A cognitive modeling approach to learning and using reference biases in language.

Area of Science:

• Psycholinguistics
• Cognitive Science
• Computational Linguistics

Background:

• Real-time language processing involves predicting upcoming input using linguistic and non-linguistic biases.
• Implicit causality bias influences predictions about entity rementioning based on causal roles in events.
• The acquisition and real-time application of these reference biases remain unclear.

Purpose of the Study:

• To investigate how implicit causality biases are acquired and utilized during real-time language processing.
• To develop and test a computational model simulating the prediction of discourse referents and their linguistic forms.
• To explore the role of domain-general learning mechanisms in shaping linguistic biases.

Main Methods:

• Developed a reference learning model within the PRIMs cognitive architecture.
• Simulated the prediction of upcoming discourse referents and their linguistic forms based on presented linguistic input.
• Introduced asymmetries in discourse continuations to train the model.

Main Results:

• The model exhibited biased prediction behavior, learning to anticipate discourse continuations based on subject- or object-biased implicit causality verbs.
• Learned biases included predicting subject referent continuations as pronouns and object referent continuations as proper names.
• These learned biases generalized to novel linguistic contexts, demonstrating robust learning.

Conclusions:

• Complex linguistic behaviors, such as implicit causality bias, can be explained by cognitively plausible, domain-general learning mechanisms.
• The study provides insights into psycholinguistic accounts of predictive language processing and language acquisition.
• Findings contribute to theories of implicit causality and reference processing in human cognition.