The NLP Task at INEX 2005

With XML information retrieval, like in traditional IR, the user's information need is loosely defined, linguistic variations are frequent, and answers are a ranked list of relevant elements. Like in database querying, structure is of importance and a simple list of keywords may not be sufficient to define an XML query. Structured query languages for XML have been developed, but appear to be difficult to use even by system-level users, let alone end-users. Therefore developing natural language interfaces for XML-IR requires innovative solutions. INEX provides a framework (documents, topics and relevance assessments) for independent evaluation of XML-IR systems and approaches. In 2002 and 2003 systems in the Ad-hoc task accepted formal language queries (i.e. elements) and produced results lists of relevant XML elements (usually well below the document root). In 2004 INEX introduced a Natural Language Processing task that operated in parallel with the Ad-hoc task (using the same topics and assessments) except that systems operated on XML natural language queries (i.e. elements). In this report we describe the motivation for using NLP in XML IR, the general approaches that were tested at INEX 2005, and comment on the results