Wednesday, July 02, 2008

Making Memories, Again

Lasry et al. , in a letter to Science, offer an interesting interpretation of work reported in a previous post, showing that testing of already learned words enhances long-term recall when assessed 1 week later, whereas repeated studying had no beneficial effects. Here are their comments:
In their Report, "The critical importance of retrieval for learning" (15 February, p. 966), J. D. Karpicke and H. L. Roediger III show that delayed recall is optimized, not with repeated studying sessions, but with repeated testing sessions. The authors conclude that "retrieval during tests produces more learning than additional encoding."

We suggest a complementary interpretation. Classically, encoded information becomes consolidated and can later be retrieved. The tacit assumption is that retrieval of a consolidated memory is a read-only mechanism, which does not affect the memory. Recent studies have shown that elicited memories are in fact labile and become reconsolidated following each retrieval. Labile elicited memories require de novo protein synthesis to be maintained, similar to that of newly acquired memories. Neurobiological differences between consolidation and reconsolidation processes were recently described in Science. On the psychological level, reconsolidation is useful for explaining false and biased memories. Reconsolidation also leads to a memory model called multiple-trace theory: Every time a memory is reactivated, a new version of it is reconsolidated, leaving multiple traces of the same memory.

With respect to Karpicke and Roediger's study, we hypothesize that repeated testing (retrieval) should lead to multiple traces (due to repeated reconsolidation), which facilitate recall. Reinterpreting Karpicke and Roediger's results from a multiple-trace reconsolidation perspective supports this hypothesis and provides a new framework for explaining the effectiveness of frequent in-class assessments in pedagogies such as Peer Instruction.

1 comment:

  1. Really cool stuff; I hadn't seen this.

    Doesn't it seem kind of inefficient to enhance retrieval just by storing multiple redundant traces? If someone could come up with a technique for ensuring better "cue specification" in the service of retrieval, they'd revolutionize college life... :)