Large-scale replication study reveals a limit on probabilistic prediction in language comprehension.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Psycholinguistics

Background:

• The N400 brain potential is modulated by word predictability, suggesting pre-activation of upcoming words.
• A 2005 study by DeLong et al. provided acclaimed evidence for phonological prediction, observing N400 effects for both nouns and articles based on cloze probability.

Purpose of the Study:

• To directly replicate the findings on phonological prediction of upcoming words.
• To investigate whether the N400 modulation by cloze probability extends to articles, as suggested by prior research.

Main Methods:

• A large-scale, multi-laboratory direct replication study (N=334).
• Pre-registered replication and exploratory Bayes factor analyses.
• Pre-registered and exploratory Bayesian single-trial analyses of electrical brain potentials (N400).

Main Results:

• Replication analyses confirmed the N400 effect for nouns but not for articles.
• Single-trial analyses showed significant effects for nouns, but not for articles.
• Bayesian analyses indicated a potentially small, non-zero effect for articles, likely too small to detect without very large samples.

Conclusions:

• The study successfully replicated the noun-related N400 effects.
• The findings do not support the routine pre-activation of phonological forms for predictable words, particularly articles.
• The evidence for article pre-activation is weak and likely requires extremely large sample sizes to observe.