Japanese Evidentials as Modals

This talk reports on a joint project with Norry Ogata on the semantics of Japanese evidentials. Until recently, evidential expressions have not received much attention in the (formal) semantic literature. This situation has changed in the last few years; influential work by Izvorski (1997) and Faller (2002), for instance, has inspired a good deal of work. The consensus at this point seems to be that evidentials have at least one characteristic in common: they must take widest scope. Even if they appear at the surface to be embedded under negation or modal operators, for instance, they cannot scope under them. In this paper we show that there are languages—namely, for instance, Japanese—in which this characterization does not hold. Many other questions are answered by the literature on evidentials; but many questions remain. Consider, for instance, hearsay evidentials. For instance, for is it enough just to overhear some information, or must one be told directly? Can inference play a role? More generally, since most of the literature on evidentials is oriented to functional and descriptive approaches (with some notable exceptions), a number of questions of interest to semanticists in the formalization of evidential facts are left unaddressed. How does evidential content affect anaphoric relations? Do evidentials block anaphora in a way similar to modals (Roberts, 1989), or not? Can evidential content itself be picked up by propositional anaphors? We address some of these questions using data from Japanese. The conclusion is that evidentials are rather similar to modals in this language in terms of their semantic behavior with respect to other operators and anaphors. They differ enough, however, that it is not obvious how to analyze them using traditional semantics for modals, as we also show. Time permitting, I will present a formal analysis which, while it makes use of possible worlds, also draws on ideas from probability theory and formulations of evidential reasoning in computer science. The data used in the talk draws on two types of evidential expressions, the inferential and hearsay evidentials. We look at four distinct inferential evidentials—mitai, (INF+)soo-da, yooda, and rashii, which also can be used as a hearsay evidential—and one pure hearsay evidential, (S+)soo-da.