Rethinking Familiarity: Remember/Know Judgments in Free Recall

Participants

Fifty University of California, San Diego (UCSD) undergraduates participated in exchange for course credit.

Materials

Three lists of 24 words were created by randomly drawing words from six different semantic categories (4 words per category). No two lists used the same semantic categories. The words were selected from category norms (Van Overschelde, Rawson, & Dunlosky, 2004), and we used categories that had at least eight words. Stimuli were presented and responses were recorded with the E-prime program (www.pstnet.com; Psychology Software Tools).

Procedure

Participants signed a consent form, listened to instructions, and were then presented with three lists that contained 24 words each (totaling 72 words). Each word appeared in the center of the screen in black courier font size 18 for 4 seconds and the inter-stimulus interval was 250 ms. After a 15-s distractor task (an adapted serial sevens test), they were prompted to type in as many words as they could recall in any order. After they typed in a word, they indicated their confidence that the word appeared during the presentation phase using a 5-point scale (ranging from 1 = low confidence to 5 = 100% sure the word was on the list). Next, they indicated whether their response was based on recollection or familiarity by making a “remember” (by pressing the “r” key) or “know” (by pressing the “k” key) judgment, respectively. After making a Remember/Know judgment, they were prompted to recall the next word. If they could not recall any more words, they pressed the enter key (once for each remaining word to be recalled) and the next list was presented. Lists 2 and 3 proceeded in the same manner.

The Remember/Know instructions were based on Gardiner and Richardson-Klavehn (2000), and they emphasized that a Remember judgment should be made if anything about the presentation of the word was recalled (such as thoughts the word elicited when it was studied), whereas a Know judgment should be made if the memory of the item was accompanied by no contextual/source detail. Prior to the experiment, participants were given a short practice list and test to acquaint them with the rating scale, and to ensure that they understood the Remember/Know distinction. Any questions that arose after the practice trial were answered before list 1 was presented. Each participant was tested individually in a quiet room.

Participants

Sixty UCSD undergraduates participated for psychology course credit.

Materials

These were the same as Experiment 1.

Procedure

The procedure was the same as Experiment 1 except that we included a source memory test. During the study phase, each word appeared with one of two questions that required an animacy judgment (is this item animate or inanimate?) or size judgment (is this item bigger than a shoebox?). Immedi ately following the stud y phase, participants were tested. After the first list, the testing procedure was the same as in Experiment 1: participants first typed in a word, then indicated their confidence that the word was presented, and then made a Remember or Know judgment. The source memory test was a surprise and was administered after the participant had completed recalling as many words from List 1 as possible. The test consisted of presenting each word that was recalled and asking whether that word was associated with the animacy question or the size question at study. For Lists 2 and 3, the procedure was similar, except that the source memory question for each word appeared immediately after the confidence rating and Remember/Know judgment was made (for these lists, the source memory test would not be a surprise, so there was no reason to delay it).

Participants

Sixty UCSD students were randomly assigned to a group (30 in the free recall group and 30 in the forced recall group).

Materials

These were the same as Experiments 1 and 2.

Procedure

The procedure was similar to Experiment 1, except we added a forced recall condition. The participants who were assigned to the forced recall condition were instructed to type 24 words after each 24-item list, even if they had to guess. They were also reminded that if they did make a random guess, they should use a “1” on the confidence rating scale.

For both the free and forced recall conditions, we also recorded the time required to recall each word. The reaction times (RTs) were measured with respect to a prompt that appeared at beginning of the recall period and that appeared again immediately after the confidence rating and Remember/Know judgment had been entered for the previously recalled word. Timing began with the presentation of each recall prompt and continued until the word was entered (indicated by pressing the enter key). Thus, the recorded data consist of a series of RTs, each of which reflects the time required to search for and then type a word.

Confidence and Accuracy of Remember and Know Judgments

As in the previous experiments, Remember judgments were indicative of stronger memory (measured by confidence and accuracy) than Know judgments, but the difference in accuracy was no longer apparent when the analysis was restricted to words recalled with high confidence. Of the 30 participants in the free recall condition, 29 made at least 1 Remember judgment and 1 Know judgment (including all recalled words regardless of confidence). The average confidence rating for Remember judgments (M = 4.75, SD = 0.48) was significantly higher than the average confidence rating for Know judgments (M = 3.96, SD = 0.95), t(28) = 3.73. Similarly, the average accuracy (i.e., correct divided by correct plus incorrect) for Remember judgments (M = 0.94, SD = 0.06) was significantly higher than the average accuracy for Know judgments (M = 0.81, SD = 0.27), t(27) = 2.68. However, the accuracy difference was no longer apparent when the analysis was restricted to words recalled with the highest confidence rating of 5. For the 21 participants who made both a high-confidence Remember judgment and a high-confidence Know judgment in the free recall condition, the average accuracy of these Remember and Know judgments we re 0.96 (SD = 0.05) and 0.98 (SD = 0.04), respectively. Thus, on average, Remember judgments indicate stronger memories than Know judgments, but the difference is eliminated when the analysis is limited to words recalled with high confidence. Once again, it is important to note that an unmeasurable difference in strength between high-confidence Remember and Know judgments might still exist (because the dependent measures are at the top of their respective scales), but it seems clear that both Remember and Know judgments reflect strong memories when the analysis is limited to words recalled with high confidence.

The same strength patterns were observed in the forced recall condition. Once again, 29 participants made at least 1 Remember judgment and 1 Know judgment. The average confidence rating for Remember judgments (M = 4.56, SD = 0.56) was significantly higher than the average confidence rating for Know judgments (M = 2.76, SD = 1.21), t(28) = 7.29. Similarly, the accuracy of Remember judgments (M = 0.86, SD = 0.11) was significantly higher than the accuracy for Know judgments (M = 0.49, SD = 0.29), t(28) = 7.21. When the analysis was limited to words recalled with high confidence, 20 participants made at least 1 Remember judgment and 1 Know judgment. The average accuracy of these Remember and Know judgments were 0.93 (SD = 0.07) and 0.85 (SD = 0.30), respectively, a difference that did not approach significance. The large standard deviation for the Know judgment accuracy score reflects the fact that two participants made only a single know judgment, and both were incorrect (so their accuracy scores were 0). Excluding those two participants, the average accuracy of the high-confidence Remember and Know judgments were 0.93 (SD = 0.07) and 0.94 (SD = 0.09), respectively.

Remember/Know Reaction Time Analyses

The results presented thus far suggest that Know judgments in free recall do not reflect familiarity-based decisions arising from a generate-recognize process, but it might be argued that they nevertheless reflect some other form of automatic memory. It is widely assumed that an automatic process occurs faster than a consciously controlled process (Yonelinas, 2002). If so, and if Know judgments reflect an automatic process, then Know judgments in free recall should be made faster than Remember judgments. By contrast, if Remember and Know judgments reflect retrieval from the same episodic memory search set, and if recall follows a relative strength rule (as is widely assumed), then the opposite result should be observed. The reason is that, as just described, Remember judgments were associated with stronger memories than Know judgments. This relative strength account further predicts that if Remember and Know judgments are equated for strength at a high level (as they appear to be when the analysis is limited to items recalled with high confidence), no difference will be observed in the speed of recall.

To test these predictions, we analyzed reaction times (RTs) measured from the onset of the prompt to recall a word to the moment the enter key was pressed (after the word was typed in). The prompt was presented at the start of recall period and again each time a Remember/Know judgment was entered for a word that had just been recalled. In the free recall condition, the mean RT for correct Remember judgments (M = 4.81 s, SD = 4.71 s) was much faster than the mean RT for correct Know judgments (M = 10.08 s, SD = 14.37 s), a difference that was marginally significant, t(27) = 1.86, p = .07. However, the Know RTs for two participants were extreme outliers (one had an RT of 61.3 s and the other 54.2 s) and the Remember RT for another s ubject was also an extreme outlier (29.0 s). Each of these scores was more than 3 standard deviations from their respective means, and, in a visual plot of the data, they stood out conspicuously from the remaining distribution of RT scores. With those three outliers excluded, the mean RT for Know judgments (M = 6.41 s, SD = 5.10 s) was still longer than the mean RT for Remember judgments (M = 3.97 s, SD = 1.13 s), and the difference was significant, t(24) = 2.71. This finding is the opposite of what would be predicted by an automatic memory interpretation of Know judgments, but it is consistent with a relative strength model of free recall (because Know judgments are associated with lower confidence and lower accuracy than Remember judgments).

The relative strength model predicts that the RT difference will be eliminated once the strength difference between Remember and Know judgments is eliminated (as it appeared to be when the analysis was limited to words recalled with high confidence). For the 21 participants in the free recall condition who made both a high-confidence Remember judgment and a high-confidence Know judgment, the mean RT for correct Remember judgments (M = 3.79 s, SD = 1.10 s) was somewhat faster than the mean RT for correct Know judgments (M = 6.08 s, SD = 10.46 s), but the difference was not significant. However, the Know RT for one subject was an extreme outlier (the RT for that subject was 51.76 s, which was more than 4 standard deviations above the mean). With that one outlier excluded, the mean RT for high-confidence Know judgments (M = 3.90 s, SD = 2.34 s) was very similar to that for high-confidence Remember judgments (M = 3.79 s, SD = 1.10 s). Thus, when Remember and Know judgments were essentially equated for confidence and accuracy – that is, they were equated for strength at a high level – the RTs were equated as well (cf. Rote llo & Zeng, 2008). This finding is again consistent with a relative strength model of free recall in which both Remember and Know judgments reflect retrieval from the same episodic memory search set.

Similar results were observed in the forced recall condition. Of the 28 participants who made at least 1 correct Remember judgment and 1 correct Know judgment, the mean RT for Remember judgments (M = 4.74 s, SD = 2.09 s) was significantly faster than the mean RT for Know judgments (M = 8.69 s, SD = 5.09 s). However, when the strength difference was minimized by limiting the analysis to words recalled with high confidence, the RT difference disappeared. For the 18 participants who made at least 1 correct high-confidence Remember judgment and 1 correct high-confidence Know judgment, the mean RT for correct Remember judgments (M = 4.02 s, SD = 1.17 s) was virtually identical to the mean RT for correct Know judgments (M = 4.15 s, SD = 1.79 s).

All of these results are consistent with a relative strength model of free recall according to which Remember and Know judgments reflect retrieval from the same episodic memory search set, but they seem hard to reconcile with the view that Know judgments in free recall reflect automatic memory.

Participants

Thirty UCSD students participated for psychology course credit.

Materials

Words that ranged in length from 3-8 letters, and ranged in concreteness (from moderate to high; 450-700), were pulled from the MRC Psycholinguist Database (Coltheart, 1981). That search yielded a large pool of words (1816) of which, 72 words were randomly selected to make up three lists of 24. Each participant studied the same 72 words, but words varied in their list location and presentation order varied for each participant.

Procedure

The procedure was the same as Experiment 1 except that words on the study list were presented for 5 s each.

What do Know Judgments in Free Recall Mean?

Tulving (1985) argued that know judgments, whether they are made during recall or recognition, reflect item-only information retrieved from semantic memory. Elsewhere, he also argued that “Access to, or actualization of, information in the episodic system tends to be deliberate and usually requires conscious effort, whereas in the semantic system it tends to be automatic” (Tulving, 1983, p. 46). Here, we propose a different idea. Our interpretation holds that both Remember and Know judgments in free recall reflect cue-dependent, consciously controlled retrieval from a single episodic memory search set. This interpretation agrees with Tulving’s (1985) account in one respect (namely, that context-free, item-only recall can occur) but not in another respect (namely, that item-only recall on a free recall test reflects semantic memory). Basically, we advance an interpretation that is more consistent with Tulving’s (1972) original view, which held that free recall for a recently presented list of items is a test of episodic memory, whereas semantic memory instead reflects the recall of information learned across multiple encoding episodes (e.g., knowledge of the number of bones in the human body). According to this view, we did not test retrieval from semantic memory in our experiments.

If this interpretation is correct, then both Remember and Know judgments would reflect the outcome of a consciously controlled search process, and the difference between them would be in the amount of source information that is retrieved when an item is recovered from the search set. That is, the difference between Remember and Know judgments in free recall would not be that one judgment reflects recollection and the other familiarity, or that one reflects a consciously controlled retrieval process and the other an automatic process. Instead, the difference would be that one reflects the consciously controlled retrieval of item-plus-source information from an episodic memory search set and the other reflects the consciously controlled retrieval of item information (with limited or no source information) from that same search set.

Our episodic memory interpretation is illustrated in Figure 7. This figure depicts a model in which the presentation of 24 items on a list creates a search set of 24 memory traces. During retrieval, these traces are sampled by a random search process with replacement. This is a simplified version of the recall process envisioned by SAM (Gillund & Shiffrin, 1984), which assumes that recall is based on a relative strength rule. If all of the items in the search set have the same memory strength (i.e., in a pure-strength list), then they would all have an equal likelihood of being sampled throughout the recall period. This equal-strength version of the model predicts that cumulative recall (i.e., the number of items recalled as a function of time spent recalling in the recall period) should be characterized by a negatively accelerated exponential growth to asymptote. The model further predicts that if some items in the search set are stronger than others (e.g., in a mixed-strength list), then the strong items would be preferentially sampled throughout the recall period (thereby delaying the recall of the weaker items). In that case, the strong items would be characterized by a faster rate of approach to asymptote than the weak items. Obviously, a more realistic model would take into account the semantic relationships between different subsets of categorized words, would include a stopping rule, and might also include provisions for output interference. However, these added complexities are not needed to illustrate the basic idea, which has often been used to characterize the dynamics of free recall (e.g., Unsworth, Spillers & Brewer, 2012; Rohrer & Wixted, 1994; Wixted, Ghadisha, & Vera, 1997; Unsworth, 2007).

Of the 24 traces shown in Figure 7, 6 are such that, when sampled, they are not sufficiently intact to recover the corresponding word (represented by the symbol “~”). As in SAM, these traces take time to sample, but they result in no response (either overt or covert). Of the remaining 18 traces, 6 are such that they contain mainly item information (represented by a “K” in Figure 7). When these traces are sampled, the originally presented word is recovered and is overtly recalled the first time it is sampled. Any attendant source information is also recovered, and a decision criterion is used to decide if there is enough source information to declare the item to be remembered. For the 6 K items, source information is sufficiently limited that it falls below the decision criterion, so a Know judgment is made. The 12 remaining traces contain more source information (represented by an “R” in Figure 7). When they are sampled, the word is overtly recalled the first time it is sampled. In addition, because the amount of recovered source information falls above the decision criterion, a Remember judgment is made.

In all 4 of our experiments, Know judgments were less accurate than Remember judgments. However, Know judgments made with high confidence were as accurate as (or nearly as accurate as) Remember judgments made with high confidence. In Experiment 3, the mean RT data reflected these differences in strength, which is consistent with the random search model just outlined. However, the random search model makes more specific predictions than that. The model predicts that when cumulative recall is analyzed, both Remember and Know judgments should be characterized by a negatively accelerated exponential growth to asymptote, with the rate of approach to asymptote being slower for weak Know judgments compared to strong Remember judgments. The model further predicts that the rates of approach to asymptote should be similar once strength is equated by limiting the analysis to Remember and Know judgments made with high confidence. To test these predictions, the individual RTs that occurred during a recall period can be used to plot cumulative recall progress throughout the recall period. The cumulative plot shows the number of items recalled as a function of time spent recalling (which equals the sum of the RTs associated with the words that have been recalled thus far). When analyzed this way, the RTs are conceptualized as interresponse times (IRTs). Because the clock was stopped while the subject made confidence ratings and Remember/Know judgments, the recall time mainly reflects search time.

Figure 8 shows the cumulative recall functions for Remember and Know judgments (summed across participants) from the free recall condition of Experiment 3. Figure 8A shows the functions for all Remember and Know judgments, and Figure 8B shows the functions for high-confidence Remember and Know judgments. The curves drawn through the data show the least-squares fits of the standard 3-parameter exponential of the form R = a*(1 − e^{−(t − c)/τ}), where R is the cumulative number of words recalled, a is the estimated asymptote, τ is mean recall latency (the parameter that governs rate of approach to asymptote), and c is an offset parameter that reflects the average time taken to type a word. Clearly, both Remember and Know judgments exhibit the typical exponential rise to asymptote that has long been known to characterize free recall (and that has been interpreted to reflect a search process from an episodic memory search set based on relative strength).

Table 5 shows the parameter estimates associated with the best-fitting exponential functions shown in Figure 8. When all Remember and Know judgments are included in the analysis (R_1-5 vs. K_1-5), Know judgments exhibit a somewhat slower rate of approach to asymptote (consistent with the fact that they reflect weaker memories). However, when the analysis is limited to high-confidence Remember and Know judgments – when the strength of Remember and Know judgments is equated (R₅ vs. K₅) – their rates of approach to asymptote are equated as well. These results are consistent with the basic RT analyses presented earlier. More to the point, this pattern of results is as it should be if both Remember and Know judgments reflect retrieval from the same episodic memory search set.

Although our theory holds that both Remember and Know judgments in free recall result from a consciously controlled search proce ss and that they should exhibit similar recall dynamics when their strengths are equated, this should not be taken to mean that Remember and Know judgments are the same in all respects (even when they are equally strong). Remember and Know judgments do differ from each other in a theoretically significant way. However, according to this view, the difference between them is not that one reflects consciously controlled search and the other automatic memory. Instead, the difference is that some retrieved items are associated with source attributes (and receive Remember judgments), whereas others are not (and receive Know judgments). This view seems quite similar to the position taken by Hamilton and Rajaram (2003), who wrote: “Thus, Remember and Know responses in free recall may be based on access to specific attributes versus strength of item memory, respectively” (p. 66). It is also similar to the view espoused by Bodner and Lindsay (2003), who argued that, in conjunction with task demands, Remember and Know judgments in recognition memory are made based on the attributes of memory that are retrieved. If the source attributes that are retrieved help to solve the task at hand, a remember judgment is made; if not, a Know judgment is made (Gruppuso, Lindsay & Kelley, 1997).

Recently, McCabe et al. (2010) presented evidence in favor of the idea that Know judgments in free recall reflect automatic memory. They found that dividing attention at study had no effect on the proportion of words from the study list that were recalled and given a Know judgment (0.14 in the full-attention condition; 0.13 in the divided-attention condition). By contrast, dividing attention at study selectively and dramatically reduced the proportion of words from the study list that were recalled and given a Remember judgment (0.35 in the full-attention condition; 0.16 in the divided-attention condition). They interpreted these results to mean that retrieval during free r ecall occurs either through a consciously controlled (memory search) process, yielding Remember judgments when successful, or through an automatic process, yielding Know judgments when items simply pop into mind without conscious effort. The same automatic process, when it occurs during recognition, theoretically gives rise to the feeling of familiarity.

If Remember and Know judgments in free recall both reflect the outcome of a consciously controlled search process, then why did McCabe et al. (2010) find that dividing attention at encoding had no effect on the proportion of list items recalled with a Know judgment? One possibility is that dividing attention reduced the quality of all of the encoded traces such that some of the K items converted to unrecoverable ~ items and some of the R traces converted both to K traces (encoded successfully enough to allow the item to be recovered, but without source information) and to ~ traces. Because K traces are both lost and gained in this scenario, the overall effect might be no change in the proportion of list items receiving a Know judgment. The idea that traces initially supporting Remember judgments may later support Know judgments when memory has been weakened has been advanced in other contexts as well (e.g., Conway, Gardiner, Perfect, Anderson & Cohen, 1997; Knowlton & Squire, 1995). Here, we add the further idea that, in free recall, traces that initially support a Know judgment may later convert into an unrecoverable trace.

One possible way to empirically differentiate between the consciously controlled vs. automatic accounts of Know judgments in free recall would be to divide attention at retrieval. If Know judgments reflect consciously controlled search, as we assume, then the rate of approach to asymptote should be slowed to the same extent as for Remember judgments. If they instead reflect automatic retrieval, then it seems reasonable to predict that Know judgments would be slowed to a much lesser extent than Remember judgments.

Remember/Know Judgments in Recognition

Our findings do not speak directly to Remember/Know judgments in recognition, but they do raise an intriguing possibility, one that might have direct relevance to a longstanding debate about the role of the hippocampus in recollection and familiarity. In studies of recognition memory, one common interpretation of Remember/Know judgments (based on dual-process theory) is that Remember judgments reflect recollection and Know judgments reflect familiarity. Another common interpretation (based on signal-detection theory) is that Remember judgments reflect strong memory and Know judgments reflect weaker memory. Although much evidence supports the latter interpretation (e.g., Wixted & Stretch, 2004), Mickes and Wixted (2010) found that Know judgments were associated with less source memory than Remember judgments even when equally strong memories were compared (a finding that our current research shows holds true of free recall as well). Thus, once memory strength is equated in terms of confidence and accuracy (unfortunately something that is rarely done), the evidence supports a distinction that goes beyond memory strength.

Mickes and Wixted (2010) assumed that high-confidence Know judgments mainly reflect familiarity-based decisions in recognition. Here, we consider the alternative possibility that while high-confidence Know judgments reflect the subjective experience of familiarity, that experience may be based on an underlying memory process that is more akin to recall than it is to processes that are ordinarily thought to underlie familiarity, such as perceptual flue ncy or automaticity. Mo re specifically, our suggestion is that if high-confidence Know judgments in free recall reflect cue-dependent retrieval from episodic memory (like Remember judgments do), then there is no reason to assume that the same recall process does not occur for high-confidence Know judgments in recognition. This view accords with Tulving’s (1985) view of cue-dependent memory of recognition memory. For example, he wrote, “from the point of view of theory, and until such time as someone produces evidence to the contrary, we should assume that all retrieval is always cued” (p. 171). One need not fully embrace that strong view to accept the possibility that recognition sometimes consists of the cue-dependent retrieval of item-only information. Although the retrieval cue differs in the two tasks (in free recall, no explicit cue is provided; in recognition, a copy cue is provided), the end result may sometimes be the same, namely, the recall of item-only information.

Why, though, would item-only recall on a recognition memory task give rise to the subjective experience of familiarity – and to a Know judgment? One possible answer is that when a copy cue elicits the recall of item-only information in a recognition task, the subject may be aware that the recalled representation corresponds to the cue that retrieved it. The conscious awareness of this correspondence may be subjectively experienced as a strong sense of familiarity. Moreover, standard Remember/Know instructions would call for a Know judgment under these conditions because of the presence of item information and the absence of source information.

If this account is correct, then the subjective experience of recollection and familiarity would not align with the underlying memory processes giving rise to those experiences. If not, then the difference between two sides in a longstanding debate in the cognitive neuroscience literature may not be as great as it now seems. More specifically, a prominent view in the cognitive neuroscience literature is that the hippocampus selectively subserves recollection (e.g., Eichenbaum, Yonelinas and Ranganath, 2007). One piece of evidence supporting this view is that activity in the hippocampus is reliably elevated for Remember judgments (compared to the activity for misses or correct rejections) but is not reliably elevated for Know judgments (e.g., Eldridge et al., 2000). Others (e.g., Squire, Wixted & Clark, 2007) have interpreted this pattern to mean that, to detect activity in the hippocampus using fMRI, memory must be strong (indicated by high confidence and high accuracy). Because Know judgments typically reflect weak recognition memory (associated with low confidence and low accuracy, on average), elevated activity might not be detected for that reason alone.

Consistent with the latter interpretation, Smith, Wixted and Squire (2011) recently found that when activity in the hippocampus was compared for high-confidence Remember and high-confidence Know judgments (both of which involved similarly high levels of old/new accuracy), elevated hippocampal activity was evident for both. This finding was interpreted to mean that the hippocampus supports strong familiarity as well as strong recollection. However, if high-confidence Know judgments often reflect the outcome of an item-only recall process (even when memory is tested by recognition), it would mean that the hippocampus supports the subjective experience of familiarity, but it might nevertheless be true that the hippocampus supports a recall-like process (as opposed to a perceptual integration or perceptual fluency process). Moreover, this interpretation is consistent with the idea that, during encoding, the hippocampus serves an associative function, binding the studied item to contextual features (Eichenbaum et al., 2007). After all, it is the copy cue – together w ith contextual cues – that retrieve the item-only information in recognition memory. Sometimes, however, additional source information is not retrieved (at least not enough to warrant a Remember judgment), and that may be when (a) the subject experiences a strong sense of familiarity and (b) makes a Know judgment. Viewed in this light, the two sides in the debate over whether or not the hippocampus supports familiarity are not as far apart as is usually assumed.

We do not mean to suggest that the familiarity of an item on a recognition memory test always reflects the recall of item-only information. Our point instead is that Know judgments made with high-confidence and high accuracy may often be based on item-only recall, whether the test involves recall or recognition. Familiarity-based decisions that are generally made with lower confidence may be based primarily on other processes, such as perceptual fluency. Indeed, as others have suggested, there may be more than one kind of familiarity (Rugg & Yonelinas, 2003; Rugg & Curran, 2007), and this is what we are also suggesting here. In a review article critiquing the field of memory, Hintzman (2011) singled out the concept of familiarity as being too nebulous. About this, he wrote:

“Familiarity is routinely invoked in formal and informal explanations of memory as though it were a concept with obvious meaning, but the term appears to mean more than one thing. This may be a case where the scientific adoption of a term from everyday life conveys explanatory power that is largely an illusion. The field could benefit from a careful analysis of the ways in which the concept of familiarity has been used.” (p. 259)

Our findings using free recall raise the possibility that one form of familiarity may be more like recall than is ordinarily assumed.