Physics instruction induces changes in neural knowledge representation during successive stages of learning
Introduction
Just knowing what some mechanical system accomplishes is often sufficient. Sometimes it is enough to know that when you aim a fire extinguisher and squeeze the handle, a fire-suppressing fluid sprays onto the fire. But how does it work? Learning how a mechanical device works through instruction is a critical part of many jobs. Understanding the psychological and neural processes that occur during such learning can now be studied with brain imaging to reveal how new technical knowledge is built up in the brain in the course of instruction.
A main aim of this research was to show how the neural representations of specific technical knowledge change as a result of acquiring new information. Of course, there are many other types of changes in the brain following training or instruction that have been reported. There are three critical differences between prior investigation of brain changes due to learning and this research, measured here are: (1) changes in representation as opposed to structural changes; (2) changes due to instruction-based learning as opposed to training, specifically in the science domain of the physics of mechanical systems; and (3) changes in the neural representation of acquired knowledge as opposed to activation changes during a performance of a task.
For example, structural brain changes due to training have been observed in the cortical responses of multiple units in cats (Merzenich, 1975), rats (Kilgard and Merzenich, 1998), and adult monkeys (Recanzone et al., 1993). At a more molar level, learning-based changes in grey and white matter have been observed in human participants (see Fields, 2011, Lövdén et al., 2013, Thomas and Baker, 2013, Zatorre et al., 2012 for reviews). For example, gray and white matter changes were observed when people were trained in juggling (Draganski et al., 2004). Evidence that structural changes occur with training also comes from visuo-motor tasks, working memory tasks, and aerobics. Many of these studies involve tasks in which learning occurs as an effect of repeated training trials rather than being due to learning from instruction (as might occur in a classroom).
A number of tasks have shown brain-based changes in activation patterns due to training or instruction-based learning. Typically, these tasks examine brain changes in activation, in which it may be difficult to separate new representations from learned processes. Examples of tasks that examine activation in training-based learning include artificial grammars (Petersson, Folia, & Hagoort, 2012), perceptual category learning (Poldrack et al., 2001), and motor learning (Toni, Krams, Turner, & Passingham, 1998). Instruction-based learning tasks such as algebra (Anderson et al., 2012 have also resulted in activation changes. One study demonstrated both activation changes and white matter changes as a result of both instruction and repeated training in word decoding in children with dyslexia (Meyler et al., 2008, Keller and Just, 2009). Unlike these previous studies, here we look not for changes in tissues or in regional activation, but in the neural representations of specific concepts using recent methods that can identify the nature of the information that is being coded by a given fMRI activation pattern.
The neural changes in our study were assessed in terms of the multi-voxel fMRI-measured activation pattern that occurs when participants think about how a particular mechanical system works. More precisely, the study assessed how their neural representation of a system changed as they learned more about its workings. The change in knowledge about specific mechanical systems should produce measureable changes in the neural representations of those systems. Furthermore, the changes may be directly related to the content of the instruction, such that instruction that describes shared properties between systems may increase their neural similarity.
Participants were taught with a series of successive increments of information about mechanical systems. The first level of explanation provided information about the components of the mechanical systems. The second increment included limited functional information. The third increment of explanation included the entire functional and causal sequence of the components of the mechanical systems. Each of these instructional steps should result in discernable neural changes. More specifically, the progressive deepening of the explanations of the systems might be expected to produce increasing involvement of cortical systems implementing higher-level psychological processes, and unchanging involvement of lower level perceptual systems that process the visual stimulus.
Despite the absence of prior neuroimaging investigations of mechanical systems, previous research in the brain bases of general cognitive processes does provide guidance as to which cortical systems might be involved. A set of eight potential cognitive processes, which have previously been associated with cortical systems, are postulated to correspond to regions or small sets of regions (networks) involved in understanding how mechanical systems work. These eight processes (and postulated cortical systems) consist of: (1) mental animation (bilateral parietal: Boronat et al., 2005), (2) causal reasoning (right temporo-parietal and medial prefrontal: Mason and Just, 2011), (3) embodied cognition (pre- and post-central: Rueschemeyer et al., 2010), (4) semantic knowledge (left temporal: Price, 2000), (5) language in context (bilateral inferior frontal: Mestres-Missé et al., 2008), (6) biological/goal-directed motion (right temporal: Pelphrey et al., 2003), (7) rule learning (middle and superior frontal: Bunge, 2004), and (8) visual processing (occipital cortex). The contributions of these various systems might be expected to shift as the instruction and learning progresses.
The goal of this study was to examine the changes in the neural representation over the course of learning and instruction, rather than establishing the correspondence between cognitive functions and brain regions. We developed several hypotheses concerning changes in representation. First, prior to instruction, during the first exposure to only the diagram and label, the hypothesis is that the participating regions would be primarily visual in nature, loading on the occipital cortex. Subsequent neural representations should involve relatively less occipital participation. Second, following the introduction of causality information, the representation could be expected to be distributed across a large set of systems including causal inference related regions (medial frontal and right temporo-parietal) for inferring causal relations among the components’ motions. Third, bilateral parietal, particularly the intraparietal sulcus, should increase in participation once components of the mechanical systems are introduced as a result of imagining components moving with respect to each other. Intuitively imagining the components moving may be a part of mental animation (Hegarty, 1992). These hypotheses provide a starting point for examining the changing involvement of cortical systems during learning.
Several recently developed methods for assessing neural knowledge representations were used in the study. One of these was the machine learning and classification of the multi-voxel activation patterns associated with each of the mechanical systems (Just et al., 2010, Mitchell et al., 2008). A related method analyzed the locations of the types of voxels whose activation levels were modulated in a consistent way by the different mechanical systems (Just et al., 2010). A third method used representational similarity analysis to assess how similarly-described systems became neurally more similar (Connolly et al., 2012). These methods can be used to converge on an account of how the neural representations change as instruction and knowledge cumulate.
Section snippets
Participants
Fourteen college students (6 females, all right handed and native speakers of English) between the ages of 18 and 26 years (M = 21.57; SD = 2.79) participated and were included in all of the analyses (no subjects were excluded). Each participant gave signed informed consent approved by the Carnegie Mellon University Institutional Review Board. Each participant received 5 minutes of practice with the experimental paradigm on a single training item (that was not included in the experimental stimuli)
Overview
The progressive stages of instruction describing how a mechanical system works resulted in corresponding changes in the states of neural knowledge representation of the systems. These successive brain states are identifiable (serve as signatures of the mechanical systems to which they correspond) and reflect the stages of cognitive processes that occur during this learning. Results will be interpreted as showing that there was a shift from visualizing how a system might move, to understanding
Overview
This research indicates the progression of neural changes the brain undergoes during the acquisition of knowledge of mechanical systems. The primary finding is that a sequence of knowledge states can be inferred by examining the snapshots of an evolving cortical representation as well as the activation during instruction. We propose that evidence from the domain of mechanical learning suggests a sequence of neurally-identifiable knowledge states, a sequence which may be general to other
Acknowledgments
This research was supported by the Office of Naval Research (Grant N00014-131-0250). We thank Tim Keller, Vladimir Cherkassky, and Andrew Bauer for help in data analysis and comments on earlier versions of the manuscript. We thank Chelsea McGrath for assistance in testing participants and running experiments. We thank the CCBI reading group for comments on an earlier draft of the article.
References (51)
- et al.
Distinctions between manipulation and function knowledge of objects: Evidence from functional magnetic resonance imaging
Cogn. Brain Res.
(2005) - et al.
Exploring the neural correlates of goal-directed action and intention understanding
NeuroImage
(2011) - et al.
Cerebral processes related to visuomotor imagery and generation of simple finger movements studied with positron emission tomography
NeuroImage
(1998) - et al.
Learning-related fMRI activation associated with a rotational visuo-motor transformation
Cogn. Brain Res.
(2005) - et al.
It takes the whole brain to make a cup of coffee: The neuropsychology of naturalistic actions involving technical devices
Neuropsychologia
(2005) - et al.
Distributed neural systems for the generation of visual images
Neuron
(2000) - et al.
Imagery in sentence comprehension: An fMRI study
NeuroImage
(2004) - et al.
Altering cortical connectivity: remediation-induced changes in the white matter of poor readers
Neuron
(2009) - et al.
Neuroanatomical distribution of five semantic components of verbs: Evidence from fMRI
Brain Lang.
(2008) - et al.
Structural brain plasticity in adult learning and development
Neurosci. Biobehav. Rev.
(2013)
Neuroimaging contributions to the understanding of discourse processes
Modifying the brain activation of poor readers during sentence comprehension with extended remedial instruction: a longitudinal study of neuroplasticity
Neuropsychologia
Patterns of fMRI activity dissociate overlapping functional brain areas that respond to biological motion
Neuron
What artificial grammar learning reveals about the neurobiology of syntax
Brain and language
Body schematics: on the role of the body schema in embodied lexical-semantic representations
Neuropsychologia
The time course of changes during motor sequence learning: a whole-brain fMRI study
NeuroImage
Teaching an adult brain new tricks: a critical review of evidence for training-dependent structural plasticity in humans
NeuroImage
Tracking children’s mental states while solving algebra equations
Hum. Brain Mapp.
Functional MRI of language: New approaches to understanding the cortical organization of semantic processing
Annu. Rev. Neurosci.
How we use rules to select actions: A review of evidence from cognitive neuroscience
Cogn. Affect. Behav. Neurosci.
The representation of biological classes in the human brain
J. Neurosci.
Decoding the representation of numerical values from brain activation patterns
Hum. Brain Mapp.
Neuroplasticity: Changes in grey matter induced by training
Nature
The extended language network: A meta-analysis of neuroimaging studies on text comprehension
Hum. Brain Mapp.
Imaging learning: The search for a memory trace
Neuroscientist
Cited by (21)
A brain-based account of “basic-level” concepts
2017, NeuroImageCitation Excerpt :The voxels used were the 200 voxels most stable over the basic concepts, disregarding occipital voxels whose activation correlated with word length. ( The choice of 200 voxels was guided by previous studies in which 100–200 stable voxels were the smallest set sizes that could still attain among the highest accuracies in classification of individual concepts, e.g. Just et al., 2010; Mason and Just, 2015). Cross-classification was expected to result in higher accuracies between basic concepts and their typical versus atypical subordinate concepts.
The neural and cognitive basis of expository text comprehension
2024, npj Science of LearningExecutive function predictors of science achievement in middle-school students
2023, Frontiers in PsychologyOverlapping semantic representations of sign and speech in novice sign language learners
2022, Proceedings of the 44th Annual Meeting of the Cognitive Science Society: Cognitive Diversity, CogSci 2022The neuroscience of advanced scientific concepts
2021, npj Science of LearningNeural alignment predicts learning outcomes in students taking an introduction to computer science course
2021, Nature Communications