Evaluating the Contribution of Intra-Linguistic and Extra-Linguistic Data to the Structure of Human Semantic Representations

We describe Bayesian models that learn semantic representations from either extra-linguistic data or intralinguistic data, or from both in combination. We evaluate the validity of these models using three human-based measures of semantic similarity. The results provide strong evidence for the hypothesis that human semantic representations are the product of the statistical combination of extraand intra-linguistic sources of data.