Is Familiarity Different Than Remembering?
Have you ever had the experience of seeing someone in the grocery store that you know you've already met, but you can't quite remember where or when you met them? In other words, you have a feeling of familiarity, even if you can't remember what, exactly, this familiarity is based upon? Compare this experience with what happens when you meet someone whom you recognize immediately, and can remember lots of relevant details about. (For instance, you can remember his or her name, or the last time you met.)
These examples illustrate the two processes that psychologists and cognitive neuroscientists believe are at work when we remember someone or something: a general feeling of familiarity and conscious recollection of specific details. The idea that recollection and familiarity are (at least psychologically) distinct is uncontroversial. There are many conditions that affect one process but not the other. A more controversial subject is whether the two processes are neurally distinct. That is, do they depend on different brain structures?
Sauvage and colleagues attempt to address this question in a recent paper in the journal Nature Neuroscience
. They took a group of rats and trained them on a recognition memory task in which they presented a series of odors mixed into a digging medium. For instance, the rats were given cumin scented wood chips or sand that smelled of thyme. Then they gave the rats a "test list" following a delay. Half of the items on the "test list" were the same odors and mediums that the rats had been exposed to earlier, whereas the other half were new combinations. Rats that got cumin wood chips now might be supplied with cumin-scented sand or glass beads coated in that spice. The rats were trained to make one response to the old pairs and another response to the rearranged pairs. The critical comparison was between the performance of a group of rats that had received a lesion to the hippocampus (a brain structure known to support declarative memory, or memory for facts and events) and a group of control rats with no lesion.
The innovative part of the study is that Sauvage and colleagues were able to vary the incentive the rats had for making correct decision by controlling the amount of the reward and the difficulty in obtaining the reward. This allowed the researchers to plot the hit rate (correct responses to a target) against the false alarm rate (incorrect responses to a new, or in this case re-paired item) at several different levels. This kind of plot is called an ROC curve
and has been used in psychology to describe people's performance on recognition memory tests. If the participant is randomly guessing, then the ROC will be a straight line from the bottom left of the plot to the top right. The better the performance, the higher the curve will fall above that diagonal. As with any curve, it can be modeled mathematically and, as with most curves, it can be modeled in more than one way.
Sauvage and colleagues consider two models. The first model assumes that the ROC can be described by a single memory-strength continuum, in which familiarity is simply a weaker form of memory than recognition, but still the same form. The second model assumes that the ROC is described by separate recollection and familiarity processes in the brain. This latter example is known as the dual process model.
Here's the tricky part: When the authors probe how well each model describes the data from the control rats and rats with hippocampal lesions, both models do a pretty good job. The difference between the predictions of the two models and the data is not significantly different from zero, although the dual process model is slightly better. In other words, there is no statistical reason to prefer one model over the other based on how well the models fit the behavioral data. The authors do offer some additional evidence for why they favor the dual process model, but this information relies on a subtle statistical argument that isn't typically used in ROC analyses.
For instance, if we assume that the single strength memory model is correct, it appears that the lesioned animals simply have weaker memories relative to the controls. (This finding shouldn't be too surprising because we know that the hippocampus
is involved in memory for facts and events). If we assume the dual process model is correct, however, then it appears that the data from the hippocampal lesion animals fall along a curve (indicating that the rodents relied entirely on familiarity to solve the task) whereas the data from the control animals falls along a straight line (indicating that they used recollection only). But both possible explanations make sense.
This last point, that the control animals have a straight ROC, is remarkable considering that one almost never sees that in the human recognition memory literature. Indeed, it is usually assumed that people rely on both recollection and familiarity in recognition memory tasks. It would seem, then, that either this task taps something other than what people use in solving recognition memory tasks, or that rats solve the task in a fundamentally different way than humans do. Both possibilities merit further investigation.
Edited by Mind Matters at 04/06/2008 1:04 PM