Language Processing as Boundedly Optimal Control of Memory, Perception, and Action

语言处理作为记忆、感知和行动的有限最优控制

• 批准号: 1152819
• 负责人: Richard Lewis
• 金额: $ 40.13万
• 依托单位: Regents of the University of Michigan - Ann Arbor
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2012
• 资助国家: 美国
• 起止时间: 2012-05-01 至 2016-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1152819&HistoricalAwards=false
• 关键词: Language; Processing; Boundedly; Optimal; Control

Language comprehension, whether listening or reading, is a rapid skill that requires the coordination of perception (e.g., recognizing printed words on a page), action (e.g., controlling the movement of the eyes), memory (e.g. remembering what came in earlier parts of a sentence to relate them to later parts), and linguistic knowledge (e.g., using the order of words to constrain possible relations among them). Understanding how these processes are combined on a moment-to-moment basis has been a major challenge for cognitive science. This project will develop and test a new theory of language comprehension from an adaptive perspective. The hypothesis is that language processing may be understood as the coordinated, adaptive control of both internal (and limited) cognitive processes and external perceptual-motor actions so as to maximize specific task goals. To apply this idea to language, the investigators will formalize and implement optimal control models of limited Bayesian perception and memory for different reading tasks. These Bounded Optimal Control models will generate predictions that will be tested by monitoring the eye-movements and responses of humans.Bounded Optimal Control is a new approach to language, thought, and action, with possibly major long-term implications for how we understand and address individual differences and deficits in reading ability. It represents a shift in scientific focus from seeking to describe the perceptual and cognitive strategies that the mind and brain employ during reading, to specifying clearly the problems for which these strategies are the solutions. There is no canonical strategy because there is no canonical problem. Thus, the best cognitive and perceptual strategies are determined by specific task goals and individual constraints. This has the potential to change our approach to reading deficits: rather than asking how an individual's reading strategies deviate from normal, it is possible to ask instead whether those strategies are the best possible strategies, taking into account both task goals and idiosyncratic cognitive constraints. Revealing the true nature of the gaps between actual behavior and "bounded optimal" behavior is an important step in understanding how to go about closing them.

语言理解，无论是听力还是阅读，都是一种需要协调感知的快速技能（例如，识别页面上的印刷字），动作（例如，控制眼睛的运动）、记忆（例如，记住句子的前面部分中出现的内容以将它们与后面的部分联系起来）和语言知识（例如，使用单词的顺序来约束它们之间可能的关系）。理解这些过程是如何在每时每刻的基础上结合起来的，一直是认知科学的一个重大挑战。 这个项目将从适应性的角度发展和测试一个新的语言理解理论。该假说认为，语言处理可以理解为内部（和有限的）认知过程和外部感知运动动作的协调，自适应控制，以最大限度地提高特定的任务目标。为了将这一想法应用于语言，研究人员将形式化并实现不同阅读任务的有限贝叶斯感知和记忆的最佳控制模型。这些有界最优控制模型将产生预测，并通过监测人类的眼球运动和反应进行测试。有界最优控制是一种新的语言、思维和行动方法，可能对我们如何理解和解决阅读能力的个体差异和缺陷产生重大的长期影响。它代表了科学焦点的转变，从试图描述思维和大脑在阅读过程中使用的感知和认知策略，到明确指出这些策略是解决问题的方法。没有规范策略，因为不存在规范问题。因此，最佳的认知和感知策略是由特定的任务目标和个人的约束条件决定的。这有可能改变我们对阅读缺陷的态度：而不是问一个人的阅读策略如何偏离正常，而是可以问这些策略是否是最好的策略，同时考虑任务目标和特殊的认知限制。 揭示实际行为和“有界最优”行为之间差距的真实本质，是理解如何缩小差距的重要一步。