Referring Expression Generation in Situated Interaction

While most current frameworks for reference handling are based on binary truth-theoretic knowledge representation, in this thesis I argue for a perspective on reference which emphasises the collaborative nature of reference. I present the Probabilistic Reference And GRounding mechanism (PRAGR) which uses flexible concept assignment based on vague property models and situational context in order to maximise the chance of communicative success. I demonstrate that PRAGR is capable of dealing with several property domains with different internal structures, such as graded adjectives, colour, shape, projective terms, and projective regions. Further, I show that PRAGR is fit to handle in an integrated fashion the most relevant challenges of Referring Expression Generation, in particular graded properties, spatial relations, and salience effects. In three empirical evaluation studies, I demonstrate the usefulness of PRAGR for situated referential communication.