When cues converge: multiple regularities in language acquisition

描述（由申请人提供）：本研究项目的目的是利用行为和计算方法来更好地了解婴儿在学习语言时如何使用多种规律。许多研究人员假设婴儿使用统计学习机制来提取这些规律。然而，这项工作主要集中在婴儿利用单一规律的能力上。在他们的自然环境中，婴儿接受的输入包含许多重叠的不同一致性的规则。很少有研究关注婴儿如何处理这种类型的输入。为了利用自然输入中可用的线索，婴儿必须能够利用这些重叠的概率规律。这项资助的第一个目的是探索婴儿如何使用多种规则在流利的言语中找到单词。到9个月大的时候，婴儿已经能够熟练地使用词汇重音和顺序统计来分词。然而，先前的研究主要集中在婴儿如何在孤立的情况下使用这些规则，或者线索的权重如何随着发育而变化。我们假设，当这些线索是概率和重叠的时候，婴儿可以使用这些线索进行分词，就像在英语等自然语言中一样。第二个目的是研究多重规则对单词学习的影响。随着孩子词汇量的增长，他们所知道的标签和指代物之间以及标签和指代物之间的映射中出现了规律。这些规律可能会影响新词的习得。然而，它们在随后的单词学习中的潜在作用尚未被仔细探索。使用计算模型，我们可以通过仔细控制词汇表中存在的规律并探索不同类型的规律如何影响单词学习来检查这些问题。这些模型可以让我们更深入地理解这些处理机制，并做出新的预测，然后在行为实验中进行测试。

DESCRIPTION (provided by applicant): The goal of this research project is to employ behavioral and computational methods to better understand how infants use multiple regularities when learning language. Many researchers have posited that infants use statistical learning mechanisms to extract these regularities. However, the majority of this work focuses on infants' abilities to exploit a single regularity. In their natural environment, infants receive input that contains many overlapping regularities of varying consistency. Little research has focused on how infants process this type of input. In order to make use of the cues available in natural input, infants must be able to exploit these overlapping probabilistic regularities. The first aim of this grant is to explore how infants use multiple regularities to find words in fluent speech. By 9 months, infants are able to robustly use both lexical stress and sequential statistics to segment words. However, prior research has focused on how infants use these regularities in isolation, or how weighting of the cues changes developmentally. We hypothesize that infants can use these cues for word segmentation when they are probabilistic and overlapping, as they are in natural languages such as English. The second aim is to investigate the effect of multiple regularities on word learning. As a child's vocabulary grows, regularities arise among the labels and referents they know, as well as within the mappings between labels and referents. These regularities may affect the acquisition of new words. However, their potential role in subsequent word learning has not been carefully explored. Using computational models, we can examine these issues by carefully controlling regularities that exist within a vocabulary and by exploring how different types of regularities may affect word learning. These models can lead to a deeper understanding of these processing mechanisms and to novel predictions, which can then be tested in behavioral experiments. Public Health Interests: The goal of this research is to better understand how typically developing infants learn language - in particular, how they use the many patterns that exist in natural languages to do so. The ability to make use of multiple regularities likely affects infants' skills at language learning. By better understanding how this process unfolds in typically developing infants, researchers will be able to investigate how children with language delays and disabilities may differ in their ability to integrate multiple cues. In the future, this information will be useful for developing treatment plans.