Bound feature combinations in visual short-term memory are fragile but influence long-term learning

We explored whether individual features and bindings between those features in VSTM tasks are completely lost from trial to trial or whether residual memory traces for these features and bindings are retained in long-term memory. Memory for arrays of coloured shapes was assessed using change detection or cued recall. Across trials, either the same colour-shape (integrated object) combinations were repeated or one feature was repeated while the other varied. Observers became sensitive to the repetition of bindings, but only if it occurred on every trial. Repetition of single features only led to learning in the cued-recall task, and was weak compared to whole-object repetitions. Results suggest that representations in visual short-term memory comprise integrated objects rather than individual features. These representations are readily displaced by new representations formed on subsequent trials. However, when a given representation is not displaced, longer term residual traces can be generated to support long-term learning, and any learning that does occur is based on integrated objects, not individual features.