Doctoral Dissertation Research: Incomplete Neutralization: The Loss and Maintenance of Contrast

语言并不总是按照书写的方式来发音。随着时间的推移，一种语言的发音会发生变化，但其正字法却保持不变。这给标准化拼写、语言教学甚至语音识别技术带来了问题，语音识别技术需要处理并将口语转换为正字法表示，反之亦然。这篇论文解决了发音随时间变化的问题，特别关注以前存在的两个声音之间的区别和正字法表达开始丢失的情况。这样的变化是逐渐发生的，声学上的差异（“声学线索”）慢慢消失，有时扬声器甚至没有意识到这一点。例如，“t”和“d”在荷兰语中是不同的发音，但随着时间的推移，“t”和“d”之间的区别在单词的末尾消失了；“bot”（骨头）和“bod”（报价）的发音都与英语单词“bought”相似。这种类型的声音变化，称为“中和”，并不总是完全的；两种声音之间可能存在微小但可测量的声学差异，尽管用裸耳听起来是一样的。研究表明，听者能够在潜意识中感知到这些差异，因为即使没有额外的上下文，他们也能以比偶然更好的速度分辨出两个看似相同的单词（例如荷兰语中的“bot”和“bod”）。目前还不清楚是什么导致了这种不完整，这个过程是如何随着时间的推移而展开的，以及什么社会特征影响了它。此外，目前还不清楚扬声器如何区分这些声音，以及区分它们所必需的声学是什么。本论文将研究显然不完全中和荷兰语和南非荷兰语的两个案例。两种语言都在失去“t”和“d”以及“p”和“b”之间的区别，无论是单词初始化还是单词最终化，尽管它们处于中和过程的不同阶段。在四个实验中，参与者将阅读含有这些辅音的单词和故事，然后用经过处理的声学方法听这些声音，看看他们能否区分它们。当我们发现听众能听到的差异时，我们就能更多地了解我们的认知能力，以及社会特征是如何影响我们的言语和听力的。这些知识可以用来告知语音识别技术，以更好地理解语音生产中的变异性。此外，我们将更好地了解语言发音随时间的变化，这可以帮助我们制定外语教学课程，特别是关注阅读和发音之间的联系。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Languages are not always pronounced the way they are written. Over time, the pronunciation of a language changes, but its orthography remains fixed. This poses problems for standardized spelling, language teaching and learning, and even speech recognition technologies, which need to process and convert spoken language into orthographic representations and vice versa. This dissertation addresses the problem of pronunciation change over time, looking specifically at cases where a distinction between two sounds that was previously present and represented orthographically begins to be lost. Changes like this happen gradually, where differences in the acoustics ("acoustic cues") slowly disappear, sometimes without speakers even realizing it. For example, "t" and "d" are different sounds in Dutch, but over time, the difference between "t" and "d" was lost at the ends of words; "bot" (bone) and "bod" (offer) are both pronounced like the English word "bought". This type of sound change, called "neutralization", is not always complete; small but measurable acoustic differences may remain between the two sounds, despite sounding identical to the naked ear. Studies have shown that listeners are able to perceive these differences subconsciously, as they can tell two seemingly-identical words apart (e.g. Dutch "bot" and "bod") at a rate better than chance, even without additional context. It is unknown what causes this incompleteness, how this process unfolds over time, and what social characteristics affect it. Additionally, it is unclear how speakers can tell these sounds apart, and what the necessary acoustics are for distinguishing between them. This dissertation will examine two cases of apparently incomplete neutralization in Dutch and Afrikaans. Both languages are losing the distinction between "t" and "d" and between "p" and "b", both word-initially and word-finally, though they are at different stages of the neutralization process. In four experiments, participants will read words and stories that have these consonants, and then listen to the sounds with manipulated acoustics and see if they can tell them apart. As we discover what differences listeners can hear, we are able to understand more about our cognitive capabilities and how social characteristics can affect our speech and hearing. This knowledge can be used to inform speech recognition technologies to better understand variability in speech production. Additionally, we will have a better understanding of language pronunciation change over time, which can help us develop curricula for teaching foreign languages, especially focusing on the link between reading and pronunciation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.