Human and computer estimations of predictability of words on written language

Abstract

When we read printed text, we continuously predict the follow words in order to integrate information and direct future eye movements to forthcoming words. Thus the Predictability has become one the most important variables when explaining human behavior and information processing during reading. In this study we present results of word predictability in long Spanish texts, estimated from human responses in a massive web-based task. We used Latent Semantic Analysis (LSA) as a way to estimate human-based predictability values computationally. We validated the human estimation of predictability with local and global properties of the text, and we showed that LSA-distance on adequate timescale captures some semantic aspects of the prediction.