A Computational Evaluation of Two Models of Retrieval Processes in Sentence Processing in Aphasia

Area of Science:

• Psycholinguistics
• Computational linguistics
• Neuroscience of language

Background:

• Individuals with aphasia (IWAs) often exhibit sentence comprehension difficulties.
• Parsing, the process of analyzing sentence structure, involves dependency completion.
• Two prominent models, activation-based and direct-access, explain dependency completion.

Purpose of the Study:

• To computationally evaluate two models of dependency completion: activation-based and direct-access (DA).
• To determine which model best explains sentence comprehension impairments in IWAs.

Main Methods:

• A self-paced listening task with subject and object relative clauses was administered to IWAs and controls.
• k-fold cross-validation was used to compare the predictive performance of the activation-based and DA models.
• Bayes factors were employed for model comparison.

Main Results:

• The activation-based model provided a better quantitative fit to the data for both IWAs and controls compared to the DA model.
• Model comparisons suggest intermittent deficiencies are a likely cause of impairments in IWAs, with potential contributions from slowed syntax and lexical access.

Conclusions:

• The activation-based model offers a more accurate account of dependency completion in sentence processing for both unimpaired individuals and IWAs.
• Intermittent deficiencies, potentially alongside slowed syntax and lexical access, are key factors in aphasia-related comprehension deficits.
• This study establishes a novel computational framework for comparing language processing models in clinical populations.