1. Introduction. The Klamath verb, as described by Gatschet (1890), and more clearly and definitively by Barker (1964), manifests a particularly complex structure, with twenty-five morphological position classes.'While a morphologically complex verb structure is hardly unusual in a North American language, Klamath presents a somewhat atypical picture, lacking some common categories such as agreement, while showing a considerable inventory of highly specific aspectual categories (eg,'beginning to do','planning to do','about to do','came close to doing','almost did', and'tried to do but failed'). On the other hand, it shares some western North American areal traits, such as instrumental and locational affixes, and it is clear that morphological parallels more specific than the general Western or California patterns can be found between Klamath and a number of other languages, including not only putatively Penutian languages such as Tsimshianic (DeLancey, Genetti, and Rude 1988), Wintuan (DeLancey 1987a; 1987b), Maiduan (DeLancey 1988), and Sahaptian (Rude 1987), but also the clearly non-Penutian Washo and Atsugewi (DeLancey 1988). This means that there is a considerable job of comparative work to be done both in seeking further parallels and in sorting these out according to their historical significance. None of the isolates or clearly established genetic groups in the area is closely related to one another, which is to say that each has a long period of independent development in which to have lost old grammatical structures and innovated new ones. Thus an important preliminary to grammatical comparison of a language like Klamath with any other is an attempt to determine on language-internal grounds which of its grammatical structures are old enough to merit comparison with possible cognate languages and which are innovative. This paper is offered as a

1 An earlier version of this paper was informally published as DeLancey (1990). Several premature suggestions and conclusions of that paper are modified or abandoned here. I am grateful to members of the audience at the 1989 Hokan-Penutian languages workshop for helpful comments, and most particularly to Noel Rude for very illuminating discussions of this material and Klamath grammar in general.