Towards a computational model of actor-based language comprehension.

Area of Science:

• Psycholinguistics
• Neuroscience
• Computational Linguistics

Background:

• Sentence comprehension relies on identifying the primary actor.
• An actor-based strategy is proposed to be cross-linguistically valid, influenced by cognitive principles and linguistic input.
• Existing models often focus on similarity-based interference, potentially overlooking actor-specific processing.

Purpose of the Study:

• To introduce and test a computational model of the actor strategy in sentence processing.
• To investigate how language-specific properties modulate actor competition.
• To differentiate between actor competition and similarity-based interference in language comprehension.

Main Methods:

• Development of an initial computational model for the actor strategy.
• Contrast of weighted and unweighted distance metrics derived from the model.
• EEG study on German word order processing to test predictive power of metrics.
• Analysis using linear mixed-effects models.

Main Results:

• A weighted metric, accounting for language-specific actor feature weighting, significantly outperformed an unweighted metric in predicting electrophysiological activity.
• Actor competition effects are not reducible to simple feature overlap or similarity-based interference.
• The model provides a neurobiologically plausible account of sentence-level meaning construction.

Conclusions:

• The actor strategy is a fundamental principle in sentence processing, modulated by language-specific features.
• Computational modeling offers valuable insights into neurocognitive mechanisms of language.
• This work challenges purely memory-based interference models and advances understanding of sentence comprehension.