Using NLP Methods to Improve the Effectiveness of a Macedonian Question Answering System

摘要

The process of retrieving particular information from a huge amount of text significantly depends on the language specific features. The most imposing one for Macedonian language is the possibility of a word to have various derivational and inflectional suffixes. In this research we investiage how particular NLP tools influence the retrieval, putting special emphasis on the use of Part-of-Speech (PoS) tagging, word forms, and stemming. In absence of a stemming algorithm for Macedonian language, we used the Dice Coefficient and the single-link clustering in order to group words with a common base form. All these features were implemented in an already existing Macedonian Question Answering System (QAS). We tested different strategies for weighting terms in the documents (the queries), as well as different approaches for query expansion with word forms and words with the same stem. The experimental results show that the word variations strongly influence the retrieval, improving our system's accuracy.