In connectionist network models of reaction time, stimulus codes are transformed into response codes through links that are set up between stimulus and response units. These connections are thought of as associative links between concepts, stored in memory. There can be both long-term memory (LTM) associations, and short-term memory (STM) associations (Barber & O’Leary, 1997).
LTM associations form between two units from repeated exposure over an extended period of time, such as the link between the stimulus unit for a written word and the response unit for saying that word, or the link between a stimulus unit for a particular location and a response unit for acting towards that location (since we often respond toward stimuli in our environment).
STM associations are temporary links formed between stimulus and response units based on a particular task at hand. For example, if you are told to press a left key when you see a blue stimulus and to press a right key when you see a green stimulus, then a STM association would form between the “blue” stimulus code and the “left” response code, and between the “green” stimulus code and the “right” response code. In the language of connectionist network models, there would be a temporary link between the “blue” stimulus unit and the “left” response unit. This means that activation in the “blue” stimulus unit would be used as input to the “left” response unit, causing activation in the “left” response unit to increase.
According to these models, both controlled (intentional, deliberate, conscious) and automatic (unintentional, reflexive, unconscious) translation of a stimulus code into a response code happens through the same mechanism: activation in a stimulus unit is transformed into output, passed along an associative link, and used as input to a response unit, causing that response unit to accumulate activation. STM associations implement the controlled translation from stimulus to response that is determined by the specific instructions of the task at hand. Because STM associations encode the instructions of the task, they always link a stimulus to the correct corresponding response. LTM associations, on the other hand, are based on previous experience, rather than the task at hand. As a result, STM associations have also been called “controlled lines,” while LTM associations have been called “automatic lines” (Kornblum et al., 1999).
It is easy to see how the combination of automatic and controlled lines can allow these models to account for the S-R consistency effect (regardless of whether one is implementing a dimensional overlap model or a response selection model, because the mechanism accounting for S-R consistency is the same in both). Consider, for example, how information processing might proceed in a typical Simon task: a blue stimulus appears on the left side, causing activation to accumulate in the blue relevant stimulus unit and the left irrelevant stimulus unit; if the instructions assign a left key-press to blue stimuli, then activation from the blue relevant stimulus is passed along a STM association to the left response unit; activation of the left irrelevant stimulus unit is passed along a LTM association to the left response unit; because the response unit is getting input from both stimulus units, it has a high input, and activation accumulates quickly, reaching the threshold for completion in a short amount of time. On the other hand, if the blue stimulus had appeared on the right side, the right irrelevant stimulus unit would have activated the right response unit through the LTM association, and input to the (correct) left response unit would have been lower. Lower input, of course, means activation accumulates more slowly, and the decision threshold is reached later.