The centering framework provides an account of ways in which patterns of interutterance reference promote a local coherence of discourse (Grosz, Weinstein, Joshi 1995). It is based on the assumption that in a coherent discourse each utterance segment contains semantic entity that provides a link to the previous discourse segment (so-called backward-looking center) and a set of ordered entities that offer potential links to the next utterance (forward-looking centers ranked according to discourse salience). As such centering models relationships among focus of attention, choice of referring expression, and perceived coherence of utterances within a discourse segment.

Centering as an account of local coherence of the discourse can be incorporated into a more general framework concerning mechanisms of reference fixation. This account emphasises the role that salience of certain objects plays in a cognitive machinery responsible for computation of linguistic reference of definite descriptions, complex demonstratives and other referring expressions. Attempts to characterize meaning of “discourse salience” concentrate of grammatical and syntactical features of expressions that realize certain objects (e.g grammatical subject of a sentence is more salient than its object) or informational status (e.g. in terms of “givenness hierarchy”). We claim that there is no essential difference between “discourse salience” and other types of salience (e.g. cognitive/attentional or contextual salience) and all of them interplay in the actual process of fixing reference of referring expressions.

We applied this account to discourse processing to provide some insights on the linguistic and attentional state factors that contribute in Polish to the coherence among utterances within a discourse segment. We will report results of a series of experiments that explore different forms of pronominal reference in Polish, in particular zeros, pronouns and proper names in subject position. Although it is believed that various referring expressions are not equivalent with respect to their effect on coherence (Gordon, Grosz, Gilliom 1993, Grosz, Weinstein, Joshi 1995), their detailed influence on discourse is unknown (for English see: Hudson, Tanenhaus, Dell 1986; Hudson-D’Zmura, Tanenhaus 1998). We will argue that given a particular attentional state, differences in perceived coherence correspond in part to the inference demands made by different types of referring expressions characteristic of a particular language.