Noisy-channel language comprehension in aphasia: A Bayesian mixture modeling approach.

Area of Science:

• Psycholinguistics
• Neuroscience
• Computational Linguistics

Background:

• Agrammatic aphasia is characterized by reliance on semantic plausibility over syntax in sentence comprehension.
• Previous views attributed this to syntactic deficits, but a recent "noisy-channel" model suggests increased noise expectation.

Purpose of the Study:

• To investigate the nature of the noisy-channel model in aphasia.
• To determine if individuals with aphasia adapt their noise processing to environmental statistics.
• To compare noisy-channel inference in aphasia versus healthy adults.

Main Methods:

• Replication of findings in healthy adults on inference by weighing prior probability and noise corruption.
• Extension of noisy-channel inference to adults with chronic post-stroke aphasia and healthy controls.
• Hierarchical latent mixture modeling to account for guessing rates and individual differences.

Main Results:

• Healthy adults adjust noise probability estimates based on input error rates.
• Individuals with aphasia show a greater tendency to draw noisy-channel inferences for semantically implausible sentences.
• Adaptation of inference rates to increased noise in individuals with aphasia remains unclear.

Conclusions:

• Aphasia comprehension may be understood through a noisy-channel framework, highlighting reliance on semantic inference.
• The findings suggest potential avenues for targeted language rehabilitation strategies.
• Further research is needed to clarify noise adaptation in aphasia.