Ontology extraction for coreference chaining

Abstract

The KunDoc project investigates coreference chaining with ontology-based methods. In this paper, we discuss knowledge-based methods for coreference chaining and in particular the use of ontologies and their acquisition from a corpus. We present the KunDoc methodology and its implementation. We use concepts and their interrelations extracted from a corpus of Norwegian newspaper articles to build up domain-specific ontologies which contribute with selectional restrictions on possible co-referents. We expect to see an improvement over methods that do not employ any semantic knowledge.