Generating Referring Expressions with OWL2

The task of generating referring expressions, an important subtask of Natural Language Generation is to generate phrases that uniquely identify domain entities. Until recently, many GRE algorithms were developed using only simple and essentially home-made formalisms. Following the fast development of ontology-based systems, reinterpretations of GRE in terms of description logic have been studied. However, the quantifiers generated are still limited, not exceeding the works covered by existing GRE approaches. In this paper, we propose an DL-based approach to GRE that exploits the full power of OWL2 to generate referring expressions that goes beyond the expressivity of previous GRE algorithms. The potential of DL reasoning in GRE is also discussed. 1 GRE and KR: the story so far Generation of Referring Expressions (GRE) is the subtask of Natural Language Generation (NLG) that focuses on the identification of objects in natural language. For example, Fig.1 depicts the relations between several individuals of women, dogs and cats. In such a scenario, a GRE system might identify a given object as “Dog” or, if this fails to identify the referent uniquely, “the Dog that loves a Cat”, which is literally unique for d1. Reference has long been a key issue in theoretical linguistics and psycholinguistics, and GRE is a crucial component of almost every practical NLG system [5, 4]. Accordingly, GRE has become one of the best developed areas of NLG, with links to many other areas of Cognitive Science. Fig. 1. An example of women, dogs and cats, in which edges from women to dogs denote feed relations, from dogs to cats denote love relations. Traditional GRE algorithms are usually based on very simple, essentially homemade, forms of Knowledge Representation (KR); in many cases all properties are exProc. 23rd Int. Workshop on Description Logics (DL2010), CEUR-WS 573, Waterloo, Canada, 2010.