Abstract
The Stroop effect—the interference in reaction time when a color word is printed in a non-matching ink color—is a fundamental phenomenon in cognitive psychology. This paper proposes that the Stroop effect is not a cognitive failure but an emergent byproduct of optimizing neural networks for speed and accuracy in natural environments. Using a model based on laterally connected Self-Organizing Maps (SOMs), we demonstrate that interference arises from the competitive dynamics of bottom-up stimuli within the brain's linguistic and perceptual organizational structures. Our results show that a specific level of lexical interference optimizes the trade-off between response time and accuracy, mirroring human behavioral patterns without the need for explicit top-down control mechanisms.
Introduction
The Stroop effect serves as a primary measure of selective attention and cognitive flexibility. While the behavioral outcome is well-documented, the underlying neural mechanisms remain a subject of debate. Traditional computational models often invoke "top-down" executive functions to modulate processing pathways. However, emerging theories suggest that "bottom-up" competitive dynamics driven by the stimuli themselves may play a more significant role than previously thought. This study investigates whether the Stroop effect can emerge naturally from a self-organizing system, providing a biologically plausible explanation rooted in neural efficiency.
Methods
The research utilizes an extension of the BiLex model, a neural network framework for the human lexicon, enhanced with lateral connections.
Model Architecture
The system consists of two 20x20 Self-Organizing Maps:
- Semantic Map: Trained on RGB color values to represent perceptual color space.
- Lexical Map: Trained on phonetic features (IPA transcriptions) of Spanish color words.
Unlike standard SOMs that utilize a simple "winner-take-all" mechanism, these maps incorporate lateral interactions—excitatory for immediate neighbors and inhibitory for distant units. This allows the network to dynamically settle into a stable state over multiple time steps. Hebbian learning is employed to create associative links between the two maps, facilitating the flow of color information into the word-processing map.
Task Simulation
To simulate the Stroop task, the model is presented with a color input (semantic) and a word input (lexical). The simulation runs until the entropy (uncertainty) in the lexical map falls below a predefined threshold. The number of iterations required to reach this threshold is recorded as the Response Time (RT). The study used a synthetic dataset of 16 colors and their corresponding Spanish names, evaluated across congruent, incongruent, and neutral conditions.
Results
The model successfully replicated qualitative patterns observed in human behavioral data, achieving an overall accuracy of 84.2%. While accuracy was perfect (100%) for congruent trials, it decreased to 83% in incongruent trials.
Response Time and Interference
Incongruent trials required significantly more processing time, taking approximately 55 steps longer than congruent trials. This delay represents the time required for lateral inhibition within the lexical map to suppress conflicting word inputs in favor of the target color name.
The Efficiency Trade-off
A key finding is that the Stroop effect represents an "optimal cost." By tuning the lexical interference strength ($r_{lex}$), the researchers identified a "sweet spot" at $r_{lex} = 0.45$. At this point, the model balances the need for rapid responses in congruent scenarios with the maintenance of acceptable accuracy in conflicting scenarios.
Discussion
The emergence of the Stroop effect in this model suggests that cognitive control may be a distributed, stimulus-driven process rather than one requiring a centralized "executive" controller. The interference is a physical consequence of the lexical map attempting to resolve two distinct activation peaks simultaneously.
Limitations and Future Directions
While the model provides a robust explanation for basic Stroop interference, it is currently limited by its small scale (16 items) and does not yet account for complex phenomena like Stroop dilution or stimulus onset asynchrony. Future work will expand the lexicon to thousands of items and explore the model's application to bilingualism and cognitive pathologies, such as aphasia, through simulated map "lesioning."
Conclusion
This research demonstrates that the Stroop effect is a natural consequence of a neural architecture optimized for the agreement between different sensory modalities. By prioritizing speed and accuracy in typical environments, the brain becomes inherently susceptible to interference in rare, conflicting scenarios.
Summary updated