Interactions between word learning and visual category development in infancy

该奖项是作为NSF的社会，行为和经济科学博士后研究奖学金（SPRF）计划的一部分提供的。SPRF计划的目标是为学术界，工业或私营部门和政府的科学事业准备有前途的早期职业博士级科学家。SPRF的奖励包括在知名科学家的赞助下进行两年的培训，并鼓励博士后研究员进行独立研究。NSF致力于促进来自科学界各部门的科学家，包括来自代表性不足的群体的科学家参与其研究计划和活动;博士后期间被认为是实现这一目标的专业发展的重要水平。每个博士后研究员必须解决推进各自学科领域的重要科学问题。要学习一个具体的名词（例如，狗），婴儿必须学习如何将这个词概括为同一类别的多个实例（例如，黄金猎犬和吉娃娃）。婴儿是如何做到这一点的呢？ 这个项目的首要目标是扩大我们对婴儿如何学习单词所指的视觉类别的理解。 具体来说，这个项目测试是否新的标签作为社会线索，以帮助婴儿形成新的视觉类别更具体的表示。为了实现这一目标，该项目将融合视觉科学和单词学习的技术，以检查婴儿在短期眼动跟踪实验中映射到新单词（与新音调）的视觉信息类型。此外，该项目将使用单词学习的计算模型来模拟婴儿的行为。因此，该项目将详细描绘婴儿映射到新单词的视觉信息。这项提议的结果可能对自闭症谱系障碍（ASD）儿童产生影响，他们在单词学习过程中往往难以处理社会线索。最终，这些短期眼动追踪任务的表现可以作为ASD风险婴儿的评估和ASD干预研究的结果指标。最后，该项目还将创建新的数据集和工具，这些数据集和工具将使用开放科学框架公开提供。婴儿如何从单个实例中进行归纳，以学习单词所指的视觉类别？也许语言在将实例结合在一起方面起着一定的作用。先前的研究表明，当婴儿听到一个新的标签（相对于一个新的音调），他们希望看到一个新的视觉类别的对象。然而，如果婴儿将新标签视为强烈的社会线索，他们应该形成一个对象类别，贝叶斯模型的单词学习预测，新标签也应该帮助婴儿形成一个更具体的表示这一类别。这个项目通过结合眼动追踪行为实验、早期视觉处理的计算模型和单词学习的贝叶斯模型来验证这一假设。首先，该项目将检查标签是否有助于婴儿形成相对于其他新奇物体的新奇类别的更紧密的视觉表征。其次，该项目将检查标签是否改变了视觉特征的种类（例如，纹理与3D形式）婴儿用来代表新的类别，利用纹理合成模型和3D打印技术。第三，该项目将检查贝叶斯单词学习的最佳模型是否能预测行为结果。该项目采用了一种新的方法来研究早期单词学习，通过结合视觉感知的现代工具，从而在语言习得和视觉认知发展领域之间建立了新的联系，同时进一步加深了我们对两者的理解。

This award was provided as part of NSF's Social, Behavioral and Economic Sciences Postdoctoral Research Fellowships (SPRF) program. The goal of the SPRF program is to prepare promising, early career doctoral-level scientists for scientific careers in academia, industry or private sector, and government. SPRF awards involve two years of training under the sponsorship of established scientists and encourage Postdoctoral Fellows to perform independent research. NSF seeks to promote the participation of scientists from all segments of the scientific community, including those from underrepresented groups, in its research programs and activities; the postdoctoral period is considered to be an important level of professional development in attaining this goal. Each Postdoctoral Fellow must address important scientific questions that advance their respective disciplinary fields. To learn a concrete noun (e.g., dog), infants must learn how to generalize this word to multiple instances from the same category (e.g., to both golden retrievers and Chihuahuas). How do infants accomplish this feat? The overarching goal of this project is to broaden our understanding of how infants learn the visual categories that words refer to. Specifically, this project tests whether novel labels serve as social cues to help infants form more specific representations of new visual categories. To accomplish this, the project will fuse techniques from vision science and word learning to examine the kind of visual information infants map to novel words (vs. novel tones) during short eye-tracking experiments. Furthermore, the project will model infants' behaviors using computational models of word learning. Thus, this project will paint a detailed picture of the visual information that infants map to novel words. The outcome of this proposal could have implications for children with autism spectrum disorders (ASD), who tend to have difficulty processing social cues during word learning. Eventually, performance on these short eye-tracking tasks could serve as assessments for infants at-risk for ASD and as outcome measures in ASD intervention studies. Finally, this project will also create new datasets and tools that will be made publicly available using the Open Science Framework.How do infants generalize from individual instances to learn the visual category that a word refers to? Perhaps language plays a role in binding instances together. Prior work suggests that when infants hear a novel label (versus a novel tone), they expect to see objects from a new visual category. However, if infants treat novel labels as strong social cues that they should form an object category, Bayesian models of word learning predict that novel labels should also help infants form a more specific representation of this category. This project tests this hypothesis by combining eye-tracking behavioral experiments, computational models of early visual processing, and Bayesian models of word learning. First, the project will examine if labels help infants form tighter visual representations of novel categories with respect to other, novel objects. Second, the project will examine if labels change the kinds of visual features (e.g., texture vs. 3D form) infants use to represent novel categories, leveraging both texture synthesis models and 3D printing techniques. Third, the project will examine if an optimal model of Bayesian word learning predicts the behavioral results. This project takes a novel approach to studying early word learning by incorporating modern tools from visual perception, and thus makes new connections between the fields of language acquisition and visual cognitive development while furthering our understanding of both.