Finding Words in a Sea of Text: Word Search as a Measure of Sensitivity to Statistical Regularities in Reading

Statistical learning (SL) is hypothesized to play a fundamental role in reading, yet the correlations between reading and SL are largely mixed. This inconsistency may result from the fact that most SL studies train participants to learn novel, nonlinguistic visual regularities, which overlooks two important factors: (a) SL performance varies across domains, and (b) most SL studies utilize tasks with short exposure phases with a limited set of novel structured stimuli. Rather than exposing participants to novel statistics, we explored how prior learning of the statistical regularities inherent in natural texts predicts individual differences in reading. We developed a novel measure of long-term orthographic SL by assessing participants’ ability to chunk letter information based on its statistical properties. Adults were prompted to find high- and low-frequency English words (derived from written-language corpora) when a single target word was embedded in an array of background distractors comprising letters that do not form words. Performance on this task was compared against three established measures of component skills of reading: lexical decision, orthographic awareness, and spelling recognition. Participants were faster and more accurate at identifying high-frequency words, replicating classic psycholinguistic results. Performance was also impacted by semantic diversity—the variation of the semantic contexts a word appears in—independent of frequency. Critically, word search performance significantly predicted each reading subtest, suggesting that the task draws upon key readingrelated skills. Sensitivity to orthographic statistical structure may serve as a crucial foundation that drives individual differences in reading, consistent with SL-based accounts of language.