Department of Psychology, University of Richmond, 28 Westhampton Way, Richmond, VA, 23173, USA.
Department of Psychological and Brain Sciences, Indiana University, Bloomington, IN, USA.
Psychon Bull Rev. 2019 Jun;26(3):993-1000. doi: 10.3758/s13423-018-1542-6.
People quickly form summary representations that capture the statistical structure in a set of simultaneously-presented objects. We present evidence that such ensemble encoding is informed not only by the presented set of objects, but also by a meta-ensemble, or prototype, that captures the structure of previously viewed stimuli. Participants viewed four objects (shaded squares in Experiment 1; emotional expressions in Experiment 2) and estimated their average by adjusting a response object. Estimates were biased toward the central value of previous stimuli, consistent with Bayesian models of how people combine hierarchical sources of information. The results suggest that an inductively learned prototype may serve as a source of prior information to adjust ensemble estimates. To the extent that real environments present statistical structure in a given moment as well as consistently over time, ensemble encoding in real-world situations ought to take advantage of both kinds of regularity.
人们会迅速形成总结性的表示，以捕捉同时呈现的一组对象中的统计结构。我们提供的证据表明，这种整体编码不仅受到呈现对象集的影响，还受到元整体或原型的影响，原型捕获了先前观看刺激的结构。参与者观看了四个对象（实验 1 中的阴影正方形；实验 2 中的情绪表达），并通过调整响应对象来估计它们的平均值。估计值偏向于先前刺激的中心值，这与人们如何组合层次信息源的贝叶斯模型一致。结果表明，一个归纳学习的原型可以作为调整整体估计的先验信息源。在某种程度上，真实环境在给定时刻以及随着时间的推移呈现统计结构，因此在真实情况下的整体编码应该利用这两种规律。
