Abstract
Recall errors from visual short-term memory are often not random. Rather, people may erroneously report information about the wrong item in memory. These have been interchangeably referred to as “swap errors” or “misbinding”.
For true swap errors, features should be incorrectly bound two ways: for the probed item as well as for the “swapee” – the other item in memory whose features are being swapped. However, previous evidence argues against true swaps, because when two features of an object have to be reported, errors are uncorrelated. Here, we demonstrate that it is possible to distinguish two mechanisms for reporting features from the wrong item: true swaps and one-way misbinding.
For one-way misbinding, only one object’s feature is swapped, whereas information about the other object’s feature might be lost, requiring participants to guess. Previous cued recall approaches cannot distinguish these two types of error.
In our free full report design, participants remembered the colours and locations of three dots shown at random locations on an imaginary circle, in random colours from a colour wheel. In cued recall, participants were cued with each of the colours, and had to reproduce the corresponding locations, or vice versa. To test free full recall, they reported both features of each object. In either case, participants reported all items in the array in self-determined order.
A new mixture model was able to account for true swap errors, as well as one-way misbinding. To do this, we introduced individual guessing parameters depending upon whether participants reported objects correctly or swapped. Furthermore, we distinguished between (i) guessing the whole object vs (ii) a feature, and between (a) guessing the swapped vs (b) non-swapped feature.
Our findings show that what has previously been labelled as swap error can, in fact, be decomposed into true swap errors, or one-way misbinding.