Learning Decouples Accuracy and Reaction Time for Rapid Decisions in a Transitive Inference Task.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Animal Behavior

Background:

• Transitive inference (TI) involves using internal representations of abstract relationships for decision-making.
• While transitive learning mechanisms are known, decision-making dynamics during learning and inference require further understanding.
• The drift diffusion model (DDM) is a framework for perceptual decision-making.

Purpose of the Study:

• To investigate if the DDM can model decision-making in a TI transfer task.
• To analyze rapid decision patterns deviating from standard accuracy and response time (RT) models.
• To explore the applicability of the DDM to symbolic reasoning and serial relational learning.

Main Methods:

• Trained six macaque monkeys on a TI transfer task involving a list of seven images.
• Recorded decisions using saccadic eye movements or reaching movements.
• Applied a generalized DDM implementation (PyDDM) to fit accuracy and RT data.

Main Results:

• Monkeys achieved consistent learning of list structure within 200-300 trials.
• Behavior showed a symbolic distance effect, with accuracy increasing with ordinal item distance.
• RTs remained stable during learning despite accuracy improvements; DDM fits captured RT distributions with increasing evidence accumulation and a collapsing decision bound.
• Learning and transfer were modeled by varying drift rate, with minimal changes in other DDM parameters.
• Eye and reaching movements exhibited similar dynamics, with non-decision time accounting for RT differences.

Conclusions:

• The DDM framework can successfully account for decision-making dynamics in transitive inference tasks.
• The study identified a distinct DDM dynamical regime applicable to symbolic reasoning and serial relational learning.
• Findings extend the DDM's utility beyond perceptual tasks to complex cognitive processes.