Is signalling responsible for the response-repetition effect in task switching? A test using two versus three response alternatives.

Task switching is used to measure humans’ cognitive control and flexibility. Participants’ responses are slower and less accurate when the task to perform changes from the task performed just before than when it repeats. Furthermore, it is often observed a RR effect, namely that participants are faster and more accurate when they have to produce the same response twice in a row when the task repeats. However, response repetitions cause costs in task switches. The RR effect could be due to the creation of an association, or binding between the task and the response. Repeating a task retrieves such association and, therefore, the previous response. Such a retrieved response will be correct in response repetitions, but incorrect in response switches. In contrast, in task switches, the binding of the task and the response leads to a partial mismatch in response repetitions, as the same response is given to a different task. Such a binding account assumes that the task and the response are bound in each trial. Then, retrieval of the previous response may or may not take place when the task repeats versus switches, respectively. However, alternative accounts exist, such as the signalling account that postulates that switching the response is easier than repeating the response when the task switches, because the switch of the task is taken as a “signal” to switch the response. In this experiment, we aim at testing a prediction from the signalling account that, if proved, would provide support for this account over the alternative feature binding and episodic retrieval account. When only two response alternatives are available (e.g., left or right key), the signalling mechanism biases towards the correct response in task switches with response switches, and towards the wrong response in task switches with response repetitions, hence the RR costs. However, if there are more than two response alternatives (e.g., left, middle, right), this mechanism would only signal to “switch away” from the previous response, but it would not facilitate responding to any of the other two responses. Therefore, in the present experiment, we contrast RR costs in two experimental sessions which are as similar as possible except for the number of possible response alternatives (two versus three). With three response alternatives, RR costs in task switches should be smaller, or absent, according to the signalling account. Not only the correct response would be “signalled” by the task switch, but also the third alternative, hence reducing the signalling benefits in task switches with response switches. However, according to the binding account, the mismatch caused by having a response repetition with a task switch is not affected by the number of response alternatives. The binding account hence predicts the same RR costs in task switches with 2 or 3 response alternatives.