Using approximate Bayesian computation for estimating parameters in the cue-based retrieval model of sentence...

Area of Science:

• Computational Linguistics
• Cognitive Science
• Psycholinguistics

Background:

• Traditional grid search for computational model parameter estimation is computationally expensive and does not quantify parameter uncertainty.
• Accurate parameter estimation is crucial for understanding cognitive models, such as the cue-based retrieval model of reading times.

Purpose of the Study:

• To present Approximate Bayesian Computation (ABC) as a superior method for parameter estimation in computational models.
• To demonstrate how ABC quantifies uncertainty in parameter estimates, unlike conventional grid search methods.
• To enable the generation of prior and posterior predictive distributions for model outputs, such as reading times.

Main Methods:

• Implemented Approximate Bayesian Computation (ABC), a Bayesian approach, for parameter estimation.
• Applied ABC to the cue-based retrieval model (Lewis and Vasishth, 2005) in the context of reading time data.
• Contrasted ABC with the conventional grid search procedure for parameter estimation.

Main Results:

• Approximate Bayesian Computation (ABC) provides a more computationally efficient method for parameter estimation compared to grid search.
• ABC successfully quantifies the uncertainty associated with parameter estimates, offering a more complete picture of model fit.
• The use of ABC facilitated the generation of prior and posterior predictive distributions for reading times.

Conclusions:

• Approximate Bayesian Computation (ABC) is a powerful and advantageous method for parameter estimation in computational cognitive models.
• ABC enhances model interpretability by quantifying parameter uncertainty and enabling predictive distribution generation.
• This Bayesian approach offers significant improvements over traditional grid search methods in computational linguistics and psycholinguistics.