Elena Cabrio - PostDoc at INRIA Multimedia Communications
Date: July 19th 2012 Location: Eurecom - Eurecom
As the social web is spreading among people and the web of data continues to grow (e.g. Linked Data initiatives), there is an increasing need to allow easy interactions between non-expert users and data available on the Web. In this perspective, we address the development of methods for a flexible mapping between natural language expressions, and concepts and relations in structured knowledge bases. In the first part of this talk, I will present an approach for open domain Question Answering over linked data. The goal is to allow end users to submit a query to an RDF triple store in English and get results in the same language, masking the complexity of SPARQL queries and RDFS/OWL inferences involved in the resolution, while profiting from the expressive power of these standards. The problem of question interpretation is addressed as the automatic identification of the set of relevant relations between entities in the natural language input question, matched against a repository of automatically collected relational patterns. Such patterns represent possible lexicalizations of ontological relations, and are associated to a SPARQL query on the knowledge base. Wikipedia is used as the source of free text for the automatic extraction of the relational patterns, and DBpedia as an example of linked data resource to be queried using a natural language interface. In the second part of the talk, I will briefly outline another research line I am currently investigating, i.e. the combination of natural language processing techniques to semantic inferences and argumentation theory to support online debates interactions.