How to build a QA system in your back-garden: application for Romanian

How to build a QA system in your back-garden: application for Romanian Constantin Or5san Computational Linguistics Group University of Wolverhampton C.Orasan@w1v.ac.uk Doina Tatar, Gabriela erban, Dana Lupsa and Adrian Onet Faculty of Mathematics and Computer Science Babe§-Bolyai University fdtatar, gabis, davram, adrianl@cs.ubbcluj.ro Abstract Even though the question answering (QA) field appeared only in recent years, there are systems for English which obtain good results for open- domain questions. The situation is very different for other languages, mainly due to the lack of NLP resources which are normally used by QA systems. In this paper, we present a project which develops a QA system for Romanian. The challenges we face and decisions we have to make are discussed. 1 Introduction Question answering (QA) emerged in the late 90s as a result of the Text Retrieval Conferences (TREC). These conferences are designed to evaluate the state-of-the-art in text retrieval and allow the participants to evaluate their systems in a consistent way by providing them a common test set. Starting with TREC-8, in 1999, these conferences contain a question answering track in which the participants try to find the answer to questions in a large collection of texts. As a result of the TREC, the QA field witnessed rapid development for English, but there are only few systems which work for languages other than English (Kim and Seo, 2002; Vetulani, 2002). Another factor which slows down the development of QA systems for other languages than English is the lack of the modules which are normally used in a QA system. In this paper, we discuss the challenges we have to face during the development of a question answering system for Romanian language. This paper is structured as follows: In Section 2, we present the structure of our QA system. The problems which need to be tackled when it is implemented for Romanian are presented in Section 3. A discussion of the project is presented in Section 4, the article finishing with conclusions. 2 The structure of our question answering system A QA system normally contains three modules: a question processor, a document processor and an answer extractor module (Harabagiu and Moldovan, 2003). In addition, QA systems also rely on a generic or specially designed search engine. Our system follows this structure. The question processor transforms a natural language question in an internal representation which can be a list of keywords or some kind of logical form. The list of keywords can contain only words from the question, or it can be expanded with words related to the ones in the question. Even if the question is transformed to a more advanced representation than a simple list of keywords, given that the QA systems rely on search engines, it is necessary to produce a list of keywords which are used to query the search engine. The more advanced form is used by the answer extraction module to locate the answer. At present, our system extracts only keywords from the question. These keywords are expanded …

Content truncated to 500 words