Examining the perception-production link in speech processing through the lens of accent imitation

每个人都认识一些非常擅长模仿口音的人，或者是“语言变色龙”，他们会自然地模仿新社区的口音。虽然接触到许多不同的口音是每天都会发生的事情，特别是在像加拿大这样语言多样的国家，但我们如何识别和处理这些不同的说话方式存在巨大的差异。从以前的研究中有证据表明，包括双语能力、音乐训练和口音熟悉度在内的几个因素可能有助于模仿，我们也知道，成年人在学习新口音方面比儿童差得多。然而，不同的研究结果有所不同，这取决于被模仿口音的社会地位和语言特征。这个项目旨在揭示是什么让口音的某些特征比其他特征更容易模仿，是什么让一些人比其他人更好地模仿，更广泛的目标是理解支配声音模仿的认知结构。成功的口音模仿是一个复杂的行为，首先需要准确地感知给定口音的特征，然后准确地再现相同的特征。我们使用了一组实验，针对儿童与成人以及单语者与双语者之间几种不同语言特征（例如元音的发音或句子的音高模式）的模仿的各个子过程。为了保持对听众听到和模仿的完全控制，我们使用人工操纵的、社会中立的口音，这些口音只在一个特征上变化。我们的方法不仅允许系统地调查哪些因素是模仿成功的预测因素，而且还允许我们通过确定语音处理链中的哪些环节解释我们在个体中看到的变异性来理解它们为什么是预测性的。 口音模仿提供了一个透镜，通过它可以更好地理解语音感知和语音产生之间的关系，这是准确的语音处理模型的基本组成部分。除了理论上的意义，研究结果将提供一个新的视角，在外国的声音学习困难，这已被证明是与模仿能力的认知过程的基础。最后，虽然在过去十年中自动语音识别（例如Siri）取得了巨大的进步，但这些技术面临的一个持续挑战是处理口音变化。因此，更好地理解人类如何感知和适应口音可能有助于语音识别算法的改进。

Everyone knows someone who is really good at imitating accents, or "linguistic chameleons" who naturally pick up the accent of a newly adopted community. While exposure to many different accents is an everyday occurrence, particularly in linguistically diverse countries like Canada, there is an enormous amount of variability in how we recognize, and process these different ways of speaking. There is evidence from previous work that several factors, including bilingualism, musical training, and accent familiarity may facilitate imitation, and we also know that adults are notoriously worse than children at picking up new accents. However, findings have differed across studies depending on the social status and linguistic features of the accent being imitated. This project aims to uncover what makes some features of accents easier to imitate than others, and what makes some people better imitators than others, with the broader goal of understanding the cognitive architecture that governs vocal imitation. Successful accent imitation is a complex behaviour, requiring first accurate perception of the features of a given accent, followed by accurate reproduction of the same features. We use a set of experiments targeting individual sub-processes of imitation across several different linguistic features (e.g. the pronunciation of a vowel, or the pitch pattern of a sentence) in children vs. adults, and in monolinguals vs. bilinguals. In order to maintain full control over what listeners hear and imitate, we use artificially manipulated, socially neutral accents that vary only in a single feature. Our method not only allows for a systematic investigation of which factors are predictive of success in imitation, but also allows us to understand why they are predictive, by determining which links in the speech processing chain account for the variability we see across individuals. Accent imitation provides a lens through which to better understand the relationship between speech perception and speech production, a fundamental component of an accurate model of speech processing. Beyond the theoretical import, the results will provide a fresh perspective on the cognitive processes underlying difficulty in foreign sound learning, which has been shown to be linked to imitative abilities. Finally, while there have been huge advances in automatic speech recognition (e.g. Siri) in the past decade, one of the persistent challenges for these technologies is dealing with accent variability. A better understanding of how humans perceive and adapt to accents may therefore contribute to improvements in speech recognition algorithms.