Finally, the Cognition paper with Shane is out. We argue that ease of learnability explains cross-linguistic universals across various semantics domains, including content and function words. We first recall our earlier result that learnability explains semantic universals in the domain of quantification and then we show that such an approach can be extended to content words. We focus on the domain of color terms. It has been suggested by Peter Gärdenfors and empirically observed by Gerhard Jäger that that color terms cross-linguistically denote convex regions of color space. We present a computational experiment with the artificial color naming system. We found out that the ease of learning of a system, measured by the ability of simple neural networks to learn the names for the colors, highly correlates with the degree of convexity of the system (see the picture). Even more, the degree of convexity uniquely explains the learning accuracies.
See here for the preprint.