Virtual ICCM Session II

Cognitive architectures (CAs) have been instrumental in in- tegrating a wide range of findings in cognitive science into unified theories of cognition. However, much less effort has been devoted to applying CAs to social phenomena, despite the high interdependence between cognitive and social processes in real-world scenarios (e.g., Ecker et al., 2022). We integrated social sampling theory (SST) and ACT-R to begin filling this gap. ACT-R is a modular, hybrid symbolic/sub-symbolic CA with a detailed memory system. SST describes how beliefs and behavior emerge from an interplay between individual and so- cial motivations. The component theories have complementary strengths and weaknesses: SST provides an account of social influence and comparison, but lacks a memory system to sup- port those processes, whereas the converse is true for ACT-R. In two simulations, we demonstrate that SST-ACT-R produces social influence dynamics not present in either component the- ory. Specifically, SST-ACT-R shows how private and publicly expressed beliefs may evolve through social interactions based on social influence and underlying memory mechanisms.

Analogical reasoning is a core cognitive process that involves mapping knowledge structures, and may depend on how mental representations are encoded and retrieved. Successful analogical reasoning can enable analogical transfer between a previous and new concept or problem. Theories and models were developed to explain analogical reasoning and transfer. However, challenges with interacting cognitive processes, generalization, and cognitive plausibility remain. Here, we attempt to address challenges by leveraging previous work with a cognitive analogical reasoning framework and a subsequent extension. The model starts with procedural knowledge about how do a problem solving task and learns its solution. It then "reads" and represents problem isomorphs, and initiates analogical transfer to solve them. We present results and limitations with our approach.

Many different theories of learning have been developed to account for human performance over time, often accounting for performance at an aggregate level. Understanding performance at an individual level is often more difficult because of multiple different factors—e.g., noise, strategy selection, or change in memory representation—, which are often not accounted for in simple learning theories. One approach used to explain the sudden changes in performance that are often observed at the individual level is to integrate change detection algorithms with psychological models. This research has shown that performance at the individual level can be understood not by a single continuous process but instead by segmented portions of multiple processes. Previous research has posited different explanations as to what features drive the inferences of change points. However, no paper has yet compared different explanations’ ability to explain the variance in inferred change points. In this paper, we use a simple model of learning to account for performance in a real-world data set with individuals performing multiple different games that tap into different task attributes (i.e., memory, attention, problem-solving) on the website Luminosity. We then conduct a statistical analysis to determine what drives change points in the dataset. The results here allow for better clarification as to what features are driving the inferences of change points at the individual level.

Prediction is argued to be a key factor in comprehending sentences in verb-final languages. The comprehender can predict the properties of the upcoming verb phrase using linguistic cues from the pre-verbal input. What are the constraints on prediction, beyond the ones posited by co-occurrence patterns in the language? We evaluate four models of verb prediction using data from a sentence completion study on Hindi. The model differs in their assumption of whether/how working memory constraints affect the prediction of the upcoming verb. The model comparison conclusively shows that working memory constraints do affect the prediction of the verb in Hindi. The results lead to a new insight into the underlying comprehension process: When the pre-verbal input is temporarily stored in memory, it probabilistically distorts to a non-veridical (or less accessible) memory representation, and this degraded representation of the context generates potentially faulty predictions of the upcoming verb.