Representation and Computation in Cognitive Models

Abstract

One of the central issues in cognitive science is the nature of human representations. We argue that symbolic representations are essential for capturing human cognitive capabilities. We start by examining some common misconceptions found in discussions of representations and models. Next we examine evidence that symbolic representations are essential for capturing human cognitive capabilities, drawing on the analogy literature. Then we examine fundamental limitations of feature vectors and other distributed representations that, despite their recent successes on various practical problems, suggest that they are insufficient to capture many aspects of human cognition. After that, we describe the implications for cognitive architecture of our view that analogy is central, and we speculate on roles for hybrid approaches. We close with an analogy that might help bridge the gap.

Language as cognitive tool kit: How language supports relational thought.

“…What kinds of learning processes might support relational learning? Clearly, simple associative processes are not sufficient (Forbus, Liang, & Rabkina, in press; Hummel, 2010; A. B. Markman, 1999).…”

Modeling Misretrieval and Feature Substitution in Agreement Attraction: A Computational Evaluation

“…The deep learning approach traditionally focuses on practical utility rather than the modeling of human psychological processes (Lakretz et al, 2021). Furthermore, deep learning models are, to some extent, "black boxes," in the sense that their behavior is not necessarily human-interpretable (Du, Liu, & Hu, 2019;Forbus, Liang, & Rabkina, 2017). However, recent studies in this domain have begun to link model behavior to plausible psychological mechanisms in humans and, possibly, human brain dynamics (Lakretz et al, 2019(Lakretz et al, , 2021Linzen & Baroni, 2021;Merkx & Frank, 2021;Ryu & Lewis, 2021).…”

Building Bridges between AI and Cognitive Psychology

“…Despite these successes Forbus, Liang, and Rabkina (2017) argued that the connectionist approach has limitations as a model of human reasoning. One limitation is that this approach requires massive amounts of data to learn-far more than required by people.…”

“…A third limitation is that it is not always apparent what is being learned in distributed representations. These limitations indicate that symbolic representations should play a central role in efforts to explain human cognition, particularly those showing structural alignments (Forbus, Liang, and Rabkina 2017).…”