Date : 06 February 2015
Speaker: Alexandre Kiburse (Joint work with Laura Perez-Beltrachini)
Abstract: The project is framed in the Quelo Tool for ontology-based data access and its Natural Language Interface (NLI). Quelo's NLI relies on Natural Language Generation (NLG) techniques to display formal queries in Natural Language (NL), namely NL queries. Internally, the Quelo Tool uses a formal query representation, namely a Description Logics (DL) query. The formal query representation is not visible to the user; in the NLI, they only see the NL version of the query. Quelo NLI keeps a link between elements of the DL query and text spans of the NL query. This link is used to provide visual hints of the underlying DL query structure on its NL version and to provide query manipulation operations. The work in this project contributes to the integration of a new NLG module based on Tree Adjoining Grammars (TAG) equipped with semantics into Quelo's NLI and involves two main tasks. One of them is the extraction of semantically annotated NL queries from the generator output. The other task consists in exploring the use of text fluency predictions to select amongst candidate NL queries produced by the surface realiser.