Probability learning and statistical inference in infancy and early childhood

婴幼儿怎么会学得这么快？我的研究项目的长期目标是通过全面了解婴幼儿和早期儿童的概率学习和推理机制来回答这个问题。近年来，我和我的学生、合作者和我在描述年轻学习者如何使用基本比率和抽样信息进行跨领域归纳推理方面取得了实质性进展。我的下一个资助周期的短期目标是在描述婴儿期和幼儿期的概率学习方面取得类似的进展。概率学习需要跟踪事件在空间和时间中的频率或比例。在许多概率学习研究中，一个人必须通过许多试验，预测一个事件将在空间的哪一边(左或右)发生。如果事件在70%的时间发生在一侧(您已经了解到了这一点)，您应该预测在那一侧发生的频率是多少？成年人通常会做两件事中的一件：要么最大化地选择频率最高的那一边，总是选择它，要么概率匹配频率，按照出现的比例选择每一边(70：30)。最大化具有更高的预期精确度(概率匹配的准确率为70%比58%)，因此，几十年来成人认知领域的工作已经研究了一些成年人为什么以及如何显示每种模式。儿童早期的概率学习很少被研究，尽管发展数据确实有机会对这个长期存在的问题进行权衡，尽管概率和统计学习在发展早期特别关键。拟议中的实验将阐明导致概率匹配行为与最大化行为的潜在认知过程。所有实验都采用简单的概率学习设计，孩子们在每次实验中预测物体将出现在哪一边，测量反应时间、选择和整体学习。系列1的目标是通过系统地操纵概率分布(50，70，90,100)并检查匹配与最大化，获得3-6岁儿童概率学习的全貌。系列2的目标是测试儿童在变化的环境中更新概率的能力。我会问，孩子们是否期望在人为环境与物理决定的环境中保持更多或更少的稳定性，以揭示某些行为模式是否源于对某些领域发生更大变化的预期。最后，系列3的目标是收集关于婴儿期概率学习的第一个广泛的数据集，以检查婴儿的这些潜在的认知过程。这些数据和发现将引起认知科学家的极大兴趣，因为他们将讲述概率学习中涉及的潜在认知机制，以及儿童在参与这些学习过程时对跨域环境稳定性的早期概念。

How do infants and young children learn so much so quickly? The long-term goal of my research program is to answer this question by developing a comprehensive picture of probabilistic learning and inference mechanisms in infancy and early childhood. In recent years, my students, collaborators, and I have made substantial progress in characterizing how young learners use base-rates and sampling information to make inductive inferences across domains. The short-term goal of my next grant cycle is to make similar progress in characterizing probability learning in infancy and early childhood. Probability learning entails tracking frequencies or proportions of events in space and time. In many probability learning studies, one must predict on which side of space (left or right) an event will occur, over many, many trials. If the event occurs on one side 70% of the time (and you have learned this), how often should you predict an occurrence on that side? Individual adults often do 1 of 2 things: either maximize on the frequent side, always choosing it, or probability match the frequencies, choosing each side in the proportion with which it occurs (70:30). Maximizing has greater expected accuracy (70% vs 58% for probability matching) and so decades of work in adult cognition has examined why and how some adults show each pattern. Probability learning in early childhood has scarcely been studied, despite the real opportunity for developmental data to weigh in on this longstanding question and despite the fact that probability and statistical learning are particularly critical early in development. The proposed experiments will shed light on the underlying cognitive processes that give rise to probability-matching versus maximizing behaviour. All experiments use a straightforward probability learning design, in which children predict on each trial which side an object will appear, measuring reaction time, choices, and overall learning. The objective of Series 1 is to obtain a full picture of 3-6-year-old children's probability learning by systematically manipulating probability distributions (50,70,90,100) and examining matching versus maximizing. The objective of Series 2 is to test children's ability to update probabilities in shifting environments. I will ask whether children expect more or less stability in human-generated versus physically determined environments to uncover whether some patterns of behaviour result from an expectation of greater shifting in some domains. Finally, the objective of Series 3 is to collect the first extensive dataset on probability learning in infancy to examine these underlying cognitive processes in infants. Together these data and findings will be of great interest to cognitive scientists, as they will speak to the underlying cognitive mechanisms involved in probability learning, as well as children's early conceptions of the stability of our environments across domains when engaging in these learning processes.