Information retrieval using a macedonian test collection for question answering

Abstract

Question answering systems solve many of the problems that users encounter when searching for focused information on the web and elsewhere. However, these systems cannot always adequately understand the user’s question posed in a natural language, primarily because any particular language has its own specifics that have to be taken into account in the search process. When designing a system for answering questions posed in a natural language, there is a need of creating an appropriate test collection that will be used for testing the system’s performance, as well as using an information retrieval method that will effectively answer questions for that collection. In this paper, we present a test collection we developed for answering questions in Macedonian language. We use this collection to test the performance of the vector space model with pivoted document length normalization. Preliminary experimental results show that our test collection can be effectively used to answer multiple-choice questions in Macedonian language.