Early neuropsychology research indicated that long-term memory and short-term memory were separable – in other words, long-term memory could be impaired by damage to the hippocampus without any corresponding deficits in short-term memory. However, this idea has come under scrutiny in recent years. Neuroimaging technology has demonstrated that the same network of brain regions is active in both long-term and short-term memory tasks, suggesting that these regions may interact more than previously assumed. As noted in Speer, Jacoby & Braver’s 2003 article, estimates of either type of memory seem to be contaminated by the other type. For example, in a simple “digit span” test of short-term memory, you also have long term memories of phone numbers as well as the digits themselves that may artificially inflate performance on this task. One way of addressing this is to manipulate subjects to preferentially engage one or the other systems in the same task.

To accomplish this, Speer et al. gave subjects lists of words to remember and then asked them to determine whether certain items were present in that list. However, for one half of the experiment the lists were on average 4 items long, whereas for the other half the lists were on average 8 items long. Speer et al. hypothesized that when subjects had formed an expectation of remembering relatively few words (as in the 4-item half), they might preferentially engage the short-term memory system (and concentrate on the active maintenance of those items, or verbal rehearsal). In contrast, when subjects expected longer lists (as in the 8-item half), they might engage the long-term memory system due to the well-known capacity limits of short-term memory (for example by elaboratively encoding the items and utilizing cued retrieval strategies).

In order to systematically compare brain activations during each half of the experiment, Speer et al. included a subset of trials that were matched in terms of the number of items, the length of the delay and the duration of encoding and retrieval. Therefore, any differences in brain activity between these matched trials would have to reflect strategy or expectancy differences on the part of the subjects, allowing a relatively clean dissociation of the neural substrates underlying short- and long-term memory strategies.

The results showed several interesting effects that seem to indicate this manipulation worked:

- Accuracy declined with increasing list length in the 4-item half but accuracy was stable across list lengths in the other half.

- Reaction times were slower with increasing list lengths only in the 4-item half – as though subjects were searching through an actively maintained store of items to accomplish the task for that half. The other half of the experiment showed no such consistent increase in reaction times with list lengths.

- A surprise recognition test at the very end of the experiment showed that subjects were better at recognizing those words they had encountered in the 8-item half than those in the other half of the experiment, suggesting they had indeed encoded these items with more of a long-term strategy.

Assuming that the manipulation worked, what do the neuroimaging results tell us about short- and long-term memory?

Although 84% of all regions showed similar activity across both task blocks, a few regions did show selective activation during either half of the experiment. For example, posterior motor, prefrontal infero-lateral and prefrontal midlateral regions were more active early in the trials in the 8-item half, but were more active at the end of the trials in the 4-item half.

What might this difference mean? Lateral posterior PFC is thought to be involved in semantic or articulatory processing, and could be expected to be active when subjects were subvocally rehearsing the items using a verbal code – indeed, it showed more activity late during the short lists, as though subjects were subvocally rehearsing the items once they had several to remember.

In contrast, during the long lists, lateral posterior PFC showed more activity early in the trials, where many subjects indicated they had been trying to elaborate on the sounds of the words (a long-term elaborative encoding strategy). This reflects an alternative use of verbal coding that nonetheless relies on the same area.

The authors conclude by noting that they found very few task-specific regions – that is, the regions to show strategy effects differed mostly in the time course of their activity rather than in their location in the brain. The take-home message is that most of the areas involved in one type of memory are also involved in the other type, although many will show temporal differences in activation.

Given that long-term memory here shows such reliance on prefrontal cortex, one might expect long-term memory deficits in populations with underfunctioning prefrontal cortices – such as children or frontal patients. This contradicts some implicit but widespread assumptions that these populations might be limited to deficits involving working memory, executive or cognitive control, planning, or other high-level functions. Furthermore it meshes nicely with the Unsworth & Engle perspective, where the relationship of the working memory system to long-term memory is more clearly specified.

One of the most interesting but unanswered questions from this research is the localization of strategy implementation: which regions trigger the use of a long- or short-term strategy? Are both strategies equally resource demanding? How optimal are the subjects’ strategy shifts, in terms of maximizing the number of memorized items? Are the strategies mutually exclusive, or alternatively are the 84% of regions with task-nonspecific activation actually engaged in both strategies all the time? These are all fascinating questions for future research into the nature of memory.