As I’m trying to consider the use of reactivating memories during lucidity, I am looking at the more unique use-cases of TMR. This paper also overlaps with my interest in using LDT to treat nightmares (although only slightly).
Despite being mentioned as a future direction in the comprehensive 2013 TiCs review from Oudiette and Paller, this appears to be the first instance of using TMR to induce forgetting (TMR-Forget; TMR-F). Of course that could be exciting that it works, or unsettling if we believe the file-drawer problem might have sucked away many failed attempts, but that’s definitely besides (yes, next to) the point and not worthy of discussion here.
The general methodology was to bring participants in for an initial memory test and cue assocation, have them nap and play tones associated with a certain memory and the “forget” cue, and then retest their memory (7 days later). They follow the memory test and method details of Rudoy et al., 2009, Science, which seems appropriate since that is kind of the “seminal” TMR paper. They just take that design and add a tone associated with forgetting.
The “forget tone” was learned via a task where a sequence of words was encoded, and if this tone played along with a word, then you should forget that word. At the end of the word list the probe was to “Please recall as many non-tone words as you can.” The object-tone associations were learned as in Rudoy et al. with the object-location learning task and sounds that semantically matched the objects playing alongside. I don’t recall if this was done in the original Rudoy paper, but they take the original learning phase data and for each participant get “encoding strength” of each memory item so that they can only apply TMR to the items around the median. Note encoding strength is a measure of their memory performance during learning/training. They say this was to “focus on memories that were neither too weak nor too strong and to match cued and control objects for encoding strength.” Sounds good to me.
So to promote forgetting during sleep, they took 5 objects from learning, and cued them all during SWS. But for all of the objects, the “forget cue” was also played along with the object sound during SWS. They then compared later memory performance for these cued objects to objects that were not cued but performance-matched prior to sleep.
In retesting memory after TMR-F (7 days later), they probed recall in 2 ways. First, the more general recall test was simply to free recall as many objects as possible from the experiment. Then they performed the Quadrant location test, which probed the specific location of where each object was on the original learning grid. Objects were presented center-screen and participants were asked to identify which of the 4 quadrants it belonged to, along with their confidence. Participants were also tested on the original word list (where they learned the forget cue), to test the long-term effects of the forget cue, although there was no TMR-F for these words.
The main finding is that participants recall less cued (TMR-F) objects than control objects. This is a very cool finding. It would be nice to have had a condition where participants were cued but without the forgetting cue. I think this seems like the most logical control, although perhaps an argument against it is that there is already so much research supporting that TMR improves memory. Still, would be nice. The other main memory test is the quadrant recognition test, and here I would expect them to report the performance on this test between control and reactivated objects. However, they instead break performance up within each of these conditions by the items that were recalled and those that were not. They find a trending difference, but this seems uninteresting – participants perform worse on detailed recall of objects they weren’t able to free recall?? This should turn out, right?
Sidenote: Figure 3 (only results figure) has significance markers above a few plots that they make clear in the legend don’t survive multiple comparisons correction. I think that’s okay and still worth reporting, but the plot markers should somehow make this distinction.
I like the paper, and even though it’s just one main effect, it seems strong and very interesting. I’m only a bit confused by the interpretation of the effect. The authors seem to argue that the forgetting cue is causing “interference” during replay of the cued object.
Regarding our statistically significant reduction in recall of reactivated objects we suppose this is the result of interference with memory transforming processes that occur when experiences are replayed during sleep.
They talk in the intro about how memory replay during sleep might be beneficial as opposed to memory replay during awake, because memories are labile when reactivated and during waking there is much interfering content. In contrast, memories can be replayed during sleep without any interference (or reduced at least). So they propose that providing interfering content would cause forgetting, and that they do that here. While I agree with the whole idea/premise, I don’t see how this experiment causes interference. They way the design is setup it seems like they are trying to promote directed forgetting during sleep. The cue is associated with directed forgetting, and then they replay it. I think a better way to test the interference idea would be to reactivate (or cue) multiple objects simultaneously and see if that has negative consequences, right? They discuss a paper that kinda did this with sounds, and found no effect of forgetting (also no memory benefit).