Highly invariant recognition memory spaces for real-world objects revealed with signal-detection analysis

Recognition memory is an essential form of memory retrieval, allowing us to determine whether a thing or episode we experience in the present has been experienced in the past. Recognition typically involves decisions on whether stimuli look "old" or "new". These decisions can be difficult, especially when the new stimuli are similar to the old ones (e.g., as in police lineup identification). Although the role of similarity between old (targets) and new (foils) in recognition is obvious, it is often hard to predict how confusable particular targets and foils would be and how much of their confusability comes from their similarity as opposed to insufficient memory. Here, we show that, even when stimuli are various real-world objects whose subjective similarity depends on numerous non-defined factors and varies across people, they form highly consistent recognition memory spaces. That is, proportions of correct (hits) and incorrect recognitions (false alarms) produced by certain target-foil combinations are distributed similarly across different participants. Moreover, using the discriminability index d' from signal-detection theory as a distance metric, we found that the recognition spaces recovered from simple two-alternative choice tests (2-AFC) correctly predicted performance in more complex 4-AFC tests not only for hits but differentiated false alarms for each foil. We also found that the decreased encoding time reduced all absolute d' distances (presumably due to lower memory strength) but preserved the relative distances. Therefore, we found evidence for highly consistent signal-detection recognition memory spaces of real-world objects invariant to memory task complexity.