Multi-factor analysis in language production: Sequential sampling models mimic and extend regression results.

Area of Science:

• Psycholinguistics
• Cognitive Science
• Computational Linguistics

Background:

• Descriptive models like linear regression dominate response time analysis in psycholinguistics.
• Empirical cognitive models, such as sequential sampling models (SSMs), offer multivariate insights but are underdeveloped in language production research.

Purpose of the Study:

• To examine sequential sampling models (SSMs) as a measurement approach for spoken word activation dynamics.
• To formally compare SSMs with regression for analyzing language production data.
• To advance the application of SSMs in the field of language production.

Main Methods:

• Analysis of two language production experiments.
• Application of sequential sampling models (SSMs) to model response activation and selection mechanisms over time.
• Comparison of SSMs with traditional regression techniques.
• Examination of a hierarchical Bayesian regression-SSM approach.

Main Results:

• SSMs successfully reproduced regression predictors and extended these findings through multivariate decomposition into cognitive parameters.
• The combined regression-SSM approach allowed for joint modeling of more conditions and achieved by-item modeling.
• Spoken word differences were primarily attributed to activation rates and production times, not activation thresholds.

Conclusions:

• SSMs provide a more detailed and informative approach to analyzing response times in language production compared to regression.
• The developed hierarchical Bayesian approach enhances the flexibility and scope of SSMs in cognitive modeling.
• Understanding the dynamics of word activation through SSMs offers new avenues for research in language production.