Changes to the realization of phonetic cues, such as vowel length or voice onset time, can have differential effects on the system of phonological categories. In some cases, variability or bias in phonetic realization may cause a contrast between categories to collapse, while in other cases, the contrast may persist through the phonologization of a redundant cue (Hyman, 1976). The goals of this dissertation are to better understand the subphonemic conditions under which a contrast is likely to survive and when it is likely to collapse, as well as to understand why certain cues are more likely to be phonologized than others.;I explore these questions by considering the categorization of speech sounds over a noisy channel (Shannon and Weaver, 1948), hypothesizing that when the precision of a contrast along one acoustic dimension is reduced, another dimension may be enhanced to compensate (the probabilistic enhancement hypothesis). Whether this results in phonologization or neutralization depends on both the degree to which the contrast is threatened as well as the informativeness of the cues that signal it.;In order to explore this hypothesis, phonological categories are modeled as finite mixtures, which provide a natural way to generate, classify, and cluster objects in a multivariate setting. These mixture models are then embedded in an agent-based simulation framework and used to simulate the ongoing process of phonologization of pitch in Seoul Korean (Kang and Guion, 2008). The results demonstrate that adaptive enhancement can account for both cue selection as well as the appearance of cue trading in phonologization. Additional data from the incomplete neutralization of final voicing in Dutch (Warner, Jongman, Sereno and Kemps, 2004) are then used to show how variation in phonetic realization can influence the loss or maintenance of phonological categories. Together, these case studies illustrate how the principles which influence the production and perception of subphonemic cues can ultimately impact the system of phonological contrasts.
展开▼
机译：语音提示的实现方式（例如元音长度或语音开始时间）的变化可能会对音位类别系统产生不同的影响。在某些情况下，语音实现的可变性或偏差可能会导致类别之间的对比崩溃，而在其他情况下，这种对比可能会通过冗余提示的语音化而持续存在（Hyman，1976年）。本文的目的是为了更好地理解对比度可能存在的亚音环境以及何时可能崩溃的情况，以及为什么某些线索比其他线索更容易被语音化。通过考虑嘈杂通道上语音的分类（Shannon和Weaver，1948年），假设当沿一个声学维度的对比度的精度降低时，可以增强另一个维度来进行补偿（概率增强假设）。是否导致语音化或中和取决于对威胁的威胁程度以及发出信号的提示的信息性。为了探究这一假设，将语音类别建模为有限混合，从而提供自然的多变量设置中生成，分类和聚类对象的方法。然后，将这些混合模型嵌入基于代理的仿真框架中，并用于模拟汉城韩语中音高的语音化过程（Kang and Guion，2008）。结果表明，自适应增强可以说明语音选择中的提示选择以及提示交易的出现。然后，使用来自荷兰语的最终语音的不完全中和的其他数据（Warner，Jongman，Sereno和Kemps，2004年）来说明语音实现的变化如何影响音位类别的丧失或维持。这些案例研究一起说明了影响次音暗示产生和感知的原理如何最终影响音系对比系统。
展开▼
