When studying references to human beings in narrative texts, we observe examples with exact and unambiguous referring expressions, like (some) proper nouns and definite descriptions, i.e., expressions that refer to a specific and precise referent. When personal pronouns are immediately interpreted as anaphora, it is then easy to identify coreferring relations. By taking into account transitivity, it is then possible to build on coreference chains. Because of its links with theme, topic continuity and referents' salience, a coreference chain is a kind of discourse structure that is taken into account by an increasing number of linguistic studies and Natural Language Processing (NLP) applications. But coreferring relations are often considered as strict: each referring expression of a text fully belongs to one coreference chain, i.e., following an all-or-nothing relationship. Some other referring expressions, for instance those that refer to groups of persons, organizations, or generic referents, raise some problems when attributing the expression to a coreference chain. In NLP applications, those fuzzy cases are often ignored or simplified. In order to go deeper into the modelling and formalization of reference and coreference, we propose in this talk a model that implies three degrees instead of all-or-nothing coreferring relations. We show how pronouns with indefinite reference, and in particular the French R-impersonal pronoun on, put a spotlight on such a model and allow reconsidering the semantic properties of such a complex marker.