Analyzing the dynamics of free recall: An integrative review of the empirical literature

Abstract

Relatively few experiments have measured the time course of free recall from episodic or semantic memory. Of those that have, most report that cumulative recall is a negatively accelerated exponential (or hyperbolic) function that is characterized by two properties: asymptotic recall and rate of approach to asymptote. The most common measure of free recall performance (viz., the number of items recalled) provides a reasonably good estimate of asymptotic recall if a relatively long recall period is used (which is rare), but the effect of experimental manipulations on the rate of approach to asymptote cannot be determined without timing when recall responses occur. The research reviewed herein suggests that the rate of approach to asymptote may offer an estimate of the breadth of search through long-term memory. The search in question, unlike most ofthose investigated in the memory literature, is unique in that it requires minutes rather than milliseconds to complete.When subjects attempt to generate items from a semantic category or to recall a recently presented list of words, they do not complete the task in an instant. Instead, their performance is almost invariably characterized by intermittent successes occurring over an extended period of time. In spite of this interesting fact, memory researchers have exhibited an overwhelming preference for recording which (or how many) items are recalled during some brief period of time, regardless of the speed with which those items have been retrieved. Does the time course of free recall offer any useful information about the nature of retrieval? Because this subject is an old one that commands little attention today, one might assume that the answer is no. However, a review of the extant literature, limited as it is, suggests otherwise.Exactly what can be learned that is not already known by studying the dynamics of free recall? First, the studies that have tracked the time course of recall usually reveal that subjects continue to make progress far beyond the 1-or 2-min recall period typically provided. Indeed, Roediger and Thorpe (1978)-to take one example that will be considered in detail later-reported that recall on an episodic memory task continued to increase more than 20 min into the recall period. Moreover, a clear difference in the level of recall between two conditions (easily named pictures vs. words) became evident only after several minutes had elapsed. Had these authors used the standard brief recall period, they might have mistakenly concluded that the two conditions produced equal levels of performance. How often are incorrect conclusions arrived at because the recall period is too brief? The answer is Correspondence concerning this article should be sent to John Wixted, Department of Psychology-OfOs, University of California, San Diego, La Jolla, CA 92093. 89 unknown because most researchers simply ignore the potential problem.Second, most of the literature reviewed below suggests that the time course of free recall, like reaction time in the Sternber...

Scale invariance in the retrieval of retrospective and prospective memories

“…All functions were negatively accelerated, consistent with random sampling with replacement from a finite search set (see Wixted & Rohrer, 1994, for a review of the dynamics of free recall). However, there was a slower rate of approach to asymptote for letter fluency than for the present task, indicative of a larger search set in the former than in the latter case (see Wixted & Rohrer, 1994, for evidence that the rate of approach to asymptote is inversely related to set size). Importantly, the cumulative output functions for the present task were similar across the different time scales, suggesting that the sizes of the search sets were also similar.…”

Dual-task influences on retrieval from semantic memory and coordination dynamics

“…The rate of generating category exemplars begins rapidly and progressively decreases as the number of generated exemplars increases toward the participant's limit. This exponential retrieval can be derived from i = (r/S)(N -n), (1) where h is the rate of recall, r is the rate at which items are sampled from the semantic set of size S, N is the asymptotic number recalled, and n is the number recalled at time t (Wixted & Rohrer, 1994).…”

“…Could concurrent coordination dynamics lower the sampling rate r or extend (defocus) the boundaries of the search set S? Either change (see Equation 1) in the present experiment would have affected the dependence of h on n (Maylor, Chater, & Brown, 2001;Wixted & Rohrer, 1994) and reduced the number of exemplars generated in the 30-sec recall period. A first step toward evaluating the preceding ideas would be new experiments in which researchers (1) manipulated both retrieval dynamics (e.g., large and small semantic categories) and coordination dynamics (e.g., LSC and MSC); (2) examined several retrieval/coordination intervals (e.g., up to 3 min; Herrmann & Murray, 1979); (3) considered in addition to i the interresponse time (IRT) of exemplar generation, given the approximation IRT, = S/[r(N -i)], where i denotes position in the sequence of responses (Wixted & Rohrer, 1994); and (4) performed a local, moving-window analysis to determine whether IRT covaries with the dynamical measures in question.…”

Verbal fluency in Huntington’s disease: a longitudinal analysis of phonemic and semantic clustering and switching

“…Observation of clustering (where related words are produced rapidly one after the other) and switching (where there is a slight pause before another cluster of related words is produced) has led to the proposition that two distinct but complementary cognitive strategies (i.e. clustering and switching) are in operation during verbal fluency tasks [14,37,40]. This concept has been refined and operationalised by Troyer et al [37] and subsequently used to investigate further the nature of verbal fluency performance in neurologically impaired patients [31,36].…”