a b s t r a c t
Article history:
Received 15 March 2015
Accepted 7 February 2016
Available online 13 February 2016
Monitoring of learning is only accurate at some time after learning. It is thought that immediate monitoring is
based on working memory, whereas later monitoring requires re-activation of stored items, yielding accurate
judgements. Such interpretations are difficult to test because they require reverse inference, which presupposes
specificity of brain activity for the hidden cognitive processes.We investigatedwhethermultivariate pattern classification
can provide this specificity. Weused a word recall task to create single trial examples of immediate and
long termretrieval and trained a learning algorithmto discriminate them. Next, participants performed a similar
task involving monitoring instead of recall. The recall-trained classifier recognized the retrieval patterns underlying
immediate and long term monitoring and classified delayed monitoring examples as long-term retrieval.
This result demonstrates the feasibility of decoding cognitive processes, instead of their content.