Cumulative faithfulness effects in phonology

One of the hallmarks of optimality theory (OT) is strict domination: multiple low-ranked constraint violations cannot gang up on a higher-ranked constraint. However, such cumulative interactions have been shown to occur. This thesis examines the subset of cumulative interactions called cumulative faithfulness effects (CFEs). CFEs occur when a single unfaithful mapping is allowed in a word, but multiple unfaithful mappings are not. In languages with CFEs, violations of multiple lower-ranked faithfulness constraints gang up on a single higher-ranked constraint to eliminate outputs that are unfaithful in multiple ways, while allowing singly-unfaithful outputs to survive. The key generalization is that for languages in which multiple repair processes could be used to repair a marked element, the least unfaithful repair process is chosen. The fact that these effects are attested in a variety of languages and language domains presents a problem for OT, which cannot account for them. Moreover, CFEs produce transparent outputs and resemble conspiracies, two phonological phenomena for which OT is typically adept at accounting. The thesis thus has three goals.;The first is descriptive: I seek to determine under what circumstances CFEs occur. A typology of CFEs is proposed, illustrating that there are at least three different ways in which faithfulness constraints can behave cumulatively. The second goal is to determine in what phonological domains CFEs occur. It is shown that CFEs are a pervasive phenomenon; examples are provided from fully-developed languages, first-language acquisition, and loanword adaptation. Finally, the third goal is analytical: what is the best constraint-based analysis for CFEs? Because OT cannot account for them, either some addendum to OT must be proposed, or some other constraint-based theory must be appealed to. Harmonic grammar (HG), a constraint-based theory in which each constraint carries a weight and candidates are evaluated based on the cumulative weight of their violations, is argued to provide the best analysis of CFEs. Multiple low-weight constraints may thus combine to overcome a single higher-weight constraint. Moreover, HG can account not only for those cases in which multiply-unfaithful outputs are disallowed, but also for cases in which multiply-unfaithful outputs occur.