Reinterpreting Transparency in Nasal Harmony*

1 . Introduction In this paper I examine crosslinguistic variation in nasal harmony. Three kinds of segment behavior are observed: target segments become nasalized in nasal harmony (/na/ → [na)]), blocking or opaque segments remain oral and block nasal spreading (/nata/ → [na) ta]), and transparent segments remain oral and do not block nasal spreading (/nata/ → [na) ta) ]). The membership of these categories varies in limited ways across languages. The aim of the present work is to establish a unified understanding of nasal harmony, so that all patterns conform to one basic character—something that has not been achieved before. A second goal is to examine the wider implications for phonological theory, particularly the consequences for analysis of transparency and locality in feature spreading. A central claim defended here is that all nasal harmony patterns are constrained by a hierarchy ranking segments according to their compatibility with nasalization. Previous work has suggested that a nasalization hierarchy is relevant only for defining sets of target segments versus blockers. However, this view is faced with a complementarity problem. First there appear to be no examples of a certain pattern predicted by the hierarchy, one in which all segments including obstruents are nasalized. Second, another system is isolated from the others, one where some obstruents act transparent and all remaining segments are targets. The crosslinguistic study presented here reveals that target and transparent segments pattern together with respect to the nasalization hierarchy: if a class of segments propagates nasal spreading (is targeted or behaves transparent), all higher-ranked classes also propagate nasalization. To explain this, I propose to analyze descriptively transparent segments together with targets of nasality spreading as a unitary class of permeable segments, i.e. segments that participate in nasal harmony. The possible outcomes for segments in feature spreading becomes accordingly simpler: they either undergo nasal spreading or they block. Systems with transparency emerge as instances where all segments undergo nasal spreading, achieving a unified typology of nasal harmony where the nasalization hierarchy exhaustively limits variation. Interestingly, this result finds support for a view of locality in which feature spreading occurs only between strictly adjacent segments (Gafos 1996; Ní Chiosáin and Padgett 1997), a notion that previously seemed to be denied by the nasal harmony data. The unified typology obtains the hierarchical variation in segments permeated by nasalization versus blockers. However, segments undergoing nasalization are noted to have two possible phonetic outcomes: nasal or oral. The latter occurs only on permeable segments near the extreme of incompatibility with nasalization, typically voiceless obstruents. I argue that this transparency outcome arises as a derivational opacity effect, a phenomenon captured under Sympathy Theory (McCarthy 1997; Itô & Mester 1997). The effect arises in a correspondence mapping between a fullynasalized but unpronounceable phonological output representation ([na) t)a) ]—unpronounceable because [t)] cannot be phonetically-realized) and a similar but phonetically-possible output ([na)ta) ]). This paper is organized as follows. In section 2 I lay out the cross-language typology of nasal harmony, drawing on generalizations established by an extensive survey of nasal harmony systems. Based on the co-patterning of target and transparent segments, I propose to merge these categories, producing a new, unified understanding of nasal harmony. In section 3 I develop an optimalitytheoretic analysis. Bearing out the predictions of the theory, the range of attested patterns is obtained by exhausting the possible rankings of a spreading constraint in relation to a fixed hierarchy of nasalized segment markedness constraints. Section 4 turns to the different realizations for permeable segments, and develops an analysis of segment transparency as derivational opacity. Section 5 presents the conclusion.