Cognitive and neural basis of memory for associations and order

联想记忆(如诺特利-艾伯塔或威廉-凯特)和秩序记忆(序列，如电话号码)对于从导航到语言再到社交行为的复杂行为至关重要。理论是通过数学模型系统地组织起来的，但这些模型也分别在联想和顺序记忆方面进行了开发和测试。我们的目标是开发一个单一的模型，既可以解释标准联想和顺序任务中的记忆行为，也可以解释结合两者元素的任务中的记忆行为。在我们的方法中，通过仔细的设计，测量记忆行为(准确性和反应时间)和与成功编码相关的大脑活动的实验可以提供强大的模型测试，导致更深层次和更普遍的关系记忆统一模型。我们最近的发现表明，模型必须重新设计。我们第一个问，当人们记住一对单词时，他们记住的联系是有秩序的还是没有秩序的(是克里斯蒂-克拉克还是克拉克-克里斯蒂)？参与者可以判断秩序，即使他们不是故意为秩序而学习。然而，顺序也可能是模棱两可的，即使在记住这对项目时也是如此。这对当前的关联模型提出了挑战：矩阵模型(与Hebb规则相关)，它以明确的顺序学习关联，以及基于卷积的模型，它存储没有顺序的关联。其次，早期的模型假设记忆是一个列表(ABCD...)是由一系列协会(AB、BC、CD等)组成的。与这些不同的是，刻意省略项目间关联的“位置编码”模型现在更受青睐。我们最近的发现清楚地表明，事实上，一次回忆确实会影响下一次回忆，推翻了位置编码模型的一个主要原则，并表明关联链是序列顺序记忆的主要部分。下一代模型将同时需要项目-项目和项目-位置关联。第三，我们挑战了之前的观点，即项目记忆任务中的一些大脑活动信号反映了联想过程。相反，额部慢波和theta节奏(4-8个周期/秒)可能对联想，也许是顺序记忆特别关键。我们将修改现有的联想和顺序记忆模型，以适应这些发现，并产生新的可测试的预测。同时，我们将进行行为和脑电实验，以更好地了解联想记忆和秩序记忆之间的关系，并测试和指导进一步的模型开发。结合数学模型、行为和大脑活动方法，我们可以在基本水平上更好地了解记忆行为，潜在地识别可能成为记忆评估或增强方案的目标的大脑活动信号，并指导人工智能体中记忆的发展。

Memory for associations (pairs, such as Notley-Alberta or William-Kate) and memory for order (sequences, such as a phone number) are critical for complex behaviour, from navigation to language to social behaviour. Theories are systematically organized through mathematical models, but these models have also been developed and tested separately for association and order memory. Our aim is to develop a single model that can explain memory behaviour in both standard association and order tasks, but also tasks that combine elements of both. In our approach, with careful design, experiments measuring memory behaviour (accuracy and response time) and brain-activity related to successful encoding can provide powerful tests of models, leading to a deeper and more generalizable unified model of relational memory.Our recent findings show that models must be redesigned. We were the first to ask, when people remember a pair of words, do they remember the association with or without order (was it Christy-Clark or Clark-Christy)? Participants could judge order, even when they did not deliberately study for order. However, order can also be ambiguous, even when the pair of items is remembered. This challenges current models of association: the matrix model (related to Hebb's rule), which learns associations with unambiguous order and convolution-based models, which store associations without order. Second, early models assumed memory of a list (ABCD...) is composed of a "chain" of associations (AB, BC, CD, ...). Pitted against these, "positional-coding" models, which deliberately omit inter-item associations, are now favoured. Our recent findings clearly show that one recall does, in fact, influence the next, overturning a major tenet of positional-coding models and suggesting that associative chaining is a major part of serial-order memory. The next generation of models will need both item-item and item-position associations. Third, we challenged previous arguments that some brain-activity signals in item-memory tasks reflected associative processes. Instead, frontal slow waves and theta rhythms (4-8 cycles/second) rhythms may be specifically critical for association, and perhaps order, memory.We will modify current models of association- and order-memory to accommodate these findings and generate new testable predictions. In parallel, we will conduct behavioural and electroencephalographic experiments to better understand the relationship between memory for associations and memory for order, and to test and guide further model development.Combining mathematical models, behavioural and brain-activity methods, we can better understand memory behaviour at a basic level, potentially identify brain-activity signals that could be targets of memory assessment or enhancement protocols, and guide the development of memories in artificial agents.