When Forgetting Is Good: Control Reductions at Retrieval

What was your 6th birthday party like?

If you successfully retrieved that memory, you may now be ever so slightly less able to remember your other childhood birthdays. A variety of behavioral evidence has shown that such "retrieval induced forgetting" of strongly competing memories is fundamental to memory retrieval. In a new article in Nature Neuroscience, Kuhl et al. provide neuroimaging evidence which ties retrieval-induced forgetting to activity in prefrontal cortex.

In the study, subjects studied a series of 240 word pairs, for example ATTIC-DUST, ATTIC-JUNK, or MOVIE-REEL. During the practice phase, subjects were given a cue word and the first letter of an associated word, and were asked to retrieve the rest of the associated word (e.g., ATTIC-D___). Critically, this practice phase included only some of the word pairs, such that half of the cue words were never practiced (e.g., MOVIE would not have been practiced) and half of the associates were never practiced (e.g., ATTIC-J___ would never have been practiced). Finally, in the testing phase, subjects were asked to recall all of the word pairs. All of this was performed within a fMRI scanner, allowing the authors to acquire spatially-precise measurements of the neural activity involved in study and retrieval.

Results demonstrated the phenomenon of retrieval induced forgetting (RIF) - that is, subjects remembered the practiced word-pairs (ATTIC-DUST) better than those that went completely unpracticed during the practice phase (MOVIE-REEL), and yet were markedly worse at recalling those associates that were themselves unpracticed but had been paired with a word cue that was practiced (ATTIC-JUNK), relative to those word pairs where neither word had ever been practiced (MOVIE-REEL). There are several theories of RIF, including some proposing that unpracticed associates are somehow "suppressed" while practicing other cue-associate pairs, either directly and strategically, or through a more incidental process of competitive activation.

However it happens, such "forgetting" of competing items is thought to be helpful for selective memory retrieval - it allows the target items in memory to be retrieved more easily. Indeed, those subjects with a greater tendency to forget these competing items also showed greater neural activity in a region of the brain thought to mitigate conflict (the anterior cingulate, or ACC), the first time they attempted to retrieve the associate for a given word pair. This suggests that these subjects may have experienced more conflict during retrieval, and perhaps then engaged control mechanisms that translated into less conflict down the line.

Similarly, activity in the hippocampus during the first retrieval attempt correlated both with ACC activity and with retrieval-induced forgetting, again as though too many competing items had been made active when attempting to retrieve the target item from memory.

However, subjects who ultimately forgot more of the competing items were not more likely to activate a region of the brain involved in selection and inhibition - the right ventrolateral prefrontal cortex - on the initial practice trial. This lightly challenges the idea that directed inhibition is mediated by the vlPFC, and also that it is important for suppressing competing memories. Such a perspective which would have predicted greater, not lower vlPFC activity among those with more forgetting.

On the other hand, the results do suggest an important role for another brain region - the dorsolateral prefrontal cortex (dlPFC) - in the selection of appropriate items from memory. Traditionally, this region is thought to be important for upregulating "task-relevant" information in cortex. Here, it was directly correlated with the amount of conflict experienced by subjects (as reflected in ACC activation) as well as the benefit of practice. This pattern of results indicates that the ACC detected conflict in these subjects' initial retrieval attempts, subsequently triggering increased top-down control by dlPFC, ultimately translating into better memory retrieval.

As subjects acquired more practice with retrieving each word pair, those who forgot more items also showed a larger decrease in activity in some of these control regions. For example, subjects who showed more forgetting also showed a greater decrease in both ACC and r.vlPFC across the course of practice. It seems that those who experienced more conflict during the first retrieval engaged control mechanisms and were thereafter able to reduce the amount of control required to retrieve the correct item on subsequent retrieval attempts.

One perspective is that retrieval induced forgetting is a necessary byproduct of selecting the appropriate memory for retrieval. By this view, competing memories must be suppressed to allow the correct memory to "pop out" to the surface, and this has a subsequent cost in terms of retrieving those competing memories. An alternative perspective is that competitive dynamics within neural networks guarantee that only a subset of all possible representations will ever dominate a network, and that the others will undergo a kind of inadvertent weakening due to "losing" the competition.

These results do little to resolve this longstanding debate, but they do clearly demonstrate that there is sometimes a benefit to forgetting: competing memories are less likely to later interfere, and so metabolically expensive activity in prefrontal cortex can be reduced during subsequent retrieval attempts.