Word Frequency and Predictability Dissociate in Naturalistic Reading.

Area of Science:

• Psycholinguistics
• Cognitive Science
• Computational Linguistics

Background:

• Readers exhibit slower reading times for infrequent or less predictable words.
• Debate exists whether word frequency and predictability effects are separate cognitive processes.
• Prior evidence for dissociation is limited by small samples, artificial materials, and simplistic modeling assumptions.

Purpose of the Study:

• To investigate if word frequency and predictability effects dissociate during naturalistic language comprehension, specifically story reading.
• To address limitations of previous studies by using large-scale datasets and advanced computational modeling.
• To determine if frequency and predictability effects are separable and additive in ordinary reading.

Main Methods:

• Analysis of a large corpus of naturalistic reading data (six datasets, >2.2 million data points).
• Estimation of word frequency and predictability using advanced statistical language models trained on extensive data.
• Application of nonlinear continuous-time regression to model reading behavior.

Main Results:

• Findings support dissociable effects of word frequency and predictability, consistent with earlier experimental studies.
• Both frequency and predictability effects were found to be additive.
• The study successfully demonstrated these dissociations at scale using naturalistic data and sophisticated modeling.

Conclusions:

• Word frequency and contextual predictability represent distinct cognitive phenomena in language processing.
• These dissociable effects are observable even in naturalistic reading of stories.
• The findings validate and extend previous research using more robust methodologies and data.