Structure & Specification in Harmony*

For as long as there has been a discipline of generative phonology—and even before that—there has been heated debate over the nature of phonological representations and processes. At one end of the spectrum we find linguists who assume hierarchical structure: morae and/or onsets and rimes grouped into syllables; syllables grouped into feet; feet grouped into words. At the opposite end of the spectrum are those who, like Neeleman & van de Koot (2006), argue that one of the fundamental ways in which syntax and phonology differ is that phonology is not hierarchical but instead completely “flat.” In this paper I focus on one phenomenon, vowel harmony, which has commonly been thought to require hierarchical or tiered representations. In the first portion of the paper, I argue that vowel harmony does not require any such structure, and can be accommodated using a linear/flat model of phonology combined with simple search and copy operations which find parallels elsewhere in linguistics, and which may not even be specific to language. In the second portion of the paper, I discuss the implications of this theory for an analysis of the harmony-altering language game in Tuvan described by Harrison & Kaun (2000, 2001), concluding that this game and others like it do not involve true harmony processes. First, let us discuss the typology of vowel harmony in terms that are as theory-neutral as possible. Vowel harmony takes the general shape in (1), taken from Turkish. A morpheme, typically a suffix, takes on a feature value—here, [α BACK]—from a vowel in the root. Thus, we see alternation in the plural suffix between [e] and [a], and in the genitive between [i] and [ɨ] depending on whether the root contains a front or back vowel. *This work benefited greatly from comments by audiences at Harvard, MIT, and NELS 38