Backward Inhibition

In task-switching situations, participants must ensure that the currently relevant task is the one that is performed. Therefore, that task must be the most active amongst all competing possibilities; that task must dominate one's attention. One theory of how humans achieve this domination is to couple activation of the relevant task with inhibition (reducing activation) of interfering tasks. In task-switching situations, when one switches to a new task (e.g. AB), the activation levels from the previous task ('A') is likely to interfere with the activation of the desired task ('B'). Therefore, to achieve selection of task B, task A is believed to be inhibited.

As an analogy to the problem of selecting a new task in the face of active competitors, imagine you are in a room with two (or more) stereos. You switch on one stereo, and turn it all the way up to volume '10'. Now, you decide to turn on a second stereo, but you can't hear the new song clearly. Therefore you try to overcome this by turning up the volume of stereo 2. However, this volume also only goes up to 10, so you have two songs playing at the same volume, and you can't hear either one clearly. Therefore, in order to listen to the second stereo, you combine turning its volume up (activation) with turning the volume of the first stereo down (inhibition). Now the song from the new stereo can be heard clearly.

This example, albeit simplistic, clearly shows the problem of task-switching, and how activation often isn't enough to achieve fluent selection. Inhibition is often essential.


The BI paradigm

In the laboratory, inhibition of disengaged tasks is inferred by adapting the traditional task-switching paradigm by increasing the task number from 2 to 3. This allows two critical comparisons: 1) returning to a task after one intermediate trial (e.g. ABA) 2) returning to a task not recently performed (e.g. CBA - where 'A', 'B', & 'C' are arbitrary labels for tasks). The logic of this is that if a task is inhibited when it is switched away from, then the suppressed levels of activation might persist for a short time and hinder re-activation of this task. This is confirmed by comparing reaction times (RTs) of the two sequences. It is a consistent finding that ABA sequences are slower and more error prone than CBA sequences. This is because task 'A' in and ABA sequence has been inhibited recently, unlike a CBA sequence, where task 'A' has not been inhibited as recently (and thus has had time to recover from its inhibited state).

This RT cost was called 'Backward Inhibition' (BI) by Mayr and Keele (2000), who were the first to report the effect (see also Arbuthnott & Frank, 2000).

What is inhibited?

The question arises as to what aspects of the task are inhibited when a switch is signaled.  The work of Iring Koch and colleagues has suggested that aspects related to response processes might be inhibited. As responses are often made on the same keys for different tasks, conflict might arise between response mappings when choosing a response in this paradigm. Removing this conflict has been shown to affect BI. For example, Schuch and Koch (2003) showed that if no response was executed for task 'B' in an ABA sequence, the BI effect disappeared.

Cue-target translation & BI

Our work has suggested that it might not exclusively be the response stages that are inhibited during task-switching. Our view is that inhibition targets those aspects of the task that generate the greatest inter-trial conflict. Of the typical trial structure, the way the task is cued has received relatively little attention (though see the work of Katherine Arbuthnott).

When a cue is presented, participants must prepare for the upcoming task. This preparation might involve activating a representation in working memory of the task to be performed (Mayr & Kliegl, 2003). If this process meets with a still-active representation used on the previous trial, conflict ensues leading us to the 'two-stereo' problem discussed above. Therefore, the irrelevant representation is inhibited allowing the new representation to dominate. This process of establishing a representation is what we refer to as "cue-target translation" (Houghton, Pritchard, & Grange, 2009).

For example, say the experiment requires the participant to respond to the location of an oval among three other ovals, all differing on one visual property (see below for an example of this). The participant knows which oval is the 'target' for a given trial based on a valid pre-cue that describes the characteristic of the target to search for (e.g. "Angled"). When the participant sees the cue, they must translate this in working memory into a target representation that will allow them to select the correct oval when the target display appears as the cue disappears before target onset (for a demonstration that removing the cue does not affect BI as some researchers have suggested , see Grange & Houghton, 2009). This translation must become dominant, so recent target translation must be inhibited (thus producing BI).

 However, what if this process of cue-target translation can be bypassed? If the cue was somehow more informative, then no translation will occur in working memory, generating no conflict. If there is reduced conflict, less (or no) inhibition should be required. Therefore, by using iconic-cues that visually present the characteristics of the oval to search for (see below), the environment provides all the information required to perform the task, and as such the cue does not need to be translated into a representation in working memory. As such, when a switch occurs, no conflict is generated in working memory, and no inhibition is required. Therefore, BI is eliminated.