Abstrakt
Centering theory approach is supposed to be able to model local coherence of the text (Grosz, Joshi and Weinstein 1995): Centering states so called Continue type of Transition to be the dominant type of discourse relationships development. To test this hypothesis I annotated (partially automatically) sample of authentic Czech texts from Prague Dependency Treebank according to the Centering theory (Centers of Attention, Types of Transition, Realization Relationships, Pronoun Rule, Subject Rule etc.).
Annotation combines the Constituent Structure of the utterance with its Information Structure and focuses on the variability of the Centers of Attention classification. Based on such annotation it was found that Centering theory is not eligible to cover all aspects of the discourse relationships' development in Czech (e.g. problems with multiple pronouns utterances, coreferential expressions within single utterance, single backward-looking center restriction).
In my paper I would like to present the Centering theory background, principles of annotation, and to propose modification(s) which could make Centering Theory plausible.