Limits on lexical prediction during reading.

Area of Science:

• Cognitive Psychology
• Psycholinguistics
• Computational Linguistics

Background:

• Efficient language processing may rely on predictive mechanisms to anticipate upcoming linguistic input.
• Understanding the role of prediction in reading comprehension is crucial for psycholinguistic theories.

Purpose of the Study:

• To investigate the extent to which prediction facilitates reading.
• To examine whether readers predict full word forms or broader semantic and syntactic information.
• To differentiate between graded prediction and full lexical prediction models.

Main Methods:

• A large-scale survey generated cloze scores for 2689 words across 55 text passages.
• An eye-tracking study measured reading behavior in response to varying levels of word predictability.
• Analysis focused on sensitivity to cloze probability and the effect of more-expected word competitors.

Main Results:

• Highly predictable words constituted only 5% of content words; most words had a more-expected competitor.
• Eye-tracking data revealed sensitivity to cloze probability but no cost associated with mis-prediction.
• Predictability of semantic and morphosyntactic information facilitated reading, even when word identity was uncertain.

Conclusions:

• Reading comprehension benefits from graded prediction, where readers anticipate likely upcoming information.
• The findings are inconsistent with models requiring full lexical prediction but support broader predictive processing.
• Language comprehension likely involves predicting contextual information rather than solely exact word forms.