Attentional control driven by statistical learning

视觉环境是高度结构化的，并且在物体如何在空间中(例如，椅子往往出现在桌子旁边)和随着时间的推移(例如，黄灯之后总是红灯)方面具有丰富的规律性。视觉系统在通过统计学习过程提取这些规则方面非常有效(Fiser&Aslin，2001；Saffran等人，1996；Turk-Browne等人，2009)。最近，人们发现，人们可以自发和隐含地将注意力吸引到规则上(赵等人，2013年)。这种对规律性的注意偏差可能会使人们能够检测结构，并增强对世界稳定方面的知识的获取。本提案的目标是探讨学习如何引导注意力的分配，以及由此对感知产生什么影响。有三个具体目标。 目标1.注意偏向规则的后果是什么？ 第一个目标是考察结构化刺激的注意优先顺序如何改变这些刺激的表征，以及由此产生的知觉错误。例如，当两个对象在空间或时间上可靠地共同出现时(例如，A总是与B一起出现)，这些对象可被更相似地表示，这可导致对象的表示的合并。 目标2.学习如何改变注意力的空间尺度？ 第二个目的是考察规则如何影响注意力的空间尺度。空间阵列中的对象可靠地共同出现可能会将注意力局部地吸引到各个对象上，并将注意力从全局集合中转移开。因此，规则性可能引起对阵列中的单个对象的局部关注，从而阻碍阵列的全局处理。这可以进一步解释统计学习和总结知觉之间的干扰。 目标3.注意偏向的时间性、持久性和灵活性是什么？ 学习对注意力的调节可能是非线性的。在广泛接触规则之后，注意偏差可能会消散，甚至逆转到结构较差的信息来源。换句话说，学习后有强烈的期望可能会将注意力从结构性刺激释放到其他地方。在缺乏规律性的情况下，这种偏见可能会持续存在于先前结构化的信息之上。最后，当规则转移到不同的空间位置时，注意力可以灵活地重新分配到新位置。 拟议的研究计划将提供对统计学习如何引导注意力的全面理解。它促使人们摆脱关于注意的外源性和内源性控制的理论上的二分法，并鼓励发展一个更广泛的框架。它还表明，当获得了关于物体之间关系的知识时(在广泛接触之后)，注意力可能会脱离，并转移到其他信息上。这一过程允许探索和获取环境中的新知识。这项工作揭示了学习引导注意力分配的新方式。目前的提案还可以为人类感知提供新的见解。也就是说，物体在空间或时间上如何共同出现，可以塑造这些物体的表示形式。最后，这项研究为检验学习诱导的注意控制提供了新的实验范式，反映并促进了该领域对理解注意机制如何与学习相互作用的日益增长的兴趣。

The visual environment is highly structured and rich with regularities in terms of how objects co-occur in space (e.g., a chair tends to appear next to a table) and over time (e.g., the yellow light is always followed by the red light). The visual system is extremely efficient at extracting these regularities through the process of statistical learning (Fiser & Aslin, 2001; Saffran et al., 1996; Turk-Browne et al., 2009). Recently, it has been found that attention can be drawn spontaneously and implicitly to regularities (Zhao et al., 2013). Such attentional bias toward regularities may enable the detection of structure and enhance the acquisition of knowledge about stable aspects of the world. The goal of the current proposal is to explore how learning guides the allocation of attention and what consequences on perception are produced as a result. There are three specific aims. Aim 1. What are the consequences of the attentional bias to regularities? The first aim examines how the attentional prioritization of structured stimuli alters the representation of these stimuli, and what kind of error in perception is produced as a result. For example, when two objects reliably co-occur over space or time (e.g., A always appears with B), these objects may be represented more similarly, which may result in the merging of representations of the objects. Aim 2. How does learning alter the spatial scale of attention? The second aim examines how regularities influence the spatial scale of attention. Reliable co-occurrence of objects in a spatial array may draw attention locally to individual objects and bias attention away from the global set. Thus, regularities may induce a local scale of attention to individual objects in an array, impeding global processing of the array. This can further explain the interference between statistical learning and summary perception. Aim 3. What are the temporal dynamics, durability and flexibility of the attentional bias? The modulation on attention by learning may be non-linear. After extensive exposure to regularities, the attentional bias may dissipate or even reverse to less structured sources of information. In other words, having strong expectations after learning might release attention from structured stimuli to elsewhere. In the absence of regularities, the bias may persist over previously structured information. Finally, when regularities shift to a different spatial location, attention may be flexibly re-allocated to the new location. The proposed program of research will offer a comprehensive understanding of how attention is directed by statistical learning. It motivates a shift away from the theoretical dichotomy on exogenous and endogenous control of attention, and encourages the development of a broader framework. It also suggests that when knowledge about relationships among objects has been acquired (after extensive exposure), attention may be disengaged and shifted to other information. This process allows exploration and acquisition of new knowledge in the environment. The work reveals new ways in which learning guides the allocation of attention. The current proposal can also offer new insights on human perception. That is, how objects co-occur in space or time can shape the representations of these objects. Finally, the proposed research provides novel experimental paradigms for examining learning-induced attentional control, which reflect and facilitate the field’s growing interest in understanding how attentional mechanisms interact with learning.