Some theories suggest that color and shape information – processed in different parts of the brain – must be integrated by attention in order to give rise to a coherent visual experience (in other words, attention is thought to solve the “binding problem”). Although that explanation is probably not satisfying to many, no doubt mostly because attention itself is an ill-defined construct, at least it’s a beginning in solving the binding problem. However, a recent article from Allen, Baddeley & Hitch argues that even this humble claim may be entirely wrong.
Allen et al. suggest that the binding of visual features may actually automatic insofar as it does not share resources with any other cognitive processes. For example, “verbal load” (i.e., repeating the word “Monday” again and again) does not interfere with individual visual short-term capacity limits. Allen et al. reasoned that perhaps other more complex tasks would also not interfere with this ability.
To demonstrate this, they first gave 8 subjects a task in which they had to remember the shape and color of four objects. Afterwards, they were asked to judge whether a single shape, color, or shape-color combination was present in the previous array. Every subject completed 384 such judgments, with negative feedback and instructions to emphasize accuracy rather than speed. In general, subjects were better at color judgments than shape or both-color-and-shape (i.e., binding) judgments. Allen et al. suggest that the similar accuracy between the shape and the binding condition may reflect an automatic process.
In a second experiment, 16 subjects repeated the task from the first experiment, and also repeated it while counting backwards by one from some number presented on the screen. A nearby experimenter monitored their performance on this secondary task. If binding is not automatic, but requires additional cognitive resources, then one would expect backward counting to more strongly affect performance in the binding condition than the other conditions. The results did show that performance was generally lower when subjects had to perform this concurrent task, but that the binding condition was similarly affected by this task as the shape- and color-only conditions. In other words, binding seemed automatic.
Based on this surprising result, the authors thought that maybe backwards counting is not demanding enough resources to show a differential impact on binding processes. To this end, they gave 24 subjects the same task as in the first experiment, except that during each trial of the task they had to remember the serial order of 6 randomized digits, and to recall these numbers correctly after each trial. (In addition, the authors made a modification to the design which included 25% feature repetition trials, because without the possibility of features repeating, memorizing just a single feature combination could result in differential benefits to the binding test, relative to shape- or color-only tests). The results showed that this digit recall test also impacted binding and shape- or color-only conditions equally, again suggesting that the maintenance of combined visual features is no more demanding than the maintenance of the features themselves.
However, the authors worried it could still be possible that binding is not automatic but rather that the digit recall task was simply not demanding enough. To this end they gave another 24 subjects the same task as in experiment 1, except for some trials subjects had to count backwards by 3s from a given number while performing the task. Their performance on the backward counting task was monitored. As before, this concurrent counting task generally decreased performance, but this was not more true of binding than the color- or shape-only conditions. In other words, binding still appeared to be automatic!
Contrary to these results, previous work has shown that binding is very fragile in one particular scenario: if subjects are presented with multiple objects during the judgment rather than a single item. Allen et al. reasoned that if this is the case, then sequential presentation of each item during the memorization period might decrease performance on binding more than in the shape- or color-only conditions. To test this, Allen et al. gave 36 subjects the same task as in the first experiment, except that items were presented sequentially during encoding and a simple articulatory suppression task was given (i.e., the “verbal load” mentioned at the start of this post, in which subjects merely repeat a simple phrase again and again). This manipulation did affect binding differentially, such that the decrease in performance resulting from sequential presentation was much greater in the binding condition than in the shape- or color-only conditions. This demonstrates clearly that binding is not invincible – but it does not appear to require the same form of attention as digit span or backwards counting.
So, it looks like binding is automatic, insofar as it is not affected by other simultaneous tasks. On the other hand, perhaps the authors didn’t run enough subjects to show a statistically significant effect. To test this, the authors combined the results from the first four experiments using meta-analytic methods, and confirmed that even after pooling all the results, there were no differential effects of concurrent tasks on the binding condition.
The authors conclude that these results support the idea that a separate “binding process” can often proceed automatically and does not appear to share resources with those involved in many secondary tasks. The authors do suggest that some forms of binding may require more active, effortful processing, such as those involving longer encoding intervals. These results are a humbling demonstration that the “binding problem” – a problem that has stumped many cognitive neuroscientists – is solved nearly effortlessly by the human brain.