A Spoken Question Answering System Based on Conditional Knowledge NICOLAE T ¸ ˘ AND ˘ AREANU University of Craiova Department of Computer Science A.I.Cuza Street 13, 200585 Craiova ROMANIA ntand@rdslink.ro MIHAELA COLHON University of Craiova Department of Computer Science A.I.Cuza Street 13, 200585 Craiova ROMANIA mcolhon@inf.ucv.ro CRISTINA ZAMFIR University of Craiova Department of Computer Science A.I.Cuza Street 13, 200585 Craiova ROMANIA cristina.zamfir@star-storage.ro Abstract: A conditional schema is a graph-based structure which is able to represent conditional knowledge. This structure was introduced in [11]. The inference mechanism corresponding to the conditional schema representa- tions was developed in [12]. In this paper we propose a question answering system that can represent and process conditional knowledge using these mechanisms. The structure of such a system is presented and an implementa- tion by means of Java platform is briefly described. Key–Words: conditional knowledge, question answering system, Java platform 1 Introduction A Spoken Dialogue System (SDS) is a software sys- tem that accepts natural language as input and pro- duces natural language as output engaging in a con- versation with an user. To successfully manage the in- teraction with users, SDS usually carry out five main tasks: Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), Dialogue Manage- ment (DM), Natural Language Generation (NLG) and Text-to Speech Synthesis (TSS). These tasks are usu- ally implemented in different modules. In general by a Question Answering System we understand a SDS designed to provide answers to questions that are for- mulated by user in natural language. To find the an- swer to a question, a question answering system may use a database, a collection of natural language doc- uments or a knowledge base. Various such systems were developed. The system START ([10]) extracts the information contained by an English text and ob- tains a knowledge base. The user has access to infor- mation by querying the knowledge base. The system formulates the answer in English. The architecture of the AskMSR question answering system is presented in [4], where the strategies for predicting when the question answering system is likely to give an incor- rect answer are also explored. Snowball ([1]) in- troduces novel strategies for generating patterns and extracting tuples from plain-text documents. This sys- tem was developed for extracting structured data from plain-text documents with minimal human participa- tion. AnswerBus ([13]) is an open-domain ques- tion answering system based on sentence level Web information retrieval. From the Web pages, Answer- Bus extracts sentences that are determined to contain answers. A mixture of Natural Language Process- ing and Information Retrieval can be encountered in Quanda system ([2], [3]). MAYA ([9]) is a question answering system in Korean that uses a predictive an- swer indexer. A model for answer extraction compo- nent of a question answering system called Sbuqa is presented in [14]. The Lexical Functional Grammar, a meaning based grammar that analyses sentences in a deeper level than syntactic parsing are used to repre- sent the question and candidate answers. The papers [6], [7] and [8] belong to a sequence of papers for Ro- manian case. The reader interested to use the question answering systems for knowledge based on the Web can find an interesting approach in [5]. In this paper we describe an architecture and a Java implementation of a spoken question answering system based on conditional knowledge. The main features of this system are the following ones: • The user-system communication is based on a voice user interface. • The information is represented into a knowledge base that uses conditional knowledge. • The implementation is platform independent be- cause Java technology is used. • The system can be extended to obtain dialogue systems based on conditional knowledge; • The product can be used successfully in automatic training, to build case-based consultancies or sys- tems for diagnosing the malfunctions of a device. LATEST TRENDS on COMPUTERS (Volume I) ISSN: 1792-4251 220 ISBN: 978-960-474-201-1