Marsha C. Lovett

Department of Psychology

Carnegie Mellon University

(Not) Just Another Model of the Stroop Effect

Extended Abstract

The Stroop effect offers a window onto mechanisms of selective attention and information processing. The standard (and quite robust) result is as follows: when people are given color word stimuli (e.g., the word “red”) printed in colored inks, the time to name the ink color of each stimulus is much slower when the word and ink conflict (e.g., the word “red” in blue ink) compared to when the word and ink match and also compared to when the stimulus is not even a word (e.g., the string “XXX”). This same slowing on conflict trials, however, does not occur when the task involves reading the word instead of naming the ink color. The basic question is, then, why do people show such difficulty maintaining attention to the ink color but not to the word?

Several models have been advanced to answer this question. The most prominent to date involves a PDP network with separate pathways for processing words and colors and with a gating mechanism to modulate task-appropriate processing. When a Stroop stimulus is presented, both pathways propagate activation in the network, but the word-processing pathway is taken to be stronger from people’s extensive practice at reading.This leads to the above asymmetry in reading versus color-naming. Note that this model presumes that the information processing of both features of the stimulus occurs in parallel.

By virtue of its subsymbolic mechanisms, ACT-R can also model the effect in terms of “stronger” components for responding to the word feature of each stimulus. This account, however, does not take full advantage of ACT-R’s symbolic level, nor does it explore the possibility of pockets of serial processing during task performance. In this talk, I will present an ACT-R model of the Stroop effect that includes both serial and parallel aspects. Specifically, the model includes parallel competition between the critical productions (one to initiate word reading and one for color naming), serial execution of each (under particular circumstances, both productions eventually execute), and then parallel competition among declarative chunks for whichever procedure is currently being implemented. I will show that this model fits the results from several Stroop experiments in the literature and discuss its implications for how the brain handles conflict.