Fine grained visual representations contribute to the content of visual long-term memory

An ongoing debate in psychology and cognitive science concerns the nature of memory representations: are they purely semantic or also visual? In order to answer this question, we were seeking to manipulate the quality of visual long-term memory representations while keeping its accessibility consistent. We used a mixture modelling approach, which provides estimates for the probability of recall and the fidelity of the recalled representations, to assess long-term memory performance of 140 participants in an object-colour association task. During encoding, participants either reproduced the colour of an object in a simultaneous presentation scenario (i.e., perceptual condition) or reproduced the colour of an object in a delayed estimation scenario (i.e., working memory condition). Crucially, all participants were instructed to memorize the colours of the objects for a subsequent memory test (i.e., long-term memory). As expected, at encoding participants in the perceptual condition revealed enhanced fidelity compared to participants in the working memory condition. Interestingly, and despite reduced fidelity at encoding, participants in the working memory condition showed enhanced long-term memory retrieval in comparison to participants in the perceptual encoding condition, while the probability of recall did not differ between the two conditions. The results imply that participants did not exclusively rely on verbal labels but also on fine grained internal visual representations.