Distance-based decay in long-distance phonological processes: a probabilistic model for Malagasy, Latin, English, and Hungarian
Many--or perhaps all--long-distance assimilatory and dissimilatory phonological processes produce lexical or free variation exhibiting a broad generalization: the likelihood of process application decreases as the transparent distance between the trigger and the target increases (a phenomenon that I call distance-based decay). This thesis provides a unified analysis of distance-based decay, drawing from thousands of data reflecting three long-distance phonological processes across four languages. I account for the data within the framework of Maximum Entropy Harmonic Grammar (Smolensky 1986, Goldwater and Johnson 2003, Hayes and Wilson 2008), which allows for an adequate treatment of variation. I argue that distance- based decay can be captured across the four surveyed languages by an invariant decay function--a simple negative power function--that interacts with two language-specific parameters: the weight of the AGREE (for assimilatory cases) or DISAGREE (for dissimilatory cases) constraint and the weight of IDENT. My account is therefore an extension of Kimper 2011, who posits a scaling factor that scales the weight of markedness to account for the decay effect present in vowel harmony in Hungarian. While it is the case that decay rates differ across the languages I survey, my analysis accounts for such differences purely with the weights of markedness and faithfulness; that is, I show that differences in decay rate can be modeled without having to posit language-specific decay functions.