The question of how the human brain represents and organizes conceptual knowledge seems impossibly complicated to answer, given how even simple nouns, such as "celery," might be expected to activate neurons in various areas, such as those involved in eating and tasting. In a Research Article in the 30 May 2008 Science, Mitchell et al. described a computational model that can actually predict the human brain activity associated with the meanings of nouns. The model is based upon a theory that represents nouns as 25-dimensional vectors, where each dimension is a sensory-motor feature such as eating or smelling. The vectors are computed from the co-occurrences of words within a trillion-word text catalog that captures typical word use. The researchers "trained" the model with observed functional magnetic resonance imaging (fMRI) data from subjects presented with 60 images of concrete nouns (such as "celery," "foot," and "cow") and found that it can identify a noun that has not been included in the training set and predict the pattern of neural activity when a subject is shown a picture of a new noun. The results thus establish a direct, predictive relationship between the statistics of word co-occurrence in text and the neural activation associated with thinking about word meanings.

Click here to read original article