Reading Time

Reading time research investigates the cognitive processes underlying text comprehension by analyzing the time spent processing individual words or phrases. Current research heavily utilizes transformer-based language models, particularly focusing on how well surprisal (a measure of word predictability) and related contextual factors, such as pointwise mutual information and contextual entropy, predict reading times. Discrepancies between model predictions and human reading times are being actively explored, with investigations into factors like word frequency, model size, and the calibration of probability scores from language models. These findings refine our understanding of human language processing and have implications for optimizing text presentation and information retrieval systems.