Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Rate-Coding Bundle Memory: A Unified Model of Memory and Control for Language Comprehension

Teun van Gils¹, Rowan P. Sommers²; ¹Utrecht University, ²Institute of Cognitive Science, Osnabrück University, Universität Osnabrück

Presenter: Teun van Gils

Neurobiological and cognitive models often tackle individual facets of cognition, yet few successfully integrate across multiple domains. For instance, (pre-transformer) connectionist approaches explain a broad range of cognitive phenomena, including graded semantics and context sensitivity, but struggle with symbolic reasoning, compositionality and dynamic variable binding, necessary components of language comprehension (Fodor & Pylyshyn, 1988; Marcus, 1998; Lake et al., 2017; Kazanina & Poeppel, 2023). Conversely, symbolic models excel at these tasks but often lack a biologically inspired explanation of how these symbolic operations might be implemented in the brain (Do & Hasselmo, 2021). We propose a unified model, inspired by (psycho-)linguistic theory (McElree et al., 2003; Seuren, 2009), that integrates these aspects, focusing on the interplay between memory, unification and control (Baggio & Hagoort, 2011). This model, which we call Rate-Coding Bundle Memory (RCBM), is designed to be both biologically plausible and capable of addressing a variety of cognitive tasks, including those that require the incremental integration and differention of entities in linguistic descriptions of scenes. We demonstrate the model’s performance on a range of tasks, including controlled storage, memory retrieval, tracking multiple memories, and semantic inference; and show, by lesioning different components of the model, that the various memory and control components are crucial for the model’s ability to perform these tasks. Our model improves upon existing working memory models of the prefrontal cortex (Manohar et al., 2019; Fiebig et al., 2020), and shows parallels with properties of grid- and place cells during hippocampal replay (Do & Hasselmo, 2021; Kurth-Nelson et al., 2023; Kazanina & Poeppel, 2023).

Topic Area: Language & Communication

Extended Abstract: Full Text PDF