CAREER: Recognition-Memory Modeling: Testing Foundations and Extending Boundaries

近几十年来，心理学家使用正式的计算模型，在理解人们的认知记忆判断(例如，我记得以前见过这个人)及其相关特征(例如，他们的准确性，归因于记忆的主观自信)的认知过程方面取得了相当大的进展。这些正式的模型为研究人员提供了测量不同认知过程的相对贡献的方法(例如，“熟悉”和“情景回忆”)，回答关于他们一生中发展的问题的方法(例如，儿童、年轻人和老年人)，并允许跨不同临床人群(例如，阿尔茨海默病患者)进行比较。正式的识别记忆模型还使研究人员能够自信地解决与社会相关的问题，例如对警察部门使用的不同目击者识别程序进行批判性评估。这项研究解决了当前的问题，即多个候选模型可能对相同的数据提供不同的表征(例如，一些候选模型假设识别判断由单个助记过程驱动，而另一些候选模型假设由两个或更多个助记过程驱动)。虽然尚未解决，但这个问题阻碍了研究人员和实践者拥有适当验证的工具来详细描述人们的助记过程。目前工作的一个关键方面是所应用方法的新颖性：该团队密切协调实验设计和数学证明，以揭示不同模型预测的“行为特征模式”。这些研究的经验结果允许团队通过解决以下问题来直接测试模型：1)助记信息、信心判断和反应偏差之间的关系，2)助记信息的表示方式(绝对与相对)，3)助记信息如何准确地支持先前的相遇(“我以前见过X吗？”)与上下文助记信息有关(“我之前在哪里看到了X？我是如何看到X的？”)，以及4)需要多少不同的提取过程才能充分表征识别判断(例如，我们是否需要假设单独的“情景回忆”过程？)。最主要的结果是可行的候选模型的数量急剧减少，并朝着识别记忆的单一验证账户趋同，从而使识别的正式建模在研究和应用环境中都发挥了其全部潜力。反过来，这项工作的教育部分将开发心理学基础上亟需的课程作为其目标，并建立一个关于形式建模的本科生培训计划，明确旨在增加历史上代表性不足的群体成员和更广泛的美国国民的研究生水平的代表性。这项工作共同解决了一些悬而未决的研究问题，并增加了美国大学生的数量，他们有能力使用最先进的正式方法来应对未来的科学挑战。不同的识别记忆模型目前正被用于研究和应用环境，以表征人们记忆判断背后的认知过程。这些模型--以信号检测或高阈值框架为基础--在它们假设的认知过程的性质和数量方面有所不同。多个候选模型的存在带来了一个问题，即相同的数据可能会以多种不兼容的方式进行解释。本研究对这些不同的模型进行了批判性的比较，目的是获得一个经过验证的帐户。这一目标是通过阐述实验设计，收集强迫选择和排序判断以及与所有候选模型(例如，Block-Marschak和Tversky-Sattath不等式)对话的正式结果来实现的，以便获得需要最少辅助假设的特权试验场。在实践中，这意味着可以把研究人员传统上使用的通用模型比较方法(模型拟合+复杂性惩罚)放在一边，重新将精力集中在特定的模型预测上，这些预测可以使用顺序约束推理方法在数据水平上进行直接测试。这里产生的测试结果能够解决一些关键问题，例如：1)置信度是直接的、无噪音的潜在强度映射吗？2)潜在强度值是用似然比表示的吗？3)为了充分描述上下文信息的检索或从高度相似的箔片中区分研究项目，是否需要(除了潜在强度之外)“情节回忆”过程？这些答案可以极大地减少识别记忆的候选模型集，并将导致为实验程序开发强有力的认知心理测量工具，例如在发展和神经认知研究中广泛使用的记忆相似性任务。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

In recent decades, psychologists have used formal, computational models to make considerable advances in understanding the cognitive processes underlying people’s recognition memory judgements (e.g., “I remember seeing this person before”) and their associated characteristics (e.g., their accuracy, the subjective confidence ascribed to the memory). These formal models provide researchers with ways to measure the relative contribution of different cognitive processes (e.g., “familiarity” versus “episodic recollection”), ways to answer questions regarding their development across the lifespan (e.g., children vs. young adults vs. older adults), and allow for comparison across different clinical populations (e.g., Alzheimer patients). Formal models of recognition memory also enable researchers to confidently tackle socially-relevant issues, such as critically evaluating different eyewitness identification procedures used by police departments. The proposed research addresses the present problem that multiple candidate models may offer alternative characterizations of the same data (e.g., some candidate models postulate that recognition judgments are driven by a single mnemonic process whereas others postulate two or more). While unresolved, this issue stands in the way of researchers and practitioners having properly validated tools for characterizing people’s mnemonic processes in detail. A key aspect of the present work is the novelty of the methods applied: the team closely coordinates experimental designs and mathematical proofs in order to reveal “behavioral signature patterns” predicted by the different models. The empirical results from these studies allow the team to directly test models by addressing questions regarding: 1) the relationship between mnemonic information, confidence judgments, and response bias, 2) the way mnemonic information is represented (in absolute versus relative terms), 3) how exactly mnemonic information supporting a previous encounter (“did I see X before?”) relates to contextual mnemonic information (“where and how did I see X before?”), and 4) how many distinct retrieval processes are necessary to adequately characterize recognition judgments (e.g., do we need to postulate separate “episodic recollection” processes?). The overarching result is a drastic reduction in the number of viable candidate models and a convergence towards a single validated account of recognition memory that brings the formal modeling of recognition to its full potential in both research and applied settings. In turn, the educational component of this work establishes as its goals the development of much-needed coursework on the foundations of psychological science as well as an undergraduate training program on formal modeling that is explicitly targeted at increasing the graduate-level representation of members of historically underrepresented groups and US nationals more broadly. Together, the work resolves a number of open research problems and increases the number of US college students equipped to take on the scientific challenges of tomorrow using state-of-the-art formal methods.Different models of recognition memory are currently being used in research and applied settings to characterize the cognitive processes behind people’s memory judgments. These models – couched in Signal Detection or High-Threshold frameworks – differ in terms of the nature and number of cognitive processes that they postulate. The existence of multiple candidate models creates a problem in that the same data can be interpreted in multiple, incompatible ways. The present research critically compares these different models with goal of obtaining a single validated account. This goal is achieved by articulating experimental designs collecting forced-choice and ranking judgments along with formal results that speak to all candidate models (e.g., Block-Marschak and Tversky-Sattath inequalities) in order to obtain privileged testing grounds that require minimal auxiliary assumptions. In practice, this means that it is possible to set aside the generic model-comparison methods (model fit + complexity penalty) traditionally used by researchers up to this point and re-focus efforts on specific model predictions that can be directly tested at the level of the data using order-constrained inferential methods. The test results produced here are able to address a number of key questions, such as: 1) are confidence ratings direct, noiseless mappings of latent strengths? 2) are latent-strength values represented in terms of likelihood ratios? 3) are “episodic recollection” processes necessary (in addition to latent strengths) to adequately describe the retrieval of contextual information or the discrimination of studied items from highly similar foils? These answers can drastically reduce the set of candidate models of recognition memory, and will lead to the development of strongly-validated cognitive psychometric tools for experimental procedures, such as the Memory Similarity Task, which is widely used in developmental and neurocognitive research.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.