Performance on perceptual word identification is mediated by discrete states

Area of Science:

• Cognitive Psychology
• Psycholinguistics
• Information Processing

Background:

• Traditional signal-detection models assume graded signal strength.
• Discrete-state models propose all-or-none information loss, previously tested with stringent assumptions.
• Critique of ROC curve linearity as a sole test for discrete models due to certainty assumptions.

Purpose of the Study:

• To contrast discrete-state and signal-detection models for English word identification.
• To assess the conditional independence property of discrete-state models.
• To evaluate model performance across varying stimulus durations and response tasks.

Main Methods:

• 50 participants identified briefly flashed English words.
• Variable flash durations were employed.
• Two response tasks were used: two-alternative forced choice with confidence ratings and yes-no with confidence ratings.
• Discrete-state and signal-detection models were fitted to participant data.

Main Results:

• Discrete-state models showed superior fit compared to signal-detection models.
• Discrete-state models outperformed signal-detection models for 90% of participants in the two-alternative task.
• Discrete-state models outperformed signal-detection models for 68% of participants in the yes-no task.

Conclusions:

• Discrete-state models provide a viable framework for predicting performance in perceptual word identification.
• The conditional independence property is a more robust assessment for discrete-state models than ROC curve linearity.
• Findings challenge the universality of graded signal strength in perceptual tasks and support discrete processing stages.