Modeling Misretrieval and Feature Substitution in Agreement Attraction: A Computational Evaluation

Area of Science:

• Computational Linguistics
• Psycholinguistics
• Cognitive Science

Background:

• Agreement attraction is a linguistic phenomenon where a verb agrees with a noun phrase other than the subject.
• Previous models primarily focus on retrieval-based mechanisms.
• Eastern Armenian presents a unique linguistic context for studying agreement attraction.

Purpose of the Study:

• To investigate number attraction effects in Eastern Armenian using computational modeling.
• To compare the predictive fit of encoding-based versus retrieval-based models of agreement attraction.
• To introduce a novel methodology combining self-paced reading with open-ended comprehension questions to analyze specific error types.

Main Methods:

• A self-paced reading study was conducted with Eastern Armenian speakers.
• Three computational models of agreement attraction were implemented within a Bayesian framework.
• K-fold cross-validation was used to compare model performance.
• Open-ended comprehension questions were analyzed to identify misinterpretations of number features and thematic roles.

Main Results:

• The data were better accounted for by an encoding-based model than a retrieval-based model.
• Evidence for both misinterpretation of the subject's number feature and misassignment of the verb's thematic subject role was found.
• These error patterns were sometimes observed within the same trial.

Conclusions:

• Encoding-based mechanisms play a significant role in agreement attraction in Eastern Armenian.
• The study provides evidence for multiple types of processing errors contributing to agreement attraction.
• Current computational models may not fully capture the complexity of agreement attraction errors observed in this study.