Abstract:

This study addresses the issue of phonological opacity. While many previous proposals have attempted to offer all-encompassing theories of opacity, the goal of this study is more modest based on the belief that further understanding of the phenomenon of opacity will come from looking at specific types of phonological processes and their interactions, as opposed to looking at all cases of opacity as a singular phenomenon. To that end, I introduce a new set of generalizations by focusing on cases involving harmony and tonal interactions. Indeed, this particular empirical domain suggests a different understanding of what it even means to be opaque. Furthermore, the general typological finding is that cases of opacity involving tonal interactions and harmony reflect greater faithfulness to the underlying representation. This can be formalized within OT by expanding on two-level markedness constraints (McCarthy, 1996) and Correspondence Theory (McCarthy & Prince, 1995) to include diagonal correspondence, a two-level faithfulness constraint. A specific prediction this approach makes is that certain types of feeding and bleeding relationships should not be found when harmonic and tonal interactions are involved. Instead, opacity is the preferred type of process interaction i.e. harmony is generally counter-fed and counter-bled by other processes, contra the hypothesis that opaque interactions are marked (Kiparsky, 1971).

One of the challenges opacity presents for theories of phonology is a learnability problem. Indeed, a theory based on increased faithfulness to input forms requires that language learners be able to abstract underlying representations for words despite potentially never hearing them. Using an artificial-grammar learning paradigm, I show that language learners are indeed able to do so. These results suggest that some process of abstraction is needed to augment theories based on surface-derived generalization such as Natural Generative Phonology (Hooper [Bybee], 1976), exemplar-theory (Pierrehumbert, 1991; Johnson, 1997), statistical theories of phonology (Goldsmith, 2005) and indeed, OT. Furthermore, under-application/counter-feeding opacity was more learnable than over-application/counter-bleeding opacity for harmony contra the hypothesis that under-application is more marked (Kiparsky, 1968). Finally, the results make predictions as to the maximum proportion of opaque to transparent forms allowed in a language.