Geminate Inalterability and Lenition

0. INTRODUCTION It is a well-known observation that phonological processes which apply to short segments frequently fail to apply to corresponding long ("geminate") segments. For example, post-vocalic spirantization of velar stops in Tigrinya yields [/a-xalIb] 'dogs' (cf. [k elbi] 'dog'), but [fekk er e] 'boasts', not [f exk er e] nor [f exx er e] (Kenstowicz 1982).1 This phenomenon of geminate "inalterability" or "blockage" has been the subject of a number of proposals within the framework of Autosegmental Phonology, most influentially Hayes 1986 and Schein & Steriade 1986.2 Subsequent research, however, has revealed that these proposals make seriously incorrect predictions as to the class of processes which display inalterability (see Inkelas & Cho 1993). As Churma (1988) observes, geminate inalterability holds true as a universally inviolable condition only in the domain of lenition phenomena, a generalization which the classic inalterability approaches fail to capture. Moreover, as Elmedlaoui 1993 notes, within the domain of lenition phenomena, the classic approaches are insufficiently restrictive: they fail to rule out processes which specifically target geminates for lenition, e.g. /kk/ -> *[xx], or which convert an underlying singleton to a lenited geminate, e.g. /k/ -> *[xx]; and they fail to draw a connection between inalterability and the general markedness of "weaker" (i.e. continuant and voiced (obstruent)) geminates, whether derived via some lenition process or present underlyingly. In light of major shifts in phonological theory which have occurred since the previous cycle of inalterability research, particularly the development of Optimality Theory (Prince & Smolensky