Evidentials in attitudes : do’s and don’ts

Abstract

This paper is devoted to evidentials in attitudinal complements. I start with two empirical observations. A. Some logically possible interpretations are systematically not attested for evidentials-in-attitudes. This new observation has no straightforward account in the current literature. B. Languages vary with respect to whether or not evidentials-in-attitudes shift, i.e. whether they are speaker-oriented (as in root declaratives) or not. The variation has been previously attributed to the semantic non-uniformity of evidentials. I argue against this view. To account for A, I propose that evidentials are self-ascriptions, which is additionaly motivated by their behavior in matrix clauses. To account for B, I propose that evidential shift is an instance of indexical shift driven by a monster operator `a la Anand and Nevins (2004), which explains previously unnoticed similarities in restrictions on both kinds of shift. Understanding what happens in attitude reports has often been key to the semantics of many phenomena, e.g. pronouns and modals. Offering the first systematic examination of evidentials-in-attitudes across languages, the paper makes a case for evidentials and broadens our understanding of perspective-sensitivity in general.