DescriptionThis dissertation presents two case studies on incomplete neutralization (IN): flapping in American English (AmE) and monomoraic lengthening in Japanese. I provide experimental evidence showing that the underlying contrast in each case is only partly neutralized. I argue that these cases represent two points on a continuum of completeness of neutralization: a plausibly perceptible surface distinction (Japanese), and an indistinguishable surface distinction (flapping). Perception experiments support the imperceptibility of IN in flapping (Experiments 3, 4, and 5). First, I argue that IN is not solely due to experimental artifacts. I show that the degree of neutralization remains constant between two tasks—one designed to increase such effects, and one designed to reduce them—suggesting that results of IN cannot be reduced to these factors (Experiments 1 and 2). Second, I address the types of contrasts which can undergo IN. Most studies on IN center on feature/segment-level contrasts (e.g., voicing). Experiments 6 and 7 show that the Japanese vowel length contrast undergoes IN in monomoraic lengthening. I argue that the processes that can lead to IN must include suprasegmental and prosodic ones. Third, given the experimental results, I claim that IN is best accounted for in a model of phonetics based on a weighted-constraint grammar (Legendre et al. 1990, Zsiga 2000, Flemming 2001). I propose two types of constraints: the first pressures segments to match a target value for a given phonetic measure (Flemming 2001). The second type of constraint, based on transderivational identity (Benua 1997, Steriade 2000), requires candidates to match a base form (determined by frequency) for a given phonetic measure. Under the weighted constraint model, constraint conflict is resolved by compromise, rather than strict domination. This compromise, combined with access to quantitative phonetic detail, allows the model to generate languages at both the plausibly perceptible end of the IN spectrum (like Japanese) and the imperceptible end (like AmE), as well as the continuum in between. Finally, I show that this model best accounts for the Directionality Observation of IN: the realization of two incompletely neutralized categories is predictable. On a continuum of the acoustic cue(s) that differentiate the categories, an incompletely neutralized segment falls between the canonical realizations of the two categories.