How trial-to-trial learning shapes mappings in the mental lexicon: Modelling lexical decision with linear...

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psycholinguistics

Background:

• Trial-to-trial effects demonstrate that stimulus processing impacts subsequent responses.
• Priming effects are often explained by error-driven learning models.
• The mental lexicon's dynamic nature and continuous learning are key research areas.

Purpose of the Study:

• To investigate the presence of trial-to-trial learning in unprimed lexical decision experiments.
• To evaluate the Discriminative Lexicon Model's (DLM) capability in detecting such learning.
• To explore individual differences in lexical processing.

Main Methods:

• Utilized the Discriminative Lexicon Model (DLM) with distributional semantics and error-driven learning (Widrow-Hoff rule).
• Simulated lexical decision experiments on a trial-by-trial basis using British Lexicon Project (BLP) data.
• Employed Generalized Additive Models (GAMs) to predict reaction times using DLM-derived measures.

Main Results:

• Learning-based DLM simulations showed a better model fit compared to non-learning simulations for most subjects.
• Measures derived from DLM simulations provided insights into lexical processing and individual differences.
• The study successfully detected trial-to-trial learning in unprimed lexical decision tasks.

Conclusions:

• The Discriminative Lexicon Model (DLM) effectively models behavioral data in lexical decision tasks.
• Trial-to-trial learning is detectable in unprimed lexical decision experiments.
• Lexical knowledge is subject to continuous, ongoing changes.