Behavioral and Neural Measures of Spoken Word Recognition in Late Language Emergence

摘要 语言出现晚（LLE）或说话晚（LT）的幼儿是18-35个月大， 口语词汇，但平均非语言能力，没有明显的感官障碍或其他发展障碍 自闭症谱系障碍等延迟（Collisson，2016; Paul & Jennings，1992）。约15%的 幼儿符合LLE标准（Singleton，2018; Paul，1992），并且持续性终身语言的风险较高 和识字不足，导致社会，学术和职业成果不佳（保罗，1993年; Rescorla，2009年; Singleton，2018）。超过16%的LT将被前瞻性地诊断为口语和/或书面语言 9岁时出现语言障碍（Paul，Murray，Clancy，& Andrews，1997），而大多数LT保留次优语言 在成年早期发挥作用（Rescorla，2002; Singleton，2018）。本研究旨在（1） 识别慢性影响的早期行为和神经标志物，以便最佳地分配稀缺的早期 干预资源，（2）检查行为表型的变化和分布，这将提供 为一系列临床和亚临床LT的更集中和有针对性的干预形式奠定基础 语言成果。 该项目将补充以前的工作LLE侧重于语言生产的评估罚款- 在LT和两个对照组（通常发展年龄， 匹配和语言匹配的对等体）使用行为（眼动跟踪）和神经（脑电图） 测量.所有参与者将完成全面的临床表征，包括评估 语言和非语言功能。在实验1中，参与者将接受简单选择任务的培训 使用八个在语音上重叠的熟悉的单词（例如，开始时，BUNNY-气泡，偏移时，KITTEN- MITTEN）和语义（例如，BUNNY-KITTEN）。眼动追踪将用于估计组和个体 词汇激活和竞争的差异随着时间的推移在选择任务。在实验2中， 将被记录在一个被动的听力范例，使用脑电图，作为参与者观看无声电影，而熟悉的 实验1中的单词被重复多次。我们将训练机器学习算法（支持向量 机器或SVM）解码每个蹒跚学步的孩子对特定单词的完整EEG反应。团体和个人 SVM相对成功的差异可以反映神经响应的一致性和保真度。ERP （事件相关电位）分析还将用于检查平均值的组和个体差异 对所说的话的反应。眼动追踪、EEG和/或ERP测量的组和个体差异 将产生新的见解的语言处理能力的LT，并将提供一个基础，为未来的工作， 找出那些最有可能导致不良结局的LT。

ABSTRACT Toddlers with late language emergence (LLE), or late talkers (LTs), are 18-35 month-olds with a limited spoken vocabulary, but average non-linguistic abilities and no overt sensory impairments or other developmental delays such as autism spectrum disorder (Collisson, 2016; Paul & Jennings, 1992). Approximately 15% of toddlers meet LLE criteria (Singleton, 2018; Paul, 1992) and are at elevated risk for persistent, lifelong language and literacy deficits that result in poor social, academic and vocational outcomes (Paul, 1993; Rescorla, 2009; Singleton, 2018). Upwards of 16% of LTs will prospectively be diagnosed with a spoken and/or written language disorder at age 9 (Paul, Murray, Clancy, & Andrews, 1997) while the majority of LTs retain suboptimal language functioning through early adulthood (Rescorla, 2002; Singleton, 2018). The proposed research seeks to (1) identify early behavioral and neural markers of chronic impact in order to optimally allocate scarce early intervention resources, and (2) examine the variation and distribution of behavioral phenotypes which will provide the foundation for more focused and targeted forms of interventions for LTs with a range of clinical and subclinical language outcomes. The project will complement prior work on LLE focused on language production by evaluating the fine- grained timecourse of spoken word recognition (SWR) in LTs and two control groups (typically developing age- matched and language-matched peers) using behavioral (eye tracking) and neural (electroencephalography) measurements. All participants will complete comprehensive clinical characterization including assessments of language and non-linguistic functioning. In Experiment 1, participants will be trained on a simple selection task using eight familiar words that overlap phonologically (e.g., at onset, BUNNY-BUBBLES, at offset, KITTEN- MITTEN) and semantically (e.g., BUNNY-KITTEN). Eye tracking will be used to estimate group and individual differences in lexical activation and competition over time in the selection task. In Experiment 2, neural responses will be recorded in a passive listening paradigm, using EEG, as participants view a silent movie while the familiar words from Experiment 1 are repeated many times. We will train a machine-learning algorithm (support vector machine, or SVM) to decode the full EEG response to specific words for each toddler. Group and individual differences in the relative success of the SVM may reflect the consistency and fidelity of neural responses. ERP (event-related potential) analysis will also be used to examine group and individual differences in mean responses to the spoken words. Group and individual differences in eye tracking, EEG and/or ERP measures will yield new insights into the language processing abilities of LTs and will provide a basis for future work aimed at identifying those LTs at greatest risk for poor outcomes.