Meaning-driven syntactic predictions in a parallel processing architecture: Theory and algorithmic modeling of ERP...

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Cognitive Neuroscience

Background:

• Syntactic and semantic processing interactions in language comprehension are complex.
• The P600 event-related potential (ERP) effect, typically indicating syntactic processing difficulty, can arise in grammatically well-formed sentences.
• This P600 effect in sentences like 'The hearty meal was devouring...' suggests a conflict between semantic predictions and syntactic roles.

Purpose of the Study:

• To investigate the nature of semantic effects on syntactic processing during language comprehension.
• To develop and test a computational model that implements a parallel processing architecture to explain the P600 effect.
• To determine if this model can identify conflicts in grammatical role assignment between semantic and syntactic systems.

Main Methods:

• Developed an algorithmic implementation of a parallel processing architecture.
• Incorporated meaning-based prediction of grammatical role labels using a probabilistic label guesser or a neural network.
• Compared predicted labels with those from a state-of-the-art dependency parser to detect labeling conflicts.

Main Results:

• The developed system successfully classified sentences from the Kim and Osterhout (2005) corpus.
• The model demonstrated the ability to detect intended labeling conflicts between semantic and syntactic role assignments.
• The findings support the hypothesis of independent semantic and syntactic systems interacting during comprehension.

Conclusions:

• The study provides a computational account for the P600 effect observed in semantically biased, syntactically correct sentences.
• The results suggest that conflicts between meaning-based predictions and syntactic parsing contribute to processing difficulty.
• The model has implications for understanding predictive mechanisms in language processing and cognitive architectures.