Why seemingly more difficult test conditions produce more accurate recognition of semantic prototype words: A recognition memory paradox?

Consciousness and Cognition xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Consciousness and Cognition journal homepage: www.elsevier.com/locate/concog

Why seemingly more difficult test conditions produce more accurate recognition of semantic prototype words: A recognition memory paradox? ⁎

Jerwen Jou , Eric E. Escamilla, Andy U. Torres, Alejandro Ortiz, Martin Perez Jr., Richard Zuniga University of Texas – Rio Grande Valley, USA

A R T IC LE I N F O

ABS TRA CT

Keywords: Recognition memory Memory interference Distractor effects on recognition Differentiation of distractors from targets

Subjects studied Deese-Roediger-McDermott semantic-associate lists and took a recognition test. The makeup and number of test probes were manipulated. In Experiments 1 and 2A, one of three or all three distractors were semantically related to the list theme. In Experiment 2B, 6 or 30 related probes were used at test. Results showed that semantically related distractors and a longer list of test words both had a beneficial effect on the accurate discrimination of the prototype lures from the studied semantic associates and on the discrimination of studied from unstudied prototype words. These findings are inconsistent with predictions of memory interference and activation theories. We propose that the counterintuitive findings can be explained by the notion of old/new recognition as categorization learning and that relatedness and a larger number of test probes provide more accurate information about the prototype lure as a distractor, thereby improving its classification as a distractor.

1. Introduction In the Deese-Roediger-McDermott (Deese, 1959; Roediger & McDermott, 1995, DRM henceforth) paradigm, subjects study 15 semantically associated words (e.g., bed, rest, awake, tired, dream, wake, snooze, blanket, doze, slumber, snore, nap, peace, yawn, drowsy) but not the thematic or prototype word sleep. At test, they produce very high percentages of false recall or false recognition of the prototype words, sometimes at levels comparable to those of the studied words. In the standard DRM recognition-test paradigm, for each list of semantic associates, three target and three distractor words are generated as 6 test words. Of the three distractor words one is the prototype lure for the list, and two were words unrelated to the theme. In this study, we manipulated semantic relatedness of the two non-prototype distractors, and examined how it affected the recognition-memory performance for the prototype words (the critical words) in a modified Deese-Roediger-McDermott (DRM) (1995) paradigm. In addition, we varied the number of related test words (both targets and distractors) and examined its effect on the recognition performance of the prototype word. We now define what we mean by “more difficult test conditions” in the recognition test. There are two competing recognitionmemory theories that are relevant to the present purpose, the item noise and the context noise model. According to the class of item noise models of recognition memory derived from the concept of global matching as the process of recognition (Malmberg, Criss, Gangwani, & Shiffrin, 2012; Gillund & Shiffrin’s, 1984 SAM; Hintzman’s, 1988 MINERVA; Neely, Schmidt, & Roediger, 1983; Shiffrin

⁎ Corresponding author at: Department of Psychological Science, University of Texas – Rio Grande Valley, 1201 W. University Drive, Edinburg, TX 78539-2999, USA. E-mail address: [email protected] (J. Jou).

https://doi.org/10.1016/j.concog.2018.06.003 Received 1 January 2018; Received in revised form 30 May 2018; Accepted 4 June 2018 1053-8100/ © 2018 Elsevier Inc. All rights reserved.

Please cite this article as: Jou, J., Consciousness and Cognition (2018), https://doi.org/10.1016/j.concog.2018.06.003

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

& Steyvers, 1997 RAM; see Clark & Gronlund, 1996 for a review), the more numerous, and the more similar, the items on the study list are, the greater the item noise in memory, and the poorer the recognition performance will be. According to the context noise model (Dennis & Humphreys, 2001), the more numerous and the more similar the test contexts in which a test probe is previously presented, the more noise or interference the recognition of that probe will suffer and the lower the recognition accuracy for that probe will be. This is so because all the traces of the items encountered prior to a particular test item will be retrieved during the decision making for that probe and can cause interference for the recognition of that particular item (Criss, Malmberg, & Shiffrin, 2011). A reconciled position between the two opposing views reaches the conclusion that both item and context noises contribute to the decrement in recognition performance (Criss & Shiffrin, 2004). The essence of this principle is the same as the well-known concept of proactive interference that posits that when increasingly more similar items are studied and retained in memory in a continuous learning and test paradigm, memory interference builds up (Watkins & Watkins, 1975; Wickens, 1972). But when the study materials are changed to an unrelated category, memory gets a release from interference, a phenomenon known as the release from proactive interference (Wickens, 1972; Wickens, Born, & Allen, 1963; Wickens, Dalezman, & Eggemeier, 1976). Therefore, by more difficult test conditions we mean either a test condition that contains distractors that are semantically related to the targets compared to one in which they are unrelated to the targets, or a test condition that has a long list of related test words of both targets and distractors compared to one that has a short list of test words. 1.1. Purposes of the study and a review of related previous studies There are two main purposes in this study. The first purpose is to find out how, in general, distractor relatedness and test-word list length affect the recognition performance for the prototype lure in a semantic-associate list learning paradigm. The second purpose is to shed light on a recognition-test design feature that may have played an important role in creating the very high rate of false memory in the DRM paradigm. In this study, we used the DRM semantic associate lists as materials (Roediger & McDermott, 1995; Stadler, Roediger, & McDermott, 1999) to examine how the makeup and the number of test probes can affect the correct recognition of the prototype words and that of the semantic associates. Specifically, how does semantic similarity of distractors to the studied semantic associates as well as the number of test words affect the accurate recognition of the prototype lure and semantic associates? The prototype word in this paradigm has some memory property that makes it different from the semantic associates, namely, its strong potential to create vivid memory illusions (Roediger, 1996). This non-studied thematic word of a semantic-associate list is commonly called the critical lure in the DRM paradigm. Although we use the semantic-associate lists in the DRM paradigm as materials, the focus of this research is not to study DRM false memory in the traditional sense. Therefore, we followed the example in Whittlesea (2002) to call this word by a more generic term prototype lure since we are studying a broader issue than the DRM false memory. Because our aim was to study a more general issue of how the semantic relatedness of test probes and the number of test probe words affect the recognition of the prototype words (and, parenthetically, the recognition of the semantic associates) rather than the DRM false memory per se, our experimental procedures also differed considerably from the DRM procedures. The findings of the present study should help better understand the mechanism underlying the recognition of the semantic prototype words and the nature of the DRM false memory. Some previous studies have shown that when the number of semantic associates on the DRM study list is increased, false recall and recognition of the prototype lure also increases (Hancock, Hicks, Marsh, & Ritschel, 2003; Jou, Arredondo, Li, Escamilla, & Zuniga, 2016; Robinson & Roediger, 1997; for a review, see Neely & Tse, 2007). This finding suggests that activation spreading is likely the underlying mechanism generating the false memory (Roediger, Balota, & Watson, 2001; Underwood, 1965). Assuming that spreading activation is driving the DRM false memory, researchers then ask the question of whether the number of semantic associates prior to the presentation of the prototype lure on the list of test probes will also contribute to the generation of the false memory since the larger the number of semantic associates is processed before the prototype lure, the greater the level of spreading activation should be from the earlier processed test words. The important question is whether this effect of spreading activation derives only from the encoding phase of the task or from both the encoding and the testing phases. In addition, the level of spreading activation can be a function of not only the number of semantic associates one encodes, but also a function of the strength of the semantic association of the words. Therefore, whether the test probes are related or unrelated to the target words may also be a contributing factor if indeed spreading activation is the underlying mechanism generating false recall and recognition of the prototype lure. Regarding how the composition of the test probes affects the recognition performance of the prototype lures as well as the semantic associates, some previous study (Gunter, Ivanko, & Bodner, 2005) found that related distractors in the DRM paradigm, in comparison to unrelated distractors, can lower the FA rate of the prototype lures. Moreover, Gunter et al. (2005) suggested that the effect is the same as that of giving warnings to subjects, since both procedures make subjects more cautious in endorsing a probe item. They further indicated that using related distractors lowers not only the FA rate of the prototype lures, but also the hit rate of the studied semantic associates, which, they argued, indicates that the related distractors do not increase the discriminability between the prototype lures and the studied semantic associates. We found that although Gunter et al. (2005) examined the distractor relatedness effect on the FA rate of the prototype lures, they did not look into the possibility that the lower FA rate of the prototype lures in the related-distractor condition could possibly be due to an improvement in the discrimination between the studied semantic associates and the prototype lures, not just due to a raise in the judgment criterion. When all the distractors (except for the prototype lure) are unrelated to the studied semantic associates as in the standard DRM paradigm (Roediger & McDermott, 1995, Experiment 2), subjects may use categorical or thematic information to decide on whether the test probe is on the list or not. The heuristic they use may be: If it is associated with the theme, it is probably on 2

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

the studied list; if it is not associated with the theme, it is probably not on the studied list. However, when the other distractors (besides the prototype lure) are also related to the theme, subjects may likely become aware that not all distractors are unrelated words, and hence they may use item-specific information, rather than categorical information, as the basis for making an old/new judgment. If that is the case, then the decision will be based on more valid information than categorical relatedness. If subjects metacognitively switch from a predominantly category-relatedness judgment in the unrelated-distractor condition to a predominantly item-specific judgment mode in the related-distractor condition, then, there should be an improvement in the discrimination between studied words and the prototype lures, rather than, or in addition to, a mere raise in the decision criterion, with no improvement in discrimination. Thus, one main specific goal of this study is to determine whether the process underlying the lowering of the prototype lure FA rate in the related-distractor condition is simply an upward adjustment of the decision criterion, with no improvement in discrimination between the prototype lures and the studied semantic associates, or whether it is more than a mere criterion shift, i.e., a criterion shift in tandem with an increase in discriminability between the two types of words. The accuracy of discrimination between the prototype lure and a studied semantic associate can be measured by a d′ score based on the hit rate of the studied semantic associates as targets and the FA rate of the prototype lures as distractors. That a drop in the prototype lure FA rate is accompanied by a concomitant drop in the hit rate of the studied semantic associates in the related-distractor condition (compared with the unrelated-distractor condition) does not necessarily mean that there is no improvement in discriminability. Whether there is an improvement in discrimination between the prototype lure and the studied semantic associate is determined by how large the decrease in the prototype lure FA rate is relative to how large the decrease in the hit rate of the studied word is. A d′ score calculated by pitting the hit rate of the semantic associates against the FA rate of the prototype lures will reveal whether there is an improvement in the discrimination between these two types of words when the unrelated distractors are replaced by related distractors. In Experiment 1, we found that distractor relatedness lowered the FA rate of the prototype lures, but, in addition, made a determination on whether there was also an improvement in the discrimination between these two types of words above and beyond a raise in decision criterion. If replacing the unrelated distractors with related distractors can achieve this result, then, it has an effect of more than making subjects more cautious; it also makes them more discerning and knowledgeable in separating these two types of words. In addition to examining the effect of distractor relatedness on the discrimination of the prototype lures from the studied semantic associates, we designed Experiments 2A and 2B in a way by which we could test for old/new discrimination within the prototype words themselves by presenting half of those words and not presenting the other half. The question of interest is whether manipulating the relatedness of distractors will also enhance the discrimination of the studied from the unstudied prototype words. This within-word-type discrimination comparison can also be done for the semantic associates to see whether the distractor-relatedness and the test probe-list length manipulation also affect the discrimination between the old and new semantic associates in the same way. One advantage of making this comparison is that it holds the baseline familiarity (or activation) level of the target and the distractor constant so that the effect of studying the words can be assessed more purely. The goal of the second manipulation performed in the present study was to investigate an unsettled issue of how the number of related test words affects the FA rate of the prototype lure. Results from the previous studies have been mixed. Some studies have found that increasing the number of related test words prior to the presentation of the prototype lure at test can increase the FA rate of the prototype lure (Coane & McBride, 2006, found an effect when related test words increased from 0 to 6 prior to the prototype lure, but no effect with further increase; Dewhurst, Knott, & Howe, 2011, found an effect only when not invoking a monitoring mechanism; Marsh & Dolan, 2007 found an effect only when decision making was speeded). Other studies have found that it has no effect on the FA rate of the prototype lure (Dodd, Sheard, & MacLeod, 2006; Marsh & Dolan, 2004 when test was self-paced; Marsh, McDermott, & Roediger, 2004). The reason for the inconsistent findings about the effects of number of related test words presented prior to the prototype lure is not known given that the materials and procedures in these studies varied extensively. In Experiment 2B, we manipulated the number of related test words from 6 in the short-test condition to 30 in the long-test condition. Although in Experiment 2B we did not precisely control the number of related test words presented before the presentation of the prototype word (since we used a random test-word presentation order), the average number of test words preceding the occurrence of the prototype word in the test should be larger in the long-test condition than in the short-test condition. With such a drastic manipulation, if the number of related test words can produce an effect on the false recognition of the prototype lure, whether in the form of a benefit or an impairment, the effect should be detected. If still no effect is produced by such a drastic manipulation, then it would be likely that priming or spreading activation during testing plays no role in generating false memory, and that false memory probably derives only from the encoding phase as some researchers suggest (Dewhurst, Bould, Knott, & Thorley, 2009; Dodd et al., 2006). Hopefully, the results will help settle the so-far still disputed issue of whether false memory derives only from encoding or from both encoding and test-induced priming. 1.2. Old/new recognition as categorization and the predictions We predict that the FA rate for the prototype lure will be lower in the related-distractor than in the unrelated-distractor test condition and that it will also be lower in the long-test than in the short-test condition. These predictions are based on the concept of old/new recognition as a form of categorization learning. The more accurately the prototype lure can be categorized, the lower will the FA rate be. According to the class of models conceptualizing old/new recognition as categorization, a probe in a recognition test is compared with either the prototype or individual exemplars (depending on the specific categorization model one espouses) of the target and the distractor category, respectively, and assigned to the category to which the probe has overall greater similarity (Estes & Maddox, 1995; Hirshman, 1995; Nosofsky, 1988). For the present purpose, the differentiation between the prototype and the 3

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

Criterion

Prototype Lures

Unrelated Distractors

SemanticAssociate Targets

Strength-of-Evidence in Memory Note: Prototype lure distribution is closer to the semantic-associate target distribution than to the unrelated distractor distribution Fig. 1A. In the unrelated-distractor condition, the prototype-lure distribution is closer to the studied semantic-associates than to the unrelated distractors.

exemplar models is not crucial since both make the same prediction. Again, according to the exemplar model of categorization, the old/new decision on a test probe in a recognition test is based on the relative total similarities of the test probe to all the exemplars in the target or the distractor category (Estes & Maddox, 1995; Nosofsky, 1988). Thus, when the majority of the distractor category is unrelated to the theme, the prototype lure has lower overall similarity to the exemplars in the distractor category than to those in the target category. In other words, the prototype lure distribution is situated closer to the target distribution than to the distractor distribution (see Fig. 1A). When all of the exemplars in the distractor category are related to the theme, the prototype lure becomes more similar to those exemplars compared with those in the unrelated-distractor condition. In other words, the prototype-lure distribution shifts a distance to the left from the unrelated- to the related-test condition (see Fig. 1B). Basically the same concept can be applied to explain the beneficial effect of the increased number of related test words on the recognition performance for the prototype words. When 15 related distractors are presented at test, the prototype lure is rendered even more similar to a representative distractor than when only 3 related distractors are presented, again, causing the prototype-lure distribution to be shifted further to the left. Some researchers suggest that people learn how to accurately classify a probe either as a target or as a distractor by constructing a representation of the target and the distractor distribution, using information available in the study list and in the test list (Estes & Maddox, 1995; Hirshman, 1995). During the study in the short-test condition, subjects saw 15 list words, but at test they saw only 3 distractors. In a certain sense, they were exposed to the “population” of the target distribution, but only a small random sample (one fifth) of the distractor population in the short-test condition compared to the long-test condition. When more test words are presented, more knowledge about the distractor distribution is obtained and hence a more accurate representation of the distractor distribution can be established. One theory in the class of categorization learning models is the range theory (Hirshman, 1995). According to that theory, people need to estimate the range (the highest and the lowest points, see Parducci, 1984) of the old- and new-item distributions to be able to accurately calibrate the criterion placement. By providing the full set of distractors in the longCriterion

Prototype Lures Related Distractors

Prototype Targets

Strength-of Evidence in Memory Note: When unrelated distractors were replaced with related distractors, prototype lure distribution shifted to the left Fig. 1B. In the related-distractor condition, the prototype-lure distribution shifts toward left and becomes more separated from the studied semantic-associates. 4

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

test condition instead of a fragment of the possible distractors, a memory representation of the target and the distractor distribution can be formed that more accurately reflects the actual difference between the two distributions, thereby generating additional benefit for the accurate recognition of the prototype lure. The same idea can be put in slightly different terms. When the number of test words is increased, the variation of the distractor distribution is increased (just as the range is increased). When the variation of the distractor distribution is increased, the prototype lure can be more easily assimilated into the distractor distribution. When more instances are sampled from a category, the chance that an atypical item is included is increased. For example, a pumpkin is an atypical instance of the fruit category because of its high similarity to vegetables. Just like a pumpkin, the prototype lure is a poor instance of the distractors because of its high similarity to the targets. However, when the set of distractors gets larger, the chance that other atypical instances (words of very high relatedness) are included is increased. As such, subjects can have a better chance of finding highly related distractors in the long test list that resemble the prototype lure than in the short test list, thereby increasing the probability of correctly classifying the prototype lure as a distractor. It is analogous to the benefit of using more varied instances in concept acquisition training (Schmidt & Bjork, 1992). In General Discussion, we also consider whether our findings may be explained by the activation/monitoring theory (Gallo & Roediger, 2002; Roediger et al., 2001), the fuzzy-trace theory (Brainerd & Reyna, 1998; Brainerd, Reyna, & Brandse, 1996; Reyna & Brainerd, 1995) of false memory, and the dual-process theory of recognition (Jacoby, 1991; Yonelinas, 2002). 2. Experiment 1 In the standard DRM procedure, six test probes are created for each list of the semantic associates. Three of the 6 test probes are targets comprised of studied list words and the other three words are distractors. Of the three distractors, one is the critical (prototype) lure and the other two are unrelated distractors. The purpose of Experiment 1 was to determine what effects changing the two unrelated distractors in the DRM paradigm to two related distractors could have on the recognition of the prototype lure and on the recognition of the semantic associates of a DRM list. 2.1. Method 2.1.1. Subjects Ninety-five introductory psychology students (65 females and 30 males) participated in the unrelated-distractor, and 90 (62 females and 28 males) in the related-distractor condition of the experiment for course credit. Data of two subjects in the unrelateddistractor condition apparently resulted from random key presses at test (the accuracy was close to 50% or chance level) and were excluded from analysis, leaving 93 subjects’ data in the unrelated-distractor condition for analysis. 2.1.2. Materials and design The two test conditions (unrelated versus related distractors) were a between-subjects factor. The materials and procedure at the study phase were the same for the two conditions but the testing procedures were different. Twenty-four lists of DRM words (of 15 words each, not including the prototype word) were adopted from the Roediger and McDermott’s (1995) study. Odd-numbered subjects studied and were tested on half of them (i.e., 12 lists), and even-numbered subjects on the other half of the 24 lists. The procedure by which the list words were presented for study, the selection of the distractor words for the recognition test, and the procedure of presenting the test words in the test were modeled after Roediger and McDermott’s (1995) Experiment 1 with the following modifications. The first change was that subjects learned one list of semantic associates rather than multiple lists before they took the recognition test. And this cycle was repeated for each of the 12 lists. Second, at the study phase, the list words were presented one at a time in a new random order for each list and each subject rather than in the same descending order of backward association strength of the semantic associates. This was done to eliminate the serial position effect as a potential confounding variable (e.g., the item at the beginning of the list might receive the benefit of a primacy effect at test). For the unrelated-distractor condition, the test words for a studied list were composed by following the DRM paradigm (Roediger & McDermott, 1995). They consisted of 6 words for each list of which one was the prototype lure for that particular list, one word selected from position 1, one word from position 8, and one word from position 10 of the studied list, and the remaining 2 distractor words were randomly chosen from a set of 24 nonpresented words (consisting of 6 prototype words for 6 nonstudied lists, and 18 other distractor words, 3 of which from each of the 6 nonstudied lists, 1 word from position 1, 1 from position 8, and 1 from position 10 of each of the 6 lists) obtained from the 12 nonpresented of the 24 lists in Roediger and McDermott (1995). Thus, in this design, all nonpresented distractor words except the prototype lure for a studied list were unrelated distractors. Here is an example of the 6 test words in the unrelated-distractor condition: mad (studied), wrath (studied), fight (studied), anger (prototype lure), butter (unrelated distractor), thimble (unrelated distractor). For the related-distractor test condition, the studied list words in the test were generated in the same way as in the unrelated distractor condition. Again, there were three distractors. One of the three distractors was the prototype lure, and the other two were obtained from the expanded DRM semantic-associate lists used in Jou, Matus, Aldridge, Rogers, and Zimmerman (2004, see the Appendix for those expanded lists). In the expanded version of a DRM word list, 14 additional semantic associates were added to each of the original DRM lists, making a total of 29 semantic associates per list, not counting the prototype word (if the prototype word is counted, the expanded list is 30-word long). The two related distractors used in the related-distractor condition were randomly sampled from the additional 14 semantic associates of the expanded list. Jou et al. (2004) selected these 14 additional semantic associates from Russell and Jenkins’ (1954) associated word norms. Roediger and McDermott (1995) developed their word lists by 5

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

taking the top 15 words in a chosen list of the Russell and Jenkins’ (1954) original word association norm. We chose our additional 14 words from each corresponding list in Russell and Jenkins’s norm by generally picking the next 14 words from that list (with some exceptions, e.g., to skip a word or two in a list to avoid already used words). The first 15 words were listed in the same order as in Roediger and McDermott’s (1995) study, and the additional 14 words that followed the first 15 words were listed in the same order as in Russell and Jenkins’ (1954) norm. Here is an example of the 6 test words in the related-distractor condition: mad (studied), wrath (studied), fight (studied), anger (prototype lure), sad (related distractor), yell (related distractor). 2.1.3. Procedure Under both test conditions, the 15 words of a list were presented in a random order one at a time at a pace of 2.5 s per word in the center of a computer monitor with a 1 s blank screen between the presentation of one word and the next. The order in which the lists were presented for learning was also random. Subjects were asked to memorize these words as best as they could. At the end of the presentation of a list of words, subjects pressed the enter key to start a 20-s backward counting task before the recognition test. The backward counting started with a random 3-digit number and subjects counted it down by 3 at a time. Subjects produced the next number by typing the number onto a blank space on the screen and pressing the enter key. When the produced number was wrong, an error message was displayed accompanied by a warning tone and an instruction to re-count from the preceding number. When the 20-s counting expired, they pressed the enter key to start the recognition test. During the recognition test, the 6 probe words were displayed in the center of the monitor one at a time in a new random order for each list and for each subject. Oddnumbered subjects were instructed to press the ‘z’ key if they judged that they saw that word at study, and the ‘/’ key if they judged they did not see that word. The mapping of the responses to the keys was reversed for the even-numbered subjects. They were provided with an index card with the word “Yes” on it to be put on either the left or right side of the keyboard to mark the positive response side in case they forgot which key was the “Yes” response key. There was a 1-s blank screen between the end of pressing the response key and the appearance of the next probe word. Subjects were told that accuracy of their responses was very important and that if their responses were random, the computer would detect the random pattern, and consequently they would be asked to repeat the experiment. However, nobody actually repeated the experiment. They were also told that they should not take a rest while the test word was being displayed but could take a brief break at the end of one round of test before they started on learning the next list of words. 2.2. Results and discussion The mean hit rate, FA rate, d′, and criterion measure c are presented in Table 1 as a function of experiment and the test probe type. The d′ score was calculated by subtracting the z score of the FA rate of distractors from the z score of the hit rate of targets (d′ = z (HR) − z(FAR)), and the c score was calculated by multiplying the sum of the above two z scores with −0.5 (c = −0.5 (z(HR) + z (FAR)) (Macmillan & Creelman, 1991). The mean FA rate for the prototype lure in the unrelated-distractor condition was 0.765, a typical prototype-lure FA rate in the standard DRM paradigm (see Stadler et al.’s, 1999), although in this experiment only one list of words was studied and tested at a time rather than multiple lists studied and tested at a time. The mean prototype-lure FA rate in the related-distractor condition was 0.575, significantly lower than in the unrelated-distractor condition (0.765), F (1, 181) = 33.67, MSE = 0.049, p < .0001 ηp2 = 0.157. Thus, when the two unrelated distractors in the three distractors were replaced by two related distractors, the false recognition rate of the prototype lure decreased significantly. This finding was consistent with that of Gunter et al. (2005). However, the question not investigated in previous studies is whether this decrease in the prototype-lure FA rate is due to a mere raise in decision criterion with no improvement in the discrimination between the prototype lure and the studied semantic associates, or a raise in criterion coupled with an improvement in the discrimination between the two types of words. The answer to this question can be obtained by computing the d′ score using the hit rate of the studied semantic associates and the FA rate of the prototype lure.1 (d′ calculated by treating the prototype lure as a distractor is marked with an asterisk in Table 1). The mean hit rate for the studied semantic associates was 0.872 in the unrelated-distractor condition, and 0.818 in the relateddistractor condition, respectively. The decrease was significant, F (1, 181) = 11.67, MSE = 0.012, p = .0008, ηp2 = 0.060. Thus, the related distractors, relative to the unrelated distractors, significantly lowered the hit rate of the studied semantic associates. However, because the drop in the FA rate of the prototype lure was even larger (from 0.765 to 0.575), there was a significant increase in d′ (discriminability) for these two types of words from the unrelated- to the related-distractor condition (from d′ = 0.394 to d′ = 0.763), F (1, 181) = 10.69, MSE = 0.582, p = 001, ηp2 = 0.056. We then used the same hit rate and the FA rate used for computing the d′ score to compute the corresponding criterion measure c. A c value of zero indicates a neutral criterion or no bias. A positive value indicates a conservative bias and a negative value indicates a liberal bias (Macmillan & Creelman, 1991). There was a significant increase in c from the unrelated- to the related-distractor condition (from c = −1.01 to c = −0.58, a decrease in the absolute value of the minus number, hence a decrease in liberal bias) (c calculated by treating the prototype lure as a distractor is marked with an asterisk in Table 1), F (1, 181) = 47.96, MSE = 0.180, p < .0001, ηp2 = 0.209. Thus, the effects of using related distractors were both improving the discrimination between the prototype lures and the studied semantic associates and raising the decision criterion in endorsing the studied semantic associates. As far as the discrimination between the studied semantic associates and unrelated distractors versus related distractors is 1 To avoid an infinite z value in computing the d′s, all hit and false alarm rates were corrected by adding 0.5 to the frequency of hits or false alarms, and dividing this adjusted frequency by N + 1 where N was either the number of old or new trials (Snodgrass & Corwin, 1988).

6

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

Table 1 Mean hit rate, FA rate, d′, c, and the significance of the difference for each measure across the two test conditions as a function of experiment and probe type. Experiment 1 Probe Type

Resp Measure

Unrelated Distr Cond

Related Distr Cond

Difference

Prototype lure

FAR

0.765 (0.024)

0.575 (0.022)

sig

Semantic Associate

HR FAR d′* c*

0.872 (0.011) – 0.394 (0.083) −1.01 (0.050)

0.818 (0.017) 0.113 (0.011) 0.763 (0.077) −0.58 (0.037)

sig

FAR

0.015 (0.004)

–

Unrelated Distractor

sig sig

Experiment 2A Probe Type

Resp Measure

Unrelated Distr Cond

Related Distr Cond

Difference

Prototype Word

HR FAR d′* c*

0.934 (0.012) 0.679 (0.026) 0.727 (0.065) −0.863 (0.047)

0.924 (0.011) 0.562 (0.022) 1.03 (0.064) −0.675 (0.035)

ns sig sig sig

Semantic Associate

HR FAR d′ c d′* c*

0.831 (0.011) 0.152 (0.017) 1.94 (0.063) −0.047 (0.035) 0.518 (0.068) −0.759 (0.049)

0.830 (0.010) 0.085 (0.008) 2.45 (0.068) 0.210 (0.026) 0.854 (0.074) −0.588 (0.033)

ns sig sig sig sig sig

Unrelated Distractor

FAR

0.052 (0.006) Experiment 2B

Probe Type

Resp Measure

Short-Test Cond

Long-Test Cond

Difference

Prototype Word

HR FAR d′* c*

0.924 (0.019) 0.562 (0.022) 1.03 (0.064) −0.675 (0.035)

0.905 (0.012) 0.476 (0.022) 1.21 (0.060) −0.510 (0.036)

ns sig sig sig

Semantic Associate

HR FAR d′ c d′* c*

0.830 (0.010) 0.085 (0.008) 2.45 (0.068) 0.210 (0.026) 0.854 (0.074) −0.588 (0.033)

0.842 (0.008) 0.112 (0.006) 2.36 (0.058) 0.107 (0.022) 1.17 (0.066) −0.489 (0.032)

ns sig ns sig sig sig

Note. The number in the parentheses is the standard error of mean. sig = significant; ns = nonsignificant; Distr = Distractor; HR = Hit Rate; FAR = False Alarm Rate; Resp = Response. d′* and c* were calculated by using the FA rate of the prototype lures as distractors.

concerned, it is expected that the former is better than the latter. As expected, mean d′ as an index of discrimination between the studied semantic associates and the unrelated distractor was 3.13, and that between the studied semantic associates and the related distractors was 2.22. The difference was significant, F (1, 181) = 93.14, MSE = 0.406, p < .0001, ηp2 = 0.337. It is also expected that the criterion measure c should be higher (more conservative) when the distractor was unrelated than related due to a higher rate of rejecting unrelated distractors compared to rejecting related distractors. Indeed, mean c of 0.355 in the unrelated-distractor condition was significantly higher than that of 0.151 in the related-distractor condition, F (1, 181) = 22.79, MSE = 0.084, p < .0001, ηp2 = 0.112. In sum, the effect of the distractor relatedness on the discrimination of the prototype lures from the studied semantic associates was a beneficial one. More important, the related distractors did not only make subjects more cautious in endorsing a test item by adopting a more strict criterion; it also made them better able to correctly discriminate the prototype lures from the studied semantic associates. One possible mechanism of causing this performance improvement may be that the related distractors provided subjects useful information about the attributes of the prototype lure (e.g., the distractor could actually be similar to the targets). 3. Experiment 2A The results of Experiment 1 indicated that related distractors, relative to unrelated distractors, lowered the FA rate of the prototype lure, and, additionally, enhanced the discriminability between these words and the studied semantic associates. A further question of interest is whether the relatedness of the distractors also helps the old/new discrimination within each type of words, i.e., 7

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

between studied and unstudied prototype words, and between studied and unstudied semantic associates. To answer this question, we presented half of the prototype words and half of the semantic associates at study in this experiment, so that both the hit rate and the FA rate can be obtained to calculate d′ for each type of words. Another purpose of Experiment 2 was to determine how the distractor relatedness manipulation would affect the discrimination of the studied semantic associates from the unstudied ones. 3.1. Method 3.1.1. Subjects A total of 250 introductory psychology students (155 females and 95 males) participated in this experiment, 122 in the unrelateddistractor and 128 in the related-distractor condition, for course credit. The data of two subjects in the unrelated-distractor, and of three subjects in the related-distractor condition showed signs of random responses and were excluded from analysis. 3.1.2. Materials, design, and procedure As in Experiment 1, there were two test conditions, the related- and the unrelated-distractor test conditions. Both the study and the test parts of this experiment were changed from Experiment 1. However, the study part across the two test conditions of this experiment was the same. The testing part of the experiment was different across the two conditions. For odd-numbered subjects, the prototype words of the odd-numbered lists (a total of 6 lists) were presented at the study phase, and the prototype words of the evennumbered lists (also 6 lists) were not presented at study. For even-numbered subjects, the arrangement was reversed. In this design, although it was not the case that one prototype word was both studied and unstudied by the same subject (which was not possible), the same prototype word was studied by half of the subjects, and not studied by the other half of the subjects. When the prototype word was presented at study, only 14 of the 15 words on the original DRM list were presented for study (one word was randomly selected to be omitted from study presentation), which made the number of studied words to be still 15 (1 prototype word plus 14 semantic associates). The presented prototype word was always used to make up one of the three target words at test for the list concerned. The other two target words were randomly selected from the 14 studied semantic associates of the original DRM list. We randomly selected the target words (except the prototype target) in this experiment rather than use the fixed target words as in the standard DRM paradigm in order to make the target words overall more representative of the whole list of the semantic associates. When the prototype word was not presented at study, it was always used at test as the prototype lure along with the other two distractors, and the three target words were randomly selected from the 15 words of the original DRM list. In the unrelated-distractor condition, if the prototype word was not presented at study, it was used as the single related distractor. The other two distractors (unrelated) were selected in the same way as in Experiment 1, and the three target words were randomly selected from the 15 studied semantic associates of the original DRM list. If the prototype word was presented at study, it was used as one of the three studied target words and the other two target words were randomly selected from the 14 studied words of the DRM list. In this condition, one semantic associate was randomly selected from the expanded section of the list to serve as the single related distractor of the distractor trio. In the related-distractor condition, if the prototype word was studied, it was used at test as one of the three target words. The other two target words were randomly selected from the 14 studied semantic associates of the original DRM list (the studied prototype word displaced a randomly selected word on the 15-word DRM list). The three related distractors for the list were randomly sampled from the 14 unstudied semantic associates in the added section of the expanded list. If the prototype word was not studied, it was used as the prototype lure with the other two related distractors randomly selected from the 14 additional semantic associates on the expanded section of the list to make up the three related distractors. The three target test words were randomly selected from the 15 studied semantic associates of the DRM list. The test procedure was the same as in Experiment 1. 3.2. Results and Discussion The mean hit rates, FA rates, mean d′, and mean c as a function of test condition (unrelated- versus related-distractor) and testprobe type are presented in Table 1. The mean hit rate of the prototype words in the unrelated-distractor condition was 0.934, and that in the related-distractor condition was 0.924. The difference was not significant, F (1, 243) = 0.41. The mean FA rate for the prototype lures in the unrelated-distractor condition was 0.679, which was significantly lower than that of 0.765 in Experiment 1, F (1, 211) = 5.45, MSE = 0.071, p = .021, ηp2 = 0.025. One possible cause of the lowering of the prototype-lure FA rate in Experiment 2A was that in Experiment 2A, subjects learned what the familiarity level of an actually studied prototype word was relative to that of an unstudied prototype word, thereby increasing the correct rejection rate of the prototype lures. This contrast was not available to the subjects in Experiment 1. The mean FA rate of 0.679 in the unrelated-distractor condition was, however, significantly higher than that of 0.562 in the related-distractor condition, F (1, 243) = 11.63, MSE = 0.071, p = .0008, ηp2 = 0.046, which was consistent with the corresponding result in Experiment 1. Thus, although the relatedness of the distractors did not have an effect on the hit rate of the prototype words, it significantly lowered the FA rate of the prototype lures, which suggested an improvement in the prototype words’ old/new discriminability. Indeed, a comparison between the two conditions on the prototype-word d′ confirmed that there was a significant increase in mean d′ from the unrelated- to the related-distractor condition. Mean d′ of the prototype words in the unrelated-distractor condition was 0.727, and that of the related-distractor condition was 1.03. The difference was significant, F (1, 244) = 10.98, MSE = 0.505, p = .001, ηp2 = 0.043. Given that there was not a significant decrease in the prototype word hit rate, but a significant decrease in its FA rate, the increase in d′ from the unrelated- to the related-distractor condition was more compatible with an account of a leftward shift of the prototype-lure distribution than an account that attributes the increased d′ solely to a 8

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

criterion rightward shift. More detailed discussion of this point is given in General Discussion. We also calculated criterion c measure to compare the two criteria subjects adopted in judging the prototype words across the two testing conditions. The mean criterion c of the prototype words in the unrelated-distractor condition was −0.863, and that in the related-distractor condition was −0.675. The difference was significant, F (1, 243) = 10.46, MSE = 0.208, p = .001, ηp2 = 0.041. Thus, when all the distractors were related words, the criterion used in judging the prototype words became less liberally biased compared with when two of the three distractors were unrelated words. However, this criterion raise was coupled with a d′ increase, indicating that the old/new discriminability of the prototype words also improved. One related question also of interest is how the distractor relatedness manipulation affected the recognition of the non-prototype related distractors (there was one single such related distractor in the unrelated-distractor condition when the prototype word was presented at study, and two and three such distractors in the related-distractor condition when the prototype word was not presented at study and was presented at study, respectively). It turned out that making all distractors related also lowered the FA rate of the non-prototype related distractors (semantic-associate distractors). The mean FA rate of the related distractors in the unrelateddistractor condition was 0.152 and that in the related-distractor condition was 0.085. The difference was significant, F (1, 243) = 13.12, MSE = 0.021, p = .0004, ηp2 = 0.051. This result was consistent with that of the prototype lures in that replacing the unrelated distractors with related distractors lowered the FA rate of the prototype lure, which was actually also a related distractor (but a super related one). In sum, the effect of using related distractors seemed to be improving the recognition accuracy of the prototype words as well as the semantic associates. Next, we looked at the effect of the distractor relatedness on the old/new discriminability of the semantic associates. In the calculation of d′ and c for the semantic associates, the prototype lure was not counted as a distractor (since the prototype lure was far more “related” than the semantic-associate lures and already counted as a distractor for assessing the prototype word recognition). Neither were the unrelated distractors in the unrelated distractor condition counted as distractors in the calculations of the semanticassociate d′ and c since they were qualitatively very different from the semantic associates. The mean hit rate for the semantic associates was 0.831 under the unrelated-distractor condition, and 0.830 under the related-distractor condition, respectively. The difference was minuscule and not significant, F (1, 243) ≈ 0. The corresponding FA rate was 0.152 and 0.085, for the unrelated and related-distractor condition, respectively. The difference was significant, F (1, 243) = 13.12, MSE = 0.021, p = .0004, ηp2 = 0.051. Since the hit rates of the two conditions were virtually the same, the decreased FA rate in the related-distractor condition should have a positive effect on the d′ score of the semantic associates in the related-distractor condition. Indeed, mean d′ of the related-distractor condition (2.45) was significantly higher than that in the unrelated condition (1.94), F (1, 243) = 30.36, MSE = 0.526, p < .0001, ηp2 = 0.111. Mean c of 0.210 in the related-distractor condition was significantly higher than that of −0.047 in the unrelateddistractor condition, F (1, 243) = 34.63, MSE = 0.117, p = .0001, ηp2 = 0.125, indicating that subjects were more conservative in their recognition decision in the related- than in the unrelated-distractor condition. However, since the hit rates were the same across the two conditions, the decreased FA rate was likely the result of a leftward shift of the distractor distribution from the unrelated to the related-distractor condition. Thus, changing the distractor words from one related to three related words raised d′ as well as the criterion for the recognition of the semantic associates just as it did for the prototype words. The manipulation benefited the old/new discrimination for the prototype words as well as for the semantic associates in a similar way. But how does distractor relatedness affect the discrimination of the studied semantic associates from the prototype lures? As in Experiment 1, we used the hit rate of the studied semantic associates as targets and FA rate of the prototype lures as distractors to compute d′ score for the two distractor conditions (those d′s were marked with an asterisk in Table 1). The mean d′ score for the unrelated-distractor condition was 0.518, and for the related-distractor condition was 0.854. The difference was significant, F (1, 243) = 11.25, MSE = 0.617, p = .0009, ηp2 = 0.044. Thus, there was a significant increase in discriminability between these two types of words from the unrelated- to the related-distractor condition. It is also expected that there was an increase in criterion involving the recognition of these two types of words. Indeed, mean c of the unrelated-distractor condition (−0.759) was significantly more liberal than that of the related-distractor (−0.588) condition (those c’s were marked with an asterisk in Table 1), F (1, 234) = 8.45, MSE = 0.210, p = .004, ηp2 = 0.034. In sum, related distractors compared with unrelated ones had a beneficial effect on the discrimination between the studied and the unstudied prototype words, as well as between the prototype lures and the studied semantic associates. 4. Experiment 2B Experiment 2B was conducted to determine how the number of test words in the test affects the recognition of the prototype word and semantic associates. In a previous study (Jou, Escamilla, Pena, Zuniga, Perez, & Garcia, 2018), the FA rate of the prototype lure dropped to below 0.50 when the test words were made up of all the words on the expanded list (one prototype word, 15 semantic associates of the DRM list plus the 14 added semantic associates of the expanded list). However, in that study, the number of test words was not manipulated. In Experiment 2B, the effect of the number of test words on the recognition of the prototype words and semantic associates was systematically examined. Only related distractors were used in this experiment and hence all test words were semantically related. The short test condition used 6 test words and long test condition used 30 test words. 4.1. Method 4.1.1. Subjects One hundred and twenty-five subjects’ data collected in the related-distractor condition of Experiment 2A were re-used as the data 9

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

for the short-test condition in Experiment 2B. For the long-test condition, 135 introductory psychology students, 85 females, and 50 males, were recruited to generate new data for Experiment 2B. The data of 5 of the 135 subjects showed signs of random responses and were excluded from analysis. 4.1.2. Materials, design, and procedure There were two conditions, a short- and a long-test condition. The study part of the two conditions was exactly the same and the same as in Experiment 2A. For the short-test condition of this experiment, we simply re-used the data of the related-distractor condition of Experiment 2A (with 6 test words per list). Although the short-test condition was originally conducted as a part of Experiment 2A, the short and long-test conditions were both run within the same semester. Again, for odd-numbered subjects, the prototype words of the odd-numbered lists were presented for study, and for even-numbered subjects, the prototype words of the even-numbered lists were presented for study. Thus, each prototype word was studied and unstudied across the subjects. There were 30 words in the long test-probe list, of which one word was the prototype word, 15 others were on the original DRM list, and remaining 14 from the expanded section of the list. In the long-test condition, all 30 words on the expanded list were presented at test as probes, 15 of which were studied, serving as target words, and 15 of which as distractors. When the prototype word was presented at study, one semantic associate from the original 15-word DRM list was randomly selected to be omitted from presentation at study. The remaining 14 words plus the prototype word were presented as 15 target words at test. The randomly selected nonpresented semantic associate on the DRM list plus the 14 additional semantic associates on the expanded list made up the 15 distractors.2 When the prototype word was not presented at study, the 15 semantic associates on the DRM list made up the 15 target words at test. The nonpresented prototype word and the 14 additional semantic associates on the expanded section of the list made up the 15 distractors. Both in the short and long test conditions, the test words were presented in a new random order for each list and for each subject. The test procedure of the longtest condition differed from that of the short-test condition in that 30 test words rather than 6 test words per list were presented. The other aspects of the methodology were the same as in Experiments 1 and 2A. 4.2. Results and discussion The mean hit rate, mean FA rate, mean d′ and mean c are presented in Table 1. The mean prototype-word hit rate was 0.924 for the short-test condition, and 0.905 for the long-test condition. The difference was not significant, F (1, 253) = 1.35, MSE = 0.017, p = .246, ηp2 = 0.005. The mean prototype-lure FA rate was 0.562 for the short test condition, and 0.476 for the long test condition (quite close to the FA rate of 0.482 under a similar learning and testing condition in Jou et al., 2018). The difference was significant, F (1, 253) = 7.84, MSE = 0.061, p = .006, ηp2 = 0.030. Thus, the FA rate for the prototype words dropped from the short-test to the long-test condition, and so did the hit rate (although not significantly), suggesting that an upward shift of the decision criterion might have occurred. Indeed, there was a significant decrease in liberal bias from the short (c = −0.675) to the long-test condition (c = −0.510), F (1, 253) = 10.78, MSE = 0.161, p = .001, ηp2 = 0.041. However, the criterion upward shift was also accompanied by an increase in d′, that is, d′ for the prototype words increased significantly from the short (d′ = 1.03) to the long test condition (d′ = 1.21), F (1, 253) = 4.34, MSE = 0.489, p = .038, ηp2 = 0.017. Thus, increasing the number of related test words had a beneficial effect on the old/new recognition of the prototype words. How did the number of test words affect the accurate recognition of the semantic associates? The mean hit rate of these words was 0.830 in the short-test condition, and 842 in the long-test condition, respectively. The difference was not significant, F (1, 253) = 0.83. The mean FA rate for the semantic associates in the short-test condition (0.085) was significantly lower than in the longtest condition (0.112), F (1, 253) = 7.48, MSE = 0.006, p = .007, ηp2 = 0.029. Consistent with the change in the FA rate, the decision criterion, mean c, for these words in the long-test condition (0.107) was significantly lower than that in the short-test condition (0.210), F (1, 253) = 9.02, MSE = 0.074, p = .003, ηp2 = 0.034. However, the mean d′s of the two conditions (2.45 in the short-test condition, and 2.36 in the long-test, respectively) did not differ significantly, F (1, 253) = 0.98, although the numerical difference was in the direction consistent with the lower FA rate in the short-test than in the long test condition. Thus, increasing the number of test words increased the FA rate and lowered the decision criterion (to a less conservative point) for the semantic associates, although the manipulation did not change the recognition accuracy (d′) for these words. Lastly, did the test list length affect the discrimination between the studied semantic associates and the prototype lures? Regarding the discrimination between these two types of words, mean d′ for the short-test condition was 0.854, and that for the longtest condition was 1.17. The difference was significant, F (1, 253) = 10.22, MSE = 0.621, p = .002, ηp2 = 0.039. Thus, the discrimination between these two types of words was better in the long-test condition. The criterion measure based on these two types of words indicated that mean c of the short-test condition (−0.588) was significantly more liberal than that of the long-test condition (−0.489). 2 In the short-test condition, all the related distractors were randomly selected from the 14-word added section of the expanded list. In the long-test condition, when the prototype word was presented, the word that was displaced by the prototype word from the original 15-word DRM list along with the 14 additional semantic associates on the extended section of the list was used as the 15 related distractors. Therefore, in the short-test condition, all distractors were from the 14-word extended section of the list, whereas in the long-test condition, in 50% of the time, one of the 15 distractor words was from the 15-word original DRM list. This confound should have had a negligible effect since that displaced word from the 15-word DRM list constituted only one of the 15 distractors. Also, the overall averaged semantic association strengths of the target and distractor words on the test lists across the short- and long-test conditions should be comparable because the test words on the short-test condition were a random sample of the full-length test list in the long-test condition.

10

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

In sum, increasing the number of test words had a beneficial effect on the recognition performance for the prototype words, both in terms of the discrimination between its studied and unstudied versions, and between the unstudied lures and the studied semantic associates, just as changing the unrelated distractor to related distractor did. On the other hand, increasing the number of test words did not seem to affect the old/new discriminability for the semantic associates although it did lower the decision criterion for these words (with a consequent increase in FA rate). Thus, increasing the number of test words affected the prototype words beneficially but seemed to only lower the decision criterion for the semantic associates. 5. General discussion We manipulated semantic relatedness of distractors at test and the number of test words in this study. The most important finding from this study is that using related compared with using unrelated distractors, and using a long test-probe list compared with using a short test-probe list made the recognition of the prototype words more accurate. Moreover, using related distractors also made the recognition of the semantic associates more accurate than using unrelated distractors. In fact, the related test words had produced an effect analogous to a reverse test-priming effect in the present experiments. That is, they enhanced the correct discrimination between the prototype lures and the studied semantic associates, as well as the correct discrimination between the studied and the unstudied prototype words. As well, a longer test-word list improved the discriminability between the studied semantic associates and the prototype lures (see Table 1, under Experiment 2B, d′*). It is remarkable that a five-fold increase in the number of related test words did not produce a detrimental effect on the recognition performance for either the prototype words or the semantic associates. This is in a stark contrast to what was found of the effect of increasing the number of semantic associates in an encoding list (Jou et al., 2016; Robinson & Roediger, 1997), namely, the FA rate of the prototype lure increased consistently and steadily with the increase in the number of semantic associates on the encoding list. These two opposite patterns of results suggest that somehow subjects can distinguish between the two task contexts and be able to selectively inhibit the build-up of the spreading activation from the test words during testing. If the spreading activation which is assumed to produce false memory (Roediger et al., 2001; Underwood, 1965) can be stopped during the test, then it is perhaps not an automatic type, but a strategic type, of activation (Neely, 1977). The spreading activation seemed to have been conditionalized on the task contexts. This idea of spreading activation being inhibitable at test may be able to explain why the supposedly increased context noises from the related distractors and the increased number of probe words at test did not make a detrimental impact on the recognition of these words. Some researchers have observed that prototype lures are typically produced late in the recall output (e.g., Roediger & McDermott, 1995), and hence suggested that they may have derived from the spreading activation from the earlier recalled words (Dewhurst et al., 2011; Marsh et al., 2004). However, Jou (2008) provided evidence that false recall of the prototype lure can be either the product of false memory generated from the encoding phase (typically output early in recall) or the product of a memory construction (metacognitive inference) process taking place late in the retrieval process, producing a late output of the prototype lures. In a recall test, subjects may feel obliged to produce as many words as they can, which may cause them to construct the prototype lure through a metacognitive inferential process after they have exhausted the items available in their memory. However, in a recognition test, subjects should not feel that they have to endorse as many words as they can. Therefore, just because the prototype lure is typically produced late in the recall output does not necessarily indicate that it is the product of priming from the related words produced earlier during the test. Thus, the first conclusion we can draw is that at least within the methodology used in the present experiments, test-induced priming does not seem to play a role in the false recognition of prototype lures. Not only did the distractor relatedness and a larger number of related test words not hurt the recognition performance for the prototype lures, but, as the results showed, the effects of the two manipulations on the recognition of the prototype words came out in the opposite direction of what the interference-based account of forgetting would have predicted (Criss & Shiffrin, 2004; Dennis & Humphreys, 2001; Gillund & Shiffrin, 1984; Hintzman, 1988; Shiffrin & Steyvers, 1997). As we indicated in the introduction section, conceptualizing old/new recognition as categorization learning (Estes & Maddox, 1995; Hirshman, 1995; Nosofsky, 1988) seems to be able to provide an adequate explanation for our results. Now we look at our findings and the processes that may have led to the obtained results from the perspective of the Signal Detection Theory. We first explicate some facts and assumptions from which our conclusions will be deduced. Although an actual criterion shift in a recognition decision process can result in a change in the measured criterion in the Signal Detection Theory paradigm, a change in the measured criterion does not necessarily translate into a real decision criterion shift. A change in the measured criterion can be derived from either a real criterion shift or a shift of the target and/or of the distractor distribution on the strength-of-evidence continuum (Cho & Neely, 2013; Jou et al., 2018; Wickens & Hirshman, 2000; Wixted & Stretch, 2000). On the other hand, a change in the measured d′ score is indicative of a change in the discriminability or separation between the target and the distractor distributions, although it can be accompanied by a measured criterion change when the target distribution’s variance is larger than the distractors’ which is typically the case (Cho & Neely, 2013; Jou, 2011; Kroll, Yonelinas, Dobbins, & Frederick, 2002; Mickes, Wixted, & Wais, 2007). This renders the d′ measure not completely independent of the criterion measure. In our data, however, there were hit-rate and FA-rate indications that the observed improved recognition performance for the prototype words was more than the result of a mere upward shift of criterion. If the positions of the memory distributions of the new and old items remain fixed (no real increase in separation between the new- and old-item distributions), and only the criterion shifts upward, then the hit rate of the old items will significantly decrease after the criterion’s upward shift. In our data, when an increased d′ score was obtained, there was either an insignificant or no decrease in the hit rate of the old items. For example, in Experiment 2A, the hit rate of the prototype words was 0.934 in the unrelated-distractor condition, and 0.924 in the related-distractor condition (see Table 1). In contrast, the FA rate decreased significantly from 0.679 to 0.562. Similarly, for the semantic associates, a virtual absence of any 11

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

decrease in hit rate from the unrelated to the related-distractor condition (from 0.831 to 0.830) was nonetheless accompanied by a significant decrease in FA rate from 0.152 to 0.085. A similar pattern of changes in hit and FA rates was obtained for the prototype words in Experiment 2B. Additionally, in Experiment 1, although the decrease in the semantic-associate hit rate and that in the prototype lure FA rate were both significant across the unrelated and the related-distractor condition (see Table 1, under Experiment 1), the decrease for the former was only 6% (0.872 versus 0.818) compared with a decrease of 25% for the latter (0.765 versus 0.575) (i.e., the decrease in the prototype-lure FA rate was four times the amount of decrease in the hit rate of the studied semantic associates). Thus, there were strong indications in the data that the improved discriminability between the prototype lures and the studied semantic associates was achieved by reducing the FA rate of the prototype lures. These results are more consistent with an account of the distractor-distribution at least making a leftward shift in the related-distractor condition (see Fig. 1B) than one that attributes the increased d′ solely to a criterion shift on two position-fixed distributions. What memory and decision processes could have led to the apparently counterintuitive results we obtained for the prototype words in Experiment 2A? The prototype words have very high memory activation levels generated from studying the semantic associates through the process of implicit associative responses (Underwood, 1965) during the encoding stage. In the unrelateddistractor condition, two-thirds, i.e., the majority, of the distractors were semantically unrelated to the targets. Under that circumstance, as far as memory activation level was concerned, the prototype lure was more similar to the target words than to the distractors. In other words, the prototype lure distribution was closer to the target distribution than to the unrelated distractor distribution, leading to the high rate of FAs to these items (see Fig. 1A). When the two unrelated distractors of the distractor triplet were replaced by related distractors, the similarity of the prototype lure to the average distractors increased, which means that the prototype lure distribution shifted a certain distance to the left, increasing its separation from the target distribution (see Fig. 1B). Thus, both the increased d′ and c for the prototype words could have been derived from a leftward shift of the prototype lure distribution. Moreover, the related distractors also helped the accurate discrimination of the studied semantic associates from the unstudied semantic associates in addition to its discrimination from the prototype lures (see Experiment 2A results in Table 1). Given that there was no decrease in the hit rate of the semantic associates but only a significant decrease in its FA rate, the same distractor-leftwardshift account can be given here to explain the result of an improved discrimination between the studied and unstudied semantic associates. An additional beneficial effect beyond the distractor relatedness effect occurred to the recognition of the prototype words in Experiment 2B as a result of increasing the number of the related test words. The mean FA rate decreased from 0.562 to 0.476 and mean d′ increased from 1.03 to 1.21 from the short- to the long-test condition, both significantly, although the mean hit rate decreased nonsignificantly from 0.924 to 0.905 (about 2% decrease) (see Experiment 2B results in Table 1). Yet, this very small decrease in hit rate was accompanied by a much larger and significant decrease in the FA rate (from 0.562 to 0.476, a 15.3% decrease) and a significant increase in d′ (from 1.03 to 1.21), consistent with an account of a leftward shift of the prototype lure distribution. Note that as shown in Table 1 (Experiment 2B, semantic associate section), the hit rate of the semantic associates actually increased numerically (from 0.830 to 0.842) from the short- to the long-test condition although d′* and c* both increased significantly. Therefore, the significant increase in d′* (from 0.854 to 1.17) cannot be interpreted as solely the result of an increase in c* without an increase in separation between the two memory distributions (because if it had been, the hit rate of the semantic associates would have decreased). Thus, the increased number of related test probes significantly improved the discrimination between the studied semantic associates and the prototype lures. Again, these beneficial effects of the related distractors and increased number of test probes on the recognition memory of the prototype words are difficult to understand from the perspective of item/context noise models. We suggest that the assumed distribution shift we referred to above is compatible with the general notion of differentiation developed within the framework of the SAM model (Criss, 2009; Kili, Criss, Malmberg, & Shifffrin, 2017; Shiffrin, Ratcliff, & Clark, 1990; see Clark & Gronlund, 1996 for a review). According to the concept of differentiation, the memory strength level of the distractor items actually decreases when the memory of the target items is strengthened. In other words, when the target items are rendered more memorable, not only is the memory strength of the target distribution increased, but the strength level of the distractor items actually decreases, rather than remains unchanged, thus increasing the differentiation between the two groups of items. Of course, we did not manipulate the strength of the target items in the present study. Instead, we changed the make-up and the number of probes of the test-item list. Despite the different manipulations we performed in this study than was done in the list strength paradigm, we propose that these manipulations can enhance the differentiation of the prototype lures from the studied semantic associates and from its own studied form by improving on the knowledge about the prototype lure as a distractor. The knowledge about what a distractor is like can be acquired only from the exposure to the distractor items at test. The composition and the sample size of the distractors can influence the quality and quantity of the knowledge about what a distractor is like. When two of the three distractors are unrelated and one distractor is highly related to the targets, subjects’ knowledge about the prototype lure as a distractor they acquire from the distractor distribution is not accurate. Under that circumstance, the lone related distractor is more similar to a target than to a representative distractor. Therefore subjects are more likely to mistake it for a target than to correctly classify it as a distractor. When all the distractors are related, the prototype lure is more similar to a representative distractor than when two thirds of the distractors are unrelated. This idea is based on an analogy between old/new classification and categorization learning. To learn the correct classification, one needs to be given a valid sample of the category (Smith & Medin, 1981). The design of the distractor set in the standard DRM experiments (Roediger & McDermott, 1995) is such (i.e., using majority unrelated distractors) that the false classification rate is maximized. It is similar to a poorly formed eyewitness identification lineup (called an unfair lineup) in which the innocent suspect is the only one with the same features as the real perpetrator. The likelihood of a false identification in such a lineup is greatly increased 12

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

(Wixted & Wells 2017) just as the prototype-lure FA rate is egregiously high in an all-other-distractors-unrelated design. In the remaining of the discussion, we consider whether the present findings can be accounted for by the extant spreading/ activation (Gallo & Roediger, 2002; Roediger et al., 2001), fuzzy-trace (Brainerd & Reyna, 1998; Brainerd et al., 1996; Reyna & Brainerd, 1995), and dual-process theory (Jacoby, 1991; Yonelinas, 2002), respectively. Numerous studies have shown that spreading activation plays a major role in causing DRM false memory (Gallo & Roediger, 2002; Hancock et al., 2003; Jou et al., 2016; Robinson & Roediger, 1997). Our results, one the other hand, indicate that this process plays no role in causing the false memory during testing. Can one argue, however, that the reduction in false recognition in the related or long-test condition is due to an enhanced monitoring process? Yes, one may argue that way. However, a more fundamental question is why monitoring is more effective in the relateddistractor or long-test than in the unrelated-distractor or short-test condition. We argue that whether monitoring is effective or not depends on what the observer monitors against, which is used as the basis for setting up an appropriate decision criterion. And what the observer monitors against depends on what constitutes a distractor in the observer’s mental representation of the distractor category. The mental representation in turn is derived from the observers’ experiences with the exemplars during the testing. For the purpose of correctly rejecting the prototype lure, a monitoring criterion that is derived from a mental representation based on 66% unrelated distractors is a suboptimal one which leads to a high rate of prototype-lure FAs. Hence, the improved monitoring mechanism originates from a change in the mental representation of what constitutes a distractor and from an adjusted decision criterion in probe classification. This is in essence a categorization learning process. Another major false memory theory, the fuzzy-trace theory (Brainerd & Reyna, 1998; Brainerd et al., 1996; Reyna & Brainerd, 1995), posits two dynamic processes, the gist-based and the verbatim-based process, working in an opponent, trade-off relationship to determine recall or recognition performance. A prototype-lure FA is supposed to be the product of a gist-based memory mechanism and a correct rejection of this item the product of a verbatim-based mechanism. The lowering of the FA rate of the prototype lure in the related-distractor condition can be attributed to an enhanced verbatim-based process. But again a more fundamental question is what causes a shift in the relative influence of these two memory mechanisms. The answer is the change in the mental representation of the distractor and the target category, which in turn is the result of a change in the make-up of the exemplars in the distractor category. However, if it is assumed that a related-distractor condition causes one to switch from a gist-dominant to a verbatimdominant processing mode in the DRM paradigm without delving into the underlying cause of the switching, then the fuzzy-trace theory does provide a good explanation for our data. Brainerd and colleagues (Brainerd, Reyna, Wright, & Mojardin, 2003) also proposed the recall-to-reject mechanism as one way by which a highly related lure can be rejected. Recall-to-reject occurs when a studied target very similar to the lure is recollected, providing evidence that the lure is not a studied item. Can this mechanism explain the decrease of false recognition in the relateddistractor condition? Compared with the unrelated-distractor condition, highly related distractors appear on the test list of the related-distractor condition. It is not clear why by being exposed to more related distractors during the test, subjects are more likely to recall targets that are highly similar to the prototype lure. A conceptually simpler mechanism through which more related distractors lower the rate of prototype-lure false recognition is that the related distractors make subjects more aware that there can be many highly related distractors. Finally, we consider whether our results can be explained by the dual-process theory of recognition memory (Jacoby, 1991; Yonelinas, 2002)? One can argue that in the unrelated-distractor condition, the recognition decision is primarily familiarity-based, and in the related-distractor condition, the decision is primarily recollection-based. Again, the crucial question is what causes the change in the decision mode? It is the change in the mental representation of the distractor category, which in turn, results from the change in the make-up of the distractor category. Again, however, if one puts aside the question of the underlying cause for the switching in decision mode from a familiarity-based to a recollection-based one, then this theory also provides an adequate explanation. At any rate, the major extant false-memory models and categorization-learning concept are not mutually incompatible. They both provide an explanation for our data, but at different levels. We propose a dual process hypothesis of memorizing and recognizing prototype words and semantic associates in the semanticassociate learning and memory paradigm (which may be applicable to taxonomic category words as well). One process is driven by spreading activation. This process is brought about by the repeated activation of similar or related concepts and operates during encoding, causing memory interference and impairing memory performance (Roediger et al., 2001; Sadeh, Ozubko, Winocur, & Moscovitch, 2014). The larger the amount of the related or similar information, the greater the interference and the poorer the memory performance is (Jou et al., 2016; Neely & Tse, 2007; Robinson & Roediger, 1997). This process is consistent with the notion of item noise (Gillund & Shiffrin, 1984; Neely et al., 1983; Shiffrin & Steyvers, 1997) and operates at the encoding stage. The other process is categorization learning that is at work during the testing stage to enhance the differentiation between the targets and the lures. Some information presented at test can provide knowledge especially useful for identifying distractors highly resembling the targets such as the prototype lures, and can also help the correct categorization of related distractors in general (e.g., see Table 1, Experiment 2A, semantic- associate section). Again, based on our findings in this study, there is no evidence of spreading activation at the testing stage as a contributing mechanism in the generation of the DRM false memry. We hope that the present research can help better understand the nature of a prototype lure in recognition memory in general, and a crucial feature of the distractor set that generates the dramatic false memory in the DRM paradigm in particular. 6. Author note We sincerely thank Fergus Craik and Henry Otgaar and two anonymous reviewers for their detailed reviews and constructive critiques, all of which added to the quality of this paper. Correspondence concerning this article can be sent to Jerwen Jou, 13

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

Department of Psychological Science, University of Texas – Rio Grande Valley, 1201W. University Drive, Edinburg, TX, 78539-2999, USA. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.concog.2018.06. 003. References Brainerd, C. J., & Reyna, V. F. (1998). Fuzzy-trace theory and children’s false memories. Journal of Experimental Child Psychology, 71, 81–129. Brainerd, C. J., Reyna, V. F., & Brandse, F. (1996). Are children’s false memories more persistent than their true memories? Psychological Science, 6, 359–364. Brainerd, C. J., Reyna, V. F., Wright, R., & Mojardin, A. H. (2003). Recollection rejection: False-memory editing in children and adults. Psychological Review, 110, 762–784. Cho, K. W., & Neely, J. H. (2013). Null category-length and target-lure relatedness effects in episodic recognition: A constraint on item-noise interference models. Quarterly Journal of Experimental Psychology, 66, 1331–1355. Clark, S. E., & Gronlund, S. (1996). Global matching models of recognition memory: How the models match the data. Psychonomic Bulletin & Review, 3, 37–60. Coane, J. H., & McBride, D. M. (2006). The role of test structure in creating false memories. Memories & Cognition, 34, 1026–1036. Criss, A. H. (2009). The distribution of subjective memory strength: List strength and response bias. Cognitive Psychology, 59, 297–319. Criss, A. H., Malmberg, K. J., & Shiffrin, R. M. (2011). Output interference in recognition memory. Journal of Memory and Language, 64, 316–326. Criss, A. H., & Shiffrin, R. M. (2004). Context noise and item noise jointly determine recognition memory: A comment on Dennis and Humphreys (2001). Psychological Review, 111, 800–807. Deese, J. (1959). On the prediction of occurrence of particular verbal intrusions in immediate recall. Journal of Experimental Psychology, 58, 17–22. Dennis, S., & Humphreys, M. S. (2001). A context noise model of episodic word recognition. Psychological Review, 108, 452–478. Dewhurst, S. A., Bould, E., Knott, L. M., & Thorley, C. (2009). The roles of encoding and retrieval processes in associative and categorical memory illusions. Journal of Memory and Language, 60, 154–164. Dewhurst, S. A., Knott, L. M., & Howe, M. L. (2011). Test-induced priming impairs source monitoring accuracy in DRM procedure. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37, 1001–1007. Dodd, M. D., Sheard, E. D., & MacLeod, C. M. (2006). Re-exposure to studied items at test does not influence false recognition. Memory, 14, 115–126. Estes, W. K., & Maddox, W. T. (1995). Interactions of stimulus attributes, base rates, and feedback in recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 1075–1095. Gallo, D. G., & Roediger, H. L. (2002). Variability among word lists in eliciting memory illusions: Evidence for associative activation and monitoring. Journal of Memory and Language, 47, 469–497. Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall. Psychological Review, 91, 1–65. Gunter, R. W., Ivanko, S. L., & Bodner, G. E. (2005). Can test list context manipulations improve recognition accuracy in the DRM paradigm? Memory, 13, 862–873. Hancock, T. W., Hicks, J. L., Marsh, R. L., & Ritschel, L. (2005). Measuring the activation level of critical lures in the Deese-Roediger-McDermott paradigm. American Journal of Psychology, 116, 1–14. Hintzman, D. L. (1988). Judgment of frequency and recognition memory in a multiple-trace memory model. Psychological Review, 95, 528–551. Hirshman, E. (1995). Decision processes in recognition memory: Criterion shift and the list-strength paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 302–313. Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Journal of Memory and Language, 30, 513–541. Jou, J. (2008). Recall latencies, confidence, and output positions of true and false memories: Implications for recall and metamemory theories. Journal of Memory and Language, 58, 1049–1064. Jou, J. (2011). Conscious and unconscious discriminations between true and false memories. Consciousness and Cognition, 20, 828–839. Jou, J., Arredondo, M. L., Li, C., Escamilla, E., & Zuniga, R. (2016). The effects of increasing semantic-associate list length on DRM false recognition memory: Dual false-memory process in retrieval from sub- and supraspan lists. Quarterly Journal of Experimental Psychology, 70, 2076–2093. Jou, J., Escamilla, E. E., Arredondo, M. L., Pena, L., Zuniga, R., Perez, M., Jr., & Garcia, C. (2018). The role of decision criterion in the DRM false recognition memory: false memory falls and rises as a function of restriction on criterion setting. Quarterly Journal of Experimental Psychology, 71, 499–521. Jou, J., Matus, Y. E., Aldridge, J. W., Rogers, D. M., & Zimmerman, R. (2004). How similar is false recognition to veridical recognition objectively and subjectively? Memory & Cognition, 32, 824–840. Kili, A., Criss, A. H., Melmberg, K. J., & Shiffrin, R. M. (2017). Models that allow us to perceive the world more accurately also allow us to remember past events more accurately via differentiation. Cognitive Psychology, 92, 65–86. Kroll, N. E. A., Yonelinas, A. P., Dobbins, I. G., & Frederick, C. M. (2002). Separating sensitivity from response bias: Implications of comparisons of yes-no and forcedchoice tests for models and measures of recognition memory. Journal of Experimental Psychology: General, 131, 241–254. Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A user’s guide. New York: Cambridge University Press. Malmberg, K. J., Criss, A. H., Gangwani, T. H., & Shiffrin, R. M. (2012). Overcoming the negative consequences of interference from recognition memory testing. Psychological Science, 23, 115–119. Marsh, E. J., & Dolan, P. O. (2007). Test-induced priming of false memories. Psychonomic Bulletin & Review, 14, 479–483. Marsh, E. J., McDermott, K. B., & Roediger, H. L. (2004). Does test-induced priming play a role in the creation of false memories? Memory, 12, 44–55. Mickes, L., Wixted, J. T., & Wais, P. (2007). A direct test of the unequal-variance signal detection model of recognition memory. Psychonomic Bulletin & Review, 14, 858–865. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226–254. Neely, J. H., Schmidt, S. R., & Roediger, H. L. (1983). Inhibition from related primes in recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 196–211. Neely, J. H., & Tse, C.-S. (2007). Semantic relatedness effects on true and false memories in episodic recognition: A methodological and empirical review. In J. S. Nairne (Ed.). The foundation of remembering: Essays in honor of Henry L. Roediger III (pp. 313–352). New York: Psychology Press. Nosofsky, R. M. (1988). Exemplar-based accounts of relations between classification, recognition, and typicality. Journal of Experimental Psychology, Learning, Memory, and Cognition, 14, 700–708. Parducci, A. (1984). Perceptual and judgment relativity. In V. Sarris, & A. Parducci (Eds.). Perspectives in psychological experimentation (pp. 135–149). Hillsdale, NJ: Erlbaum. Reyna, V. F., & Brainerd, C. J. (1995). Fuzzy-trace theory: An interim synthesis. Learning and Individual Differences, 7, 1–75. Robinson, K. J., & Roediger, H. L. (1997). Associative processes in false recall and false recognition. Psychological Science, 8, 231–237. Roediger, H. L. (1996). Memory illusions. Journal of Memory and Language, 35, 76–100. Roediger, H. L., Balota, D. A., & Watson, J. M. (2001). Spreading activation and the arousal of false memories. In H. L. Roediger, J. S. Nairne, I. Neath, & A. M. Surprenant (Eds.). The nature of remembering: Essays in honor of Robert G. Crowder (pp. 95–115). Washington, DC: American Psychological Association.

14

Consciousness and Cognition xxx (xxxx) xxx–xxx

J. Jou et al.

Roediger, H. L., & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 803–814. Russell, W. A., & Jenkins, J. J. (1954). The complete Minnesota norms for responses to 100 words from the Kent-Rosanoff Word Association Test. (Tech Rep. No. 11, Contract N8 ONR 66216, Office of Naval Research). University of Minnesota. Saheh, T., Ozubko, J. D., Winocur, G., & Moscovitch, M. (2014). How we forget may depend on how we remember. Trends in Cognitive Sciences, 18, 26–36. Schmidt, R. A., & Bjork, R. (1992). New conceptualization of practice: Common principles in three paradigms suggest new concepts for training. Psychological Science, 3, 207–217. Shiffrin, R. M., Ratcliff, R., & Clark, S. (1990). List-strength effect: II. Theoretical mechanisms. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 179–185. Shiffrin, R. M., & Steyvers, M. (1997). A model for recognition memory: REM—retrieving effective from memory. Psychonomic Bulletin & Review, 4, 145–166. Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard University Press. Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory: Applications to dementia and amnesia. Journal of Experimental Psychology: General, 117, 34–50. Stadler, M. A., Roediger, H. L., & McDermott, K. B. (1999). Norms for word lists that create false memories. Memory & Cognition, 27, 494–500. Underwood, B. J. (1965). False recognition produced by implicit verbal responses. Journal of Experimental Psychology, 70, 122–129. Watkins, O. C., & Watkins, M. J. (1975). Buildup of proactive inhibition as a cue-overload effect. Journal of Experimental Psychology: Human Learning and Memory, 104, 442–452. Whittlesea, B. W. A. (2002). False memory and discrepancy-attribution hypothesis: The prototype-familiarity illusion. Journal of Experimental Psychology: General, 131, 96–115. Wickens, D. D. (1972). Characteristics of word encoding. In A. W. Melton, & E. Martin (Eds.). Coding processes in human memory (pp. 192–215). New York: Wiley. Wickens, D. D., Born, D. G., & Allen, C. K. (1963). Proactive inhibition and item similarity in short-term memory. Journal of Verbal Learning and Verbal Behavior, 2, 440–445. Wickens, D. D., Dalezman, R. E., & Eggemeier, F. T. (1976). Multiple encoding of word attributes in memory. Memory & Cognition, 4, 307–310. Wickens, T. D., & Hirshman, E. (2000). False memories and statistical decision theory: Comment on Miller and Wolford (1999) and Roediger and McDermott (1999). Psychological Review, 107, 377–383. Wixted, J. T., & Stretch, V. (2000). The case against a criterion-shift account of false memory. Psychological Review, 107, 368–376. Wixted, J. T., & Wells, G. L. (2017). The relationship between eyewitness confidence and identification accuracy: A new synthesis. Psychological Science in the Public Interest, 18, 10–65. Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of 30 years of research. Journal of Memory and Language, 46, 441–517.

15