Phonological processes interacting with the lexicon : variable and non-regular effects in Japanese phonology

• Main Author: Rosen, Eric Robert
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/13157

In current generative linguistic theory, a speaker of a natural human language possesses a language faculty that includes a lexicon: a set of language-specific input forms, and a grammar: a set of constraints or rules that derive the surface or output forms of the language from a structured combination of input forms. In Optimality Theory (Prince & Smolensky (1993)), output forms are determined by an ordered set of constraints. Constraints operate only on output forms and not input forms: thus we should expect input forms not to show the effects of a grammar and to occur in statistically random patterns. Conversely, we expect output forms to occur in patterns that are favoured by the grammar. This thesis examines apparent exceptions to both these predicted tendencies, with respect to two phonological phenomena in Japanese: (a) rendaku voicing, which causes voicing at a morpheme juncture in a compound word and (b) pitch-accent patterns of monomorphemic Yamato (native) nouns, which fail to show a predicted randomness of statistical patterns. I argue that a compound word is prone to exceptions to processes like rendaku voicing because its input form must occur partly as a minimal lexical item of its own rather than simply a concatenation of two constituent input forms. This opens up the possibility for a compound word's input form to include independent phonological features that can lexically block a process like rendaku. Here, the lexicon affects the regularity of output forms. For noun pitch-accent, I argue that apparent non-randomness of input forms is actually due to the lack of constraints on input forms in Optimality Theory. If any type of input is possible, then the surface forms of monomorphemic nouns do not necessarily reflect their input forms. Apparent non-randomness of inputs can arguably be due to a convergence of several input types on one output type, as required by the grammar, and can thus be explained as non-randomness of outputs, not inputs. In current linguistic theory, both of these claims have important implications for the nature of the lexicon, the grammar, and the way they interact.