Word predictability effects are linear, not logarithmic: Implications for probabilistic models of sentence...

Area of Science:

• Cognitive Psychology
• Psycholinguistics
• Computational Linguistics

Background:

• Contextual predictability significantly influences word processing during language comprehension.
• Existing models propose differing relationships (linear vs. logarithmic) between predictability and processing difficulty.
• Resolving this debate is crucial for understanding predictive language processing mechanisms.

Purpose of the Study:

• To empirically investigate the function linking lexical predictability and word processing times.
• To differentiate between linear and logarithmic predictions of processing difficulty.
• To constrain theoretical models of predictive language comprehension.

Main Methods:

• Conducted two high-powered experiments: self-paced reading (N=216) and cross-modal picture naming (N=36).
• Performed a meta-analysis of existing eye-tracking while reading studies (N=218).
• Analyzed the relationship between contextual predictability and word processing times.

Main Results:

• A robust linear relationship was consistently observed between lexical predictability and word processing times across all studies.
• Findings contradict predictions derived from surprisal theory.
• Results support a proportional pre-activation account of lexical prediction effects.

Conclusions:

• The study provides strong evidence for a linear linking function in lexical prediction.
• These findings have significant implications for computational models of language comprehension.
• The results challenge surprisal theory and support alternative predictive processing accounts.