Review
Category learning and the memory systems debate

https://doi.org/10.1016/j.neubiorev.2007.07.007Get rights and content

Abstract

A substantial and growing body of evidence from cognitive neuroscience supports the concept of multiple memory systems (MMS). However, the existence of multiple systems has been questioned by theorists who instead propose that dissociations can be accounted for within a single memory system. We present convergent evidence from neuroimaging and neuropsychological studies of category learning in favor of the existence of MMS for category learning and declarative knowledge. Whereas single-system theorists have argued that their approach is more parsimonious because it only postulates a single form of memory representation, we show that the MMS approach is superior in its ability to account for a broad range of data from psychology and neuroscience.

Section snippets

Neuropsychological dissociations

Early studies of intact procedural learning in amnesia focused on the learning of motor skills (e.g., Brooks and Baddeley, 1976; Corkin, 1968), perceptual skills (e.g., mirror reading: Cohen and Squire, 1980; Martone et al., 1984), and cognitive skills (e.g., Tower of Hanoi: Cohen et al., 1985). In each of these cases, dissociations were found between skill learning and explicit memory for the study episode. However, these tasks did not generally make strong contact with models of learning and

Behavioral dissociations

If classification learning does not require declarative memory, then it should be possible to dissociate the two in normal subjects by manipulations that affect declarative but not procedural memory systems. In a recent study (Foerde et al., 2007), we examined whether the presence of a secondary task during learning would affect declarative memory more than category learning. Participants learned the PCT either alone or while additionally performing a secondary tonecounting task. Inclusion of

Neuroimaging

Although studies of patients with brain damage may provide evidence regarding the necessity of particular brain systems for particular cognitive functions, they can only provide limited insight into the neural architecture of normal behavior. In particular, if memory systems are not completely independent then their function in the normal individual will differ. Neuroimaging methods such as functional magnetic resonance imaging (fMRI) can fill this gap, by elucidating the dynamics of neural

Single-system approaches

Although the foregoing data seem to compel the notion of MMS, a number of theorists have posited that dissociations between memory phenomena can be accounted for using a single memory system. Their main argument has relied on the use of simple computational models to show that the behavioral data demonstrating dissociations can be fit equally well by models that posit just a single memory mechanism. For example, Nosofsky and Zaki (1998) showed that the dissociation between perceptual category

Can single-system models account for the data?

The modeling and empirical results presented by the single-system theorists have provoked important and useful debate within the field and have highlighted both the need for stronger experimental control over memory phenomena and the limitations of dissociation logic. We do not dispute the fact that single-system models can often account for particular dissociations, or that dissociations themselves do not compel MMS (cf. Poldrack et al., 1999b). However, we will argue here that although the

Proliferation of memory systems or parameters?

Opponents of MMS views have often posed a reductio ad absurdum in which memory systems proliferate to the point where every single dissociation is explained by a different memory system; this argument has appeared in the debate over systems versus processing theories of amnesia in the 1990s (e.g., Roediger et al., 1990) and the more recent debate over single versus multiple system models of category learning (e.g., Palmeri and Flanery, 2002). Further, it is suggested that the single-system

Neural models and the single-system approach

The models proposed by single-system theorists to account for memory dissociations have generally been simple mathematical or computational models that do not attempt to model the known neural circuitry underlying the behaviors being modeled. In contrast, there is a growing move toward models that are directly informed by the neurocomputational architecture of the relevant neural systems identified by cognitive neuroscientists, taking advantage of the recent explosion in computational

Conclusions

The study of category learning using a cognitive neuroscience approach has provided substantial insights into the cognitive and neural systems that support multiple forms of learning, yet some theorists continue to insist on the viability of a single memory system. We have argued here that, contrary to the claims of single-system theorists, the single-system approach lacks parsimony because it fails to successfully explain a large body of data from neuropsychological, neuroimaging, and animal

Acknowledgments

The work outlined here was supported by grants from the National Science Foundation (BCS-0223843) and Whitehall Foundation to R. Poldrack and by a National Science Foundation Graduate Fellowship to K. Foerde.

References (52)

  • R.A. Poldrack

    Can cognitive processes be inferred from neuroimaging data?

    Trends in Cognitive Sciences

    (2006)
  • R.A. Poldrack et al.

    Competition among multiple memory systems: converging evidence from animal and human brain studies

    Neuropsychologia

    (2003)
  • R.A. Poldrack et al.

    How do memory systems interact? Evidence from human classification learning

    Neurobiology of Learning and Memory

    (2004)
  • J.N. Reynolds et al.

    Dopamine-dependent plasticity of corticostriatal synapses

    Neural Networks

    (2002)
  • J.G. Smith et al.

    When artificial grammar acquisition in Parkinson's disease is impaired: the case of learning via trial-by-trial feedback

    Brain Research

    (2006)
  • D.C. Van Essen

    A population-average, landmark- and surface-based (PALS) atlas of human cerebral cortex

    Neuroimage

    (2005)
  • A.R. Aron et al.

    Human midbrain sensitivity to cognitive feedback and uncertainty during classification learning

    Journal Of Neurophysiology

    (2004)
  • N.J. Cohen et al.

    Preserved learning and retention of pattern-analyzing skill in amnesia: Dissociation of knowing how and knowing that

    Science

    (1980)
  • N.J. Cohen et al.

    Different memory systems underlying acquisition of procedural and declarative knowledge

    Annals of the New York Academy of Sciences

    (1985)
  • K. Foerde et al.

    Modulation of competing memory systems by distraction

    Proceedings of the National Academy of Sciences USA

    (2006)
  • K. Foerde et al.

    Secondary task effects on classification earning

    Memory & Cognition

    (2007)
  • M.J. Frank et al.

    By carrot or by stick: cognitive reinforcement learning in parkinsonism

    Science

    (2004)
  • M.A. Gluck et al.

    From conditioning to category learning: an adaptive network model

    Journal of Experimental Psychology General

    (1988)
  • M.A. Gluck et al.

    Hippocampal mediation of stimulus representation: a computational theory

    Hippocampus

    (1993)
  • A. Kinder et al.

    Neuropsychological dissociations between priming and recognition: a single-system connectionist account

    Psychological Review

    (2003)
  • B.J. Knowlton et al.

    The learning of categories: parallel brain systems for item memory and category knowledge

    Science

    (1993)
  • Cited by (142)

    • Information overload, stress, and emergency managerial thinking

      2020, International Journal of Disaster Risk Reduction
    • Generalization and the hippocampus: More than one story?

      2020, Neurobiology of Learning and Memory
    View all citing articles on Scopus
    View full text