Application of Fuzzy Logic to Collocation Extraction

TL;DR

This paper introduces a fuzzy logic approach to collocation extraction in NLP, allowing for degrees of membership rather than binary classification, and demonstrates improved results over traditional statistical methods.

Contribution

It presents a novel fuzzy logic-based method for collocation extraction that combines existing statistical measures to overcome their limitations.

Findings

Fuzzy logic provides a nuanced measure of collocation strength.

The method outperforms mutual information and t-test in accuracy.

Demonstrated on a large corpus of literary texts.

Abstract

Collocations are important for many tasks of Natural language processing such as information retrieval, machine translation, computational lexicography etc. So far many statistical methods have been used for collocation extraction. Almost all the methods form a classical crisp set of collocation. We propose a fuzzy logic approach of collocation extraction to form a fuzzy set of collocations in which each word combination has a certain grade of membership for being collocation. Fuzzy logic provides an easy way to express natural language into fuzzy logic rules. Two existing methods; Mutual information and t-test have been utilized for the input of the fuzzy inference system. The resulting membership function could be easily seen and demonstrated. To show the utility of the fuzzy logic some word pairs have been examined as an example. The working data has been based on a corpus of about one million words contained in different novels constituting project Gutenberg available on www.gutenberg.org. The proposed method has all the advantages of the two methods, while overcoming their drawbacks. Hence it provides a better result than the two methods.