Infant Statistical Learning in Natural Language Acquisition

自然语言习得中的婴儿统计学习

• 批准号: 8124530
• 负责人: Casey Lew-Williams
• 金额: $ 4.84万
• 依托单位: UNIVERSITY OF WISCONSIN-MADISON
• 依托单位国家: 美国
• 财政年份: 2011
• 资助国家: 美国
• 起止时间: 2011-04-01 至 2013-03-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8124530
• 关键词: Address; Adult; Affect; Attention; Caregivers; Characteristics; Child; Clinical; Cochlear Implants; Complex; Cues; Detection; Dimensions; Early Diagnosis; Early treatment; Environment; Exhibits; Exposure to; Female; Future; Goals; Hearing; Hearing Impaired Persons; Home environment; Impairment; Infant; Knowledge; Laboratories; Language; Language Delays; Language Development; Learning; Length; Life; Linguistics; Link; Machine Learning; Outcome Study; Pattern; Play; Population; Probability; Process; Property; Research; Research Personnel; Research Training; Risk; Role; Shapes; Source; Speech; Stream; Structure; Testing; Time; Variant; Vocabulary; Voice; Yang; autism spectrum disorder; base; expectation; experience; male; natural language; novel; research study; simulation; specific language impairment; tool

DESCRIPTION (provided by applicant): Infants are adept at tracking statistical regularities to segment words in continuous speech. Researchers have documented infants' learning abilities using highly simplified artificial languages as speech input. However, natural language is replete with variability. Infants listening to speech encounter different speakers, different word lengths, and numerous other dimensions of complexity. The proposed experiments will use speech that captures key aspects of the natural variation observed in infants' language environments to test whether infants use statistical learning mechanisms in identifying word boundaries. This research will provide a rigorous test of whether statistical learning is in fact linked to early language acquisition. Specific Aim 1 is to examine how infants segment words given variability in utterance types-specifically, the presence of both continuous speech and isolated words. A preliminary study showed that isolated words enhance infants' attention to statistical regularities in fluent, natural speech. Experiment 1 will test the hypothesis that hearing words in isolation helps infants discover other words in continuous speech. Experiment 2 will explore word segmentation given variation in voices, testing whether isolated words enhance statistical learning when speech comes from multiple talkers. Specific Aim 2 is to investigate how variability in language experience influences infants' subsequent detection of statistical regularities in fluent speech. In Experiment 3, infants will first receive brief laboratory exposure to novel isolated words. This pre-familiarization is expected to constrain infants' abilities to establish word boundaries in continuous speech. Experiment 4 will test whether cross-linguistic differences in the proportion of multisyllabic words in infant-directed speech shape infants' word segmentation abilities. Spanish-learning infants, who hear far more multisyllabic words than English-learning infants over the first year, are expected to show greater facility in segmenting trisyllabic words from continuous speech than English-learning infants. The results will assess the degree to which previous learning leads to expectations that facilitate processing future speech, or whether infants' computational abilities operate continuously at each moment in time. The outcomes of these studies will inform future research exploring statistical learning abilities in young children with emergent language impairments. PUBLIC HEALTH RELEVANCE: The goal of this research is to understand how typically developing infants discover structure in the complex speech they hear from caregivers. Future studies will use these tasks to assess language learning in children exhibiting difficulties with language (e.g., children at risk for specific language impairment, children with emergent autism spectrum disorder, or deaf children with cochlear implants). The eventual goals are to investigate whether the mechanisms responsible for learning are compromised in these clinical populations, and to develop more appropriate tools for early diagnosis and intervention.

描述（由申请人提供）：婴儿善于跟踪统计的语音来分割连续语音中的单词。研究人员已经记录了婴儿使用高度简化的人工语言作为语音输入的学习能力。然而，自然语言充满了可变性。婴儿在听演讲时会遇到不同的说话者、不同的单词长度以及许多其他复杂的维度。拟议的实验将使用捕捉婴儿语言环境中观察到的自然变化的关键方面的语音来测试婴儿是否使用统计学习机制来识别单词边界。这项研究将提供一个严格的测试，是否统计学习实际上是与早期语言习得。具体目标1是研究婴儿如何段的话，在话语类型的变化，特别是，存在的连续语音和孤立的话。一项初步研究表明，孤立的单词可以增强婴儿对流利自然语言中统计数据的注意力。实验1将测试这一假设，即孤立地听单词有助于婴儿在连续的语音中发现其他单词。实验2将探讨在语音变化的情况下进行分词，测试当语音来自多个说话者时，孤立的单词是否会增强统计学习。具体目标2是研究语言经验的变化如何影响婴儿随后在流利言语中检测统计学差异。在实验3中，婴儿将首先接受短暂的实验室暴露于新的孤立词。这种预熟悉预期会限制婴儿在连续语音中建立单词边界的能力。实验四将检验婴儿引导言语中多音节词比例的跨语言差异是否影响婴儿的分词能力。学习西班牙语的婴儿，谁听到更多的多音节词比英语学习的婴儿在第一年，预计将显示出更大的设施分割三音节词从连续讲话比英语学习的婴儿。研究结果将评估先前的学习在多大程度上导致了促进处理未来语音的期望，或者婴儿的计算能力是否在每一时刻都在持续运作。这些研究的结果将为未来的研究提供信息，以探索具有紧急语言障碍的幼儿的统计学习能力。 公共卫生相关性：这项研究的目的是了解典型的发育中的婴儿如何发现他们从照顾者那里听到的复杂语音的结构。未来的研究将使用这些任务来评估表现出语言困难的儿童的语言学习（例如，有特殊语言障碍风险的儿童、患有紧急自闭症谱系障碍的儿童或植入人工耳蜗的聋哑儿童）。最终的目标是调查这些临床人群中负责学习的机制是否受到损害，并开发更合适的早期诊断和干预工具。