Differential hippocampal and prefrontal-striatal contributions to instance-based and rule-based learning
Introduction
Learning regularities across multiple episodes is a core cognitive ability. Controversy currently surrounds whether humans learn the surface structure of regular input pattern based on the superficial similarity between learning instances or whether humans acquire abstract rule knowledge (cf. Pothos, 2005, Shanks and St. John, 1994, Shanks, 1995). Three main learning tasks have been used extensively in experimental psychology: artificial grammar learning, category learning, and sequence learning tasks. Results of several studies provide evidence for the notion that learning in these tasks is partly based on the knowledge of (fragments of) learning instances (Nosofsky, 1986, Perruchet, 1994). In contrast, rule-based accounts assume that subjects acquire a set of abstract rules, defining an input pattern as grammatical, as a category member, or as a regular sequence, respectively (Ashby and Perrin, 1988, Reber, 1989). Alternative views posit that learning is subserved by both instance-based and rule-based processes (Dominey et al., 1998, Erickson and Kruschke, 1998, Knowlton and Squire, 1996, Meulemans and Van der Linden, 1997, Shanks and St. John, 1994, Shanks, 1995).
Tightly coupled with the debate about instance-based vs. rule-based learning is the question which brain structures might subserve either mechanism. However, only a few studies examined the neural correlates of instance-based vs. rule-based learning. In an artificial grammar learning study, Fletcher et al. (1999) demonstrated that learning within experimental blocks is mediated by the right lateral prefrontal cortex (PFC), whereas the left lateral PFC subserves learning across the entire experiment. The authors argue that within-block learning effects mainly rely on explicit retrieval of individual items based on the surface structure of items, i.e., instance-based learning. Further, the authors propose that, in contrast, across-block learning is based on the acquisition of abstract rule knowledge. In another study (Strange et al., 2001), subjects were required to learn rules which define the category membership of four-letter strings. Changes in abstract rules were associated with an increase of anterior PFC activity, whereas hippocampal activation was modulated by the introduction of new instances. Recent fMRI and patient studies revealed that the hippocampus is associated with instance-based learning (Lieberman et al., 2004, Opitz and Friederici, 2004), whereas the basal ganglia (Lieberman et al., 2004, Teichmann et al., 2005) and the lateral PFC (Opitz and Friederici, 2004) subserve rule-based learning. However, a computational model has implicated the basal ganglia in instance-based learning and the lateral PFC in rule-based learning, respectively (Dominey et al., 1998). Apart from the Dominey et al. (1998) and the Teichmann et al. (2005) study (which used letter sequences and arithmetic operations, respectively), most of the studies examined artificial grammar learning. Taken together, these studies converge to suggest that the lateral PFC subserves rule-based learning, However, the exact location of lateral PFC activation varies between studies, depending on stimulus properties and task requirements. Furthermore, most of the studies point to an involvement of the basal ganglia in rule-based learning and a specific role of the hippocampus in instance-based learning.
In the present fMRI study, we investigated this issue by adopting the experimental logic of the Fletcher et al. (1999) study. Using a slightly modified version of a recently developed learning paradigm (Doeller et al., 2005), we aimed at investigating instance-based vs. rule-based learning in a non-linguistic domain. In two experiments, subjects were required to memorize six object–position conjunctions in each trial of several experimental blocks. Both experiments included two conditions, a learning condition and a control condition. In the learning condition, either objects (Experiment 1) or positions (Experiment 2) were held constant in each trial of the experimental blocks (systematic repetition of specific objects or positions), by this introducing regularities across episodes, whereas in the control condition, object–position conjunctions were trial-unique (no systematic repetition of specific objects or positions). Hence, we define learning as the successful adoption of these invariant objects and positions in object–position conjunctions across trials, i.e., object or spatial regularities. We expected that the introduction of invariant objects and positions entails increased task performance across trials within-blocks of the learning condition. These within-block learning effects were supposed to reflect mainly instance-based learning, since subjects' judgments could rely solely on the similarity between study and test items. In contrast to our recent fMRI study (Doeller et al., 2005), trials were blocked by condition to minimize the probability that subjects based their judgement on a common strategy for both conditions. Critically, the set of invariant objects and positions changed from block to block in the learning condition. By this, subjects were able to transfer their knowledge about regularities to new instances when a new block starts. This transfer has been implicated as a possible experimental test to dissociate instance-based and rule-based learning (e.g., Gomez and Schvaneveldt, 1994, Mathews et al., 1989). If subjects acquire abstract rule knowledge, they should be able to transfer this knowledge to new instances. Thus, a performance modulation across learning blocks was assumed to be a main index of rule-based learning. Based on the above reviewed literature and our previous results (Doeller et al., 2005), we predicted a learning-related decrease of hippocampal and an increase of prefrontal-striatal activation as a function of learning within-blocks. In contrast, we expected prefrontal-striatal – but not hippocampal – involvement during across-block learning (cf. Fletcher et al., 1999, Lieberman et al., 2004, Opitz and Friederici, 2004).
Section snippets
Subjects
Twenty-four subjects participated in the study, 12 subjects in Experiment 1 (aged 22–33, mean age 24.6 years, 6 females) and 12 subjects in Experiment 2 (aged 22–29, mean age 24.3 years, 5 females). All subjects were right-handed with normal or corrected-to-normal vision and were paid for participating. Informed consent was obtained before scanning. All participants reported to be in good health with no history of neurological disease.
Stimuli, task, and design
Sixteen pictures denoting real-life objects were used as
Behavioral results: within-block learning
In a first step, within-block learning effects were analyzed (Fig. 3). In both experiments, performance (adjusted Pr values) increased across trials in the learning condition, but not in the control condition. To examine changes of performance during the time course of the blocks, comparisons between the first (trials 1–9) and last (trials 28–36) quarter of trials within-blocks were conducted. In Experiment 1 (Fig. 3A, left panel), mean adjusted Pr values increased from the first to the last
Discussion
The present study aimed at specifying the neural correlates of instance-based and rule-based learning. The first learning-process should be mainly reflected in a performance increase within experimental blocks (within-block learning), whereas the latter learning process should predominantly entail a performance increase across learning blocks (across-block learning). In Experiment 1 and Experiment 2, subjects had to learn object regularities and spatial regularities, respectively. Our
Acknowledgments
This work was supported by grants from the German Research Foundation, DFG, Research Group FOR-448. The authors thank Lea Meyer and Henning Loebbecke for their assistance during data acquisition and Ali Jeewajee for helpful comments on a previous version of the manuscript.
References (75)
- et al.
Modeling geometric deformations in EPI time series
NeuroImage
(2001) - et al.
Hippocampus, cortex, and basal ganglia: insights from computational models of complementary learning systems
Neurobiol. Learn. Mem.
(2004) - et al.
Encoding strategies dissociate prefrontal activity from working memory demand
Neuron
(2003) - et al.
Event-related fMRI: characterizing differential responses
NeuroImage
(1998) - et al.
Classical and Bayesian inference in neuroimaging: theory
NeuroImage
(2002) - et al.
Hemispheric asymmetries of memory: the HERA model revisited
Trends Cogn. Sci.
(2003) - et al.
Temporal prediction errors in a passive learning task activate human striatum
Neuron
(2003) - et al.
Temporal difference models and reward-related learning in the human brain
Neuron
(2003) - et al.
Interactions of the hippocampal system and the prefrontal cortex in learning language-like rules
NeuroImage
(2003) - et al.
Doubts about double dissociations between short- and long-term memory
Trends Cogn. Sci.
(2005)
Medial temporal lobe activity associated with active maintenance of novel information
Neuron
An fMRI study of the medial temporal lobe in implicit and explicit sequence learning
Neuron
Getting formal with dopamine and reward
Neuron
Information theory, novelty and hippocampal responses: unpredicted or unpredictable?
Neural Netw.
Toward a unified theory of similarity and recognition
Psychol. Rev.
A computational model of prefrontal control in free recall: strategic memory use in the California verbal memory task
J. Cogn. Neurosci.
On the control of control: the role of dopamine in regulating prefrontal function and working memory
Recognition memory: what are the roles of the perirhinal cortex and hippocampus?
Nat. Rev., Neurosci.
Prefrontal regions involved in keeping information in and out of mind
Brain
Hippocampal system and declarative (relational) memory: summarizing the data from functional neuroimaging studies
Hippocampus
Hippocampal contributions to episodic encoding: insights from relational and item-based learning
J. Neurophysiol.
Learning and selective attention
Nat. Neurosci.
Cognitive and neural mechanisms subserving regularity learning
Prefrontal–hippocampal dynamics involved in learning regularities across episodes
Cereb. Cortex
Dissociable processes for learning the surface structure and abstract structure of sensorimotor sequences
J. Cogn. Neurosci.
An adaptive coding model of neural function in prefrontal cortex
Nat. Rev., Neurosci.
A cortical–hippocampal system for declarative memory
Nat. Rev., Neurosci.
Rules and exemplars in category learning
J. Exp. Psychol. Gen.
The effect of context on discrimination and bias in recognition memory for pictures and words
Mem. Cogn.
Learning-related neuronal responses in prefrontal cortex studied with functional neuroimaging
Cereb. Cortex
Responses of human frontal cortex to surprising events are predicted by formal associative learning theory
Nat. Neurosci.
On the benefits of not trying: brain activity and connectivity reflecting the interactions of explicit and implicit sequence learning
Cereb. Cortex
Interaction between frontal cortex and basal ganglia in working memory: a computational model
Cogn. Affect. Behav. Neurosci.
Categorial representation of visual stimuli in the primate prefrontal cortex
Science
Cerebral specialization and interhemispheric communication: does the corpus callosum enable the human condition?
Brain
Regional and cellular fractionation of working memory
Proc. Natl. Acad. Sci. U. S. A.
What is learned from artificial grammars? Transfer tests of simple association
J. Exp. Psychol., Learn. Mem. Cogn.
Cited by (13)
Neurocomputational Dynamics of Sequence Learning
2018, NeuronCitation Excerpt :Note that activity in the hippocampus, another area important for storing sequential representations (Bornstein and Daw, 2012; Vilberg and Rugg, 2012), did not differentiate between the pattern lengths at the end of the sequence (Figure 4E), and thus it is unlikely to be involved in retrieval. Another potential candidate region for storing sequential information is the left IFG (Doeller et al., 2006; van der Graaf et al., 2006; Huettel et al., 2002); however, some prior work has demonstrated that activity in this region could be simply related to the use of verbal encoding strategies (Van Opstal et al., 2009). We have found it to be primarily correlated with Bayesian surprise, which could be an indication that the IFG encodes sequential violations.
Multiple brain networks underpinning word learning from fluent speech revealed by independent component analysis
2015, NeuroImageCitation Excerpt :Its engagement during word form learning, even when there is no semantic component, agrees with previous results indicating a prominent role of this ventral stream when support to the dorsal stream is needed (López-Barroso et al., 2011; Saur et al., 2010). In addition, the caudate nucleus forms part of this network, which agrees with the importance of this area for the concatenation of sequences forming a chunk (Koechlin and Jubault, 2006) in artificial language learning from visual or auditory sequences (Bahlmann et al., 2008; De Diego-Balaguer et al., 2008; Doeller et al., 2006; Lieberman, 2000). Nevertheless, the limited replication of this network, marginally significant in the main cohort, may go in the direction of a secondary implication of this network compared to the dorsal networks previously described.
Neural binding mechanisms in learning and memory
2010, Neuroscience and Biobehavioral ReviewsCitation Excerpt :The lateral PFC has been posited to subserve such reorganization operations during memory encoding (Fletcher and Henson, 2001; Simons and Spiers, 2003). Doeller et al. (2006) further explored whether the representational bindings are built on the surface structure of a regular input pattern based on the superficial similarity between learning instances or on abstract rules. Capitalizing on the successful transfer of knowledge about regularities to new instances as a main index of rule-based learning (Mathews et al., 1989), an increase of activity in the dorsolateral PFC associated with knowledge transfer was observed.
Altered effective connectivity and anomalous anatomy in the basal ganglia-thalamocortical circuit of stuttering speakers
2010, CortexCitation Excerpt :In the present study, we used the structural equation modeling (SEM) technique to explore how the nodes of the BGTC interacted with one another in stuttering and non-stuttering speakers (see Fig. 1). Since its introduction by McIntosh et al. (1994), neural connectivity analysis has been used in a wide range of functional MRI studies (e.g., Büchel and Friston, 1997; Doeller et al., 2006; Fu et al., 2006; Goncalves et al., 2001; Honey et al., 2003; Iidaka et al., 2006; Schlosser et al., 2006; Zhuang et al., 2005). Several multivariate techniques have been employed to investigate effective connectivity.
Neural and Cognitive Plasticity: From Maps to Minds
2008, Psychological BulletinExperimental Studies on Trading Decision
2019, SSRN