<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2510">
  <Title>Ontological resources and question answering</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> This paper discusses the possibility of building an ontology-based question answering system in the context of the Semantic Web presenting a proof-of-concept system. The system is under development in the MOSES European Project.</Paragraph>
    <Paragraph position="1"> Introduction Question Answering (QA) systems (as QA track of the Text Retrieval Conference (TREC-QA) competitions (Voorhees 2001)), are able both to understand questions in natural language and to produce answers in the form of selected paragraphs extracted from very large collections of text. Generally, they are open-domain systems, and do not rely on specialised conceptual knowledge as they use a mixture of statistical techniques and shallow linguistic analysis. Ontological Question Answering systems, e.g. (Woods et al. 1972, Zajac 2000) propose to attack the problem by means of an internal unambiguous knowledge representation. As any knowledge intensive application, ontological QA systems have as intrinsic limitation related to the small scale of the underlying syntactic-semantic models of natural language.</Paragraph>
    <Paragraph position="2"> While limitations are well-known, we are still questioning if any improvement has occurred since the development of the first ontological QA system LUNAR. Several important facts have emerged that could influence related research approaches: a0 a growing availability of lexical knowledge bases that model and structure words: WordNet (Miller 1995) and EuroWordNet (Vossen 1998) among others; some open-domain QA systems have proven the usefulness of these resources, e.g. WordNet in the system described in (Harabagiu et al. 2001).</Paragraph>
    <Paragraph position="3"> a0 the vision of a Web populated by &amp;quot;ontologically&amp;quot; tagged documents which the semantic Web initiative has promoted; in case this vision becomes a reality, it will require a world-wide collaborative work for building interrelated &amp;quot;conceptualisations&amp;quot; of domain specific knowledge a0 the trend in building shallow, modular, and robust natural language processing systems (Abney 1996, Hobbs et al. 1996, Ait-Moktar&amp;Chanod 1997, Basili&amp;Zanzotto 2002) which is making them appealing in the context of ontological QA systems, both for text interpretation (Andreasen et al. 2002) and for database access (Popescu et al. 2003).</Paragraph>
    <Paragraph position="4"> Given this background, we are investigating a new approach to ontology-based QA in which users ask questions in natural language to knowledge bases of facts extracted from a federation of Web sites and organised in topic map repositories (Garshol 2003). Our approach is being investigated in the context of EU project MOSES1, with the explicit objective of developing an ontology-based methodology to search, create, maintain and adapt semantically structured Web contents according to the vision of the Semantic Web. MOSES is taking advantage of expertise coming from several fields: software agent technology, NLP, graph theory  sity of Roma Tor Vergata and ParaBotS.</Paragraph>
    <Paragraph position="5"> and text mining. The test-bed chosen in the project is related to the development of an ontology-based knowledge management system and an ontology-based search engine that will both accept questions and produce answers in natural language for the Web sites of two European universities. The challenges of the project are: a0 building an ontological QA system; a0 developing a multilingual environment which implies the ability to treat several languages, and, importantly, several conceptualisations.</Paragraph>
    <Paragraph position="6"> In this paper, after briefly describing how the project is trying to comply with the semantic Web vision, we will focus on question processing, and in particular on the way in which NLP techniques and ontological knowledge interact in order to support questions to specific sites or to site federations.</Paragraph>
    <Paragraph position="7"> An ontology-based approach to question answering In our ontological QA system, both questions and domain knowledge are represented by the same ontological language. It is foreseen to develop the QA system in two steps. First a prototypical implementation is planned to answer questions related to the current &amp;quot;state-of-affairs&amp;quot; of the site to which the question is posed. In a second step, given a &amp;quot;federation&amp;quot; of sites within the same domain, we will investigate whether and how an ontological approach could support QA across the sites. Answering a question can then be seen as a collaborative task between ontological nodes belonging to the same QA system. Since each node has its own version of the domain ontology, the task of passing a question from node to node may be reduced to a mapping task between (similar) conceptual representations. To make such an approach feasible, a number of difficult problems must still be solved. In this paper, we will provide details on how: a0 to build on existing ontologies and interface between them and language resources; a0 to interpret questions wrt the ontological language; a0 to model the mapping task for federated questions.</Paragraph>
    <Paragraph position="8"> Building on off-the-shelf semantic Web ontologies null One of the results of the Semantic Web initiative will be the production of many interrelated domain-specific ontologies that provide the formal language for describing the content of Web documents. In spite of the freedom allowed in the production of new conceptualisations, it is reasonable to expect that a first knowledge representation jungle will leave room to a more orderly place where only the more appreciated conceptualisations have survived. This is a prerequisite for achieving interoperability among software agents. In view of this, and since publicly available non-toy ontology examples are already available, the effort of adapting an existing ontology to a specific application is both useful and possible. This experiment is being conducted in MOSES to treat the university domain.</Paragraph>
    <Paragraph position="9"> Ontologies for the Semantic Web are written in formal languages (OWL, DAML+OIL, SHOE) that are generalisations/restrictions of Description Logics (Baader et al. 2003). TBox assertions describe concepts and relations. A typical entry for a concept is:</Paragraph>
  </Section>
class="xml-element"></Paper>