On generating coherent multilingual descriptions of museum objects from Semantic Web ontologies

During the last decade, there has been a shift from developing natural language generation systems to developing generic systems that are capable of producing natural language descriptions directly from Web ontologies. To make these descriptions coherent and accessible in different languages, a methodology is needed for identifying the general principles that would determine the distribution of referential forms. Previous work has proved through crosslinguistic investigations that strategies for building coreference are language dependent. However, to our knowledge, there is no language generation methodology that makes a distinction between languages about the generation of referential chains. To determine the principles governing referential chains, we gathered data from three languages: English, Swedish and Hebrew, and studied how coreference is expressed in a discourse. As a result of the study, a set of language specific coreference strategies were identified. Using these strategies, an ontology-based multilingual grammar for generating written natural language descriptions about paintings was implemented in the Grammatical Framework. A preliminary evaluation of our method shows languagedependent coreference strategies lead to better generation results. createdBy (Guernica, PabloPicasso) currentLocation (Guernica, MuseoReinaSofía) hasColor (Guernica, White) hasColor (Guernica, Gray) hasColor (Guernica, Black) Guernica is created by Pablo Picasso. Guernica has as current location the Museo Reina Sofía. Guernica has as color White, Gray and Black. Figure 1: A natural language description generated from a set of ontology statements.