Decoding continuous speech in infancy

解码婴儿期的连续语音

• 批准号: 355894-2008
• 负责人: Johnson, Elizabeth
• 金额: $ 1.67万
• 依托单位: University of Toronto
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2012
• 资助国家: 加拿大
• 起止时间: 2012-01-01 至 2013-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=497170
• 关键词: Decoding; continuous; speech; infancy

Thinking of language acquisition as a process that begins when children utter their first meaningful words overlooks an important stage of language development. Long before infants are even close to attaching meanings to words or producing grammatical utterances, they must first learn to decode the speech signal. Research on adult speech perception has demonstrated that decoding the speech signal is a complicated task, requiring a sophisticated understanding of the speech signal. Even identifying word boundaries in speech is challenging because there are no reliable acoustic cues to word boundaries (e.g., silences between words), and no word is produced in an identical fashion twice. Despite these difficulties, infants begin recognizing words remarkably early. Most current models of language acquisition place a heavy burden on infants' ability to extract regularities from their language input. A great number of studies have been reported which support this statistical learning view on early language development, but many of these studies assume an idealized speech input and adult-like attention to the speech signal. But these assumptions are not necessarily justified. In this project, I will address these issues with the goal of better aligning current developmental models of speech perception with the real-life perceptual and cognitive capabilities of normally developing infants. Specific issues that will be addressed include 1) whether infants are more attentive to prosody than adults, 2) whether it is easier for infants to notice mispronunciations in some contexts than others, and 3) whether infants can track statistical patterns in real speech as easily as they do so with the simplified speech used in many laboratory studies.

如果认为语言习得是从儿童说出第一个有意义的单词时开始的过程，那么就忽视了语言发展的一个重要阶段。 早在婴儿即将赋予单词意义或产生语法话语之前，他们就必须首先学会解码语音信号。 对成人语音感知的研究表明，解码语音信号是一项复杂的任务，需要对语音信号有深入的理解。即使识别语音中的单词边界也具有挑战性，因为没有可靠的单词边界声学线索（例如单词之间的沉默），并且没有单词以相同的方式产生两次。尽管存在这些困难，婴儿还是很早就开始识别单词。当前大多数语言习得模型对婴儿从语言输入中提取规律的能力造成了沉重负担。已经报道的大量研究支持这种关于早期语言发展的统计学习观点，但其中许多研究假设理想化的语音输入和对语音信号的成人般的关注。但这些假设并不一定合理。在这个项目中，我将解决这些问题，目标是更好地将当前的言语感知发育模型与正常发育的婴儿的现实感知和认知能力结合起来。将要解决的具体问题包括：1）婴儿是否比成人更注意韵律；2）婴儿在某些情况下是否比其他情况更容易注意到发音错误；3）婴儿是否可以像使用许多实验室研究中使用的简化语音一样轻松地跟踪真实语音中的统计模式。