Dr. Rajeev Raizada
A key aim of Cognitive Science is to understand how people's mental representations are structured, and how they give rise to behaviour. However, fMRI has had very little to say about this issue: merely showing which part of the brain lights up for a particular task does not in itself tells us anything about the mechanisms via which that task is carried out. I will present recent work aimed at addressing this problem. Drawing upon machine-learning research, we can treat brain scans not just as a collection of individual voxels lighting up, but instead as multivariate distributions of spatial patterns. These distributions have structure, which we can quantify and then seek to relate to behaviour. In a study of Japanese and English speakers listening to /ra/ and /la/, we found that the statistical separability of the neural patterns elicited by these sounds predicted individual differences in people's ability behaviourally to tell the sounds apart. Moving to the domain of numerical cognition, we have found that the separability of people's fMRI patterns in a "number sense" task correlated with their scores on standardised math tests. Both of these results hold true even when the whole brain is analysed at once in a single statistical test, without the need for any selection of a region-of-interest. These results, if they turn out to hold more generally, could allow fMRI to serve as a diagnostic tool, distinguishing neural representational competence from behaviourally measured performance. Possible implications for the diagnosis and remediation of learning disabilities will be discussed.
|
