The article argues that, contrary to a widespread view (e.g. Haiman, 1985; Palmer, 1984), agreement in those languages which exhibit it is not a purely redundant, semantically empty and grammatically predictable phenomenon, but performs several important functions at the level of discourse.
Taking French as the example language, I will argue (section 2.1) that agreement signals the function-argument interpretation to be assigned to pairs of expressions of various kinds; and second, that it may also code anaphorically the high-focus status of particular discourse referents (section 2.2). Section 3 compares certain written errors in agreement marking made by advanced learners of French, with certain other interpretative errors in their reading of French articles - errors based on agreement relations and leading to the mis-assignment of reference to an agreement target or personal pronoun. Finally, section 4 argues that third person personal pronouns should be treated differently from the (essentially predicative) agreement targets discussed in sections 2 and 3, claiming that they do not participate in agreement stricto sensu.