The Statistical Reader: The Role of Orthographic Regularities in Reading

Recent statistical learning views of reading posit that writing systems present to their readers a wide range of statistical regularities which are leveraged to process printed texts. While substantial research has focused on the “vertical” correlations between orthographic, phonological, and semantic units in a given writing system, here we employ information-theoretic measures to further consider “horizontal” regularities—the extent to which printed units predict and are predicted by other printed units, in one writing system compared to another. As a first step, we present a novel information-theoretic measure that captures how horizontal regularities constrain lexical access given the distribution of orthographic information in a writing system and considering realistic retinal and cognitive constraints. We then present a series of empirical studies serving as proof of concept, from both single-word reading experiments and analyses of eye movements during naturalistic reading, which examine how a reader who has internalized these regularities could leverage them for efficient uncertainty reduction regarding printed information while reading on-the-fly. Our findings converge on high-order general principles fleshed out in terms of explicit computational mechanisms that simultaneously apply to a wide range of writing systems and that can potentially explain behavioral outcomes across the trajectory of reading development and reading skill.