Abstract

Centering Theory (CT) as articulated by Grosz et al. (1995) is a theory intended to model some of the factors that influence local coherence in a discourse. The idea is that at any one time there are a number of entities that are at the center of attention. Each utterance n that makes up a discourse potentially has two sorts of discourse ‘centers’, an ordered set of forward-looking centers, Cf(uttn), that provide potential links to upcoming utterances, and a single backward-looking center, Cb(uttn), that links back to the previous utterance, in the sense that it is identical to one of the members of the set of forward-looking centers of the previous utterance. Roughly, one can think of Cb(uttn) as the current topic of the conversation. The members of the set Cf(uttn) on the other hand are the entities mentioned in the utterance and which are candidates to become the next topic of conversation. The members of Cf(uttn) are ranked in order of their salience. The most highly ranked member of Cf(uttn) is the preferred center, Cp(uttn), and ideally it will become the backward-looking center of the next utterance, Cb(uttn+1). When Cb(uttn) is the most highly ranked member of Cf(uttn), this indicates that it will continue to be the topic of conversation in the next utterance. If this does indeed happen, we have what is known in CT as a CONTINUE transition. When some entity other than Cb(uttn) is the most highly ranked member of Cf(uttn), an upcoming topic shift is signaled. This situation is known in CT as a RETAIN. A topic shift following a RETAIN will be a SMOOTH SHIFT. If the topic shifts to an entity that is neither the preferred center of the current nor of the previous utterance, then the shift will be a ROUGH..