The repetition of errors in recall: a review of four 'fragmentation' experiments

This review reanalyses the data from four experiments originally designed to test the fragmentation hypothesis. Participants were asked to recall triple or quadruple associates, cued by each of their components in turn, and to guess if they could not remember. There were many errors in recall and many of those errors were repetitions of previous errors. This reanalysis focuses, not on the fragmentation hypothesis, but on the repetition of errors. It works backwards through sequences of test trials to discover the best prior match to the responses on each trial. It reports frequencies of different categories of repetition, conditional probabilities of repetition, correct recalls, and the probability of repetition in relation to the lag between trial and match in the test sequence. These results may be summarised as (1) every event (a stimulus or a response or just a retrieval) to which the participant attends is separately recorded in memory, creating an ordered record of those events that have engaged the participant's attention; (2) the compilation of the record is automatic; while attention to a stimulus is at the participant's disposal, the consequent entry into memory is not, and (3) the retrieval of a potential response from memory is spontaneous; that retrieval becomes an overt response if it is compatible with the cue. This makes sense of a number of historic anomalies in the study of recall and informs some contemporary problems in the study of short-term memory.