How 'rational' is semantic prediction? A critique and re-analysis of

Area of Science:

• Cognitive Neuroscience
• Psycholinguistics
• Computational Linguistics

Background:

• Event-related potential (ERP) studies investigate semantic processing and prediction.
• Previous research suggested Bayesian adaptation in semantic predictions based on trial frequency.
• A recent study by Delaney-Busch et al. (2019) claimed evidence for rational, Bayesian adaptation of semantic predictions.

Purpose of the Study:

• To critique the analyses and conclusions of Delaney-Busch et al. (2019) regarding rational semantic prediction.
• To re-analyze ERP data to investigate the adaptation of semantic predictions.
• To determine if N400 effects support rational or probabilistic semantic prediction.

Main Methods:

• Re-analysis of event-related potential (ERP) data from Lau, Holcomb, and Kuperberg (2013).
• Examination of N400 amplitudes in response to associatively related and unrelated prime-target word pairs across blocks with varying relatedness frequencies.
• Statistical analysis addressing confounds of lexical frequency and including data from all experimental blocks.

Main Results:

• The N400 effect for related words did not support rational adaptation of semantic predictions as claimed.
• Confounding of prediction effects with lexical frequency and neglect of initial block data were identified as critical flaws in the original analysis.
• The effect of relatedness gradually strengthened across blocks, indicating improved prediction accuracy rather than rational adaptation.

Conclusions:

• The N400 data do not provide evidence for rational or probabilistic adaptation of semantic predictions.
• The findings suggest that participants improved their ability to predict target words based on prime words, but not through a rational Bayesian process.
• Under the original framework, the observed semantic predictions might be considered 'irrational'.