xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001993681
007 cr mnu|||uuuuu
008 090326s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002402
A global memory model of intentional forgetting
h [electronic resource] /
by Melissa Lehman.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 85 pages.
Thesis (M.A.)--University of South Florida, 2008.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: Intentional forgetting is a phenomenon that has been studied by memory researchers since 1968 (Bjork, LaBerge, & Legrand, 1968), however a formal model to explain directed forgetting has not yet been developed. In this paper, I will review the literature on directed forgetting and discuss the results six experiments used assess directed forgetting in highly controlled manner. The striking findings are a.) that directed forgetting phenomena are observed for both free recall and recognition memory when the list method is utilized, b.) that almost the entire effect in free recall is the result of the ability to initially recall the item from the first serial position, and c.) that the costs and benefits are separately affected by an increase in the retention interval. After extensive model analyses, no simple rehearsal or context based model was identified that can handle the full data set. Here I describe a Retrieving Effectively from Memory model (REM; Shiffrin & Steyvers, 1997) that does account for the full range of findings by blurring the traditional distinctions between these classical approached to directed forgetting phenomena.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Kenneth Malmberg, Ph.D.
t USF Electronic Theses and Dissertations.
A Global Memory Model of Intentional Forgetting by Melissa Lehman A thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts Department of Psychology College of Arts and Sciences University of South Florida Major Professor: Kenneth Malmberg, Ph.D. Douglas Nelson, Ph.D. Mark Goldman, Ph.D. Jon Rottenberg, Ph.D. Date of Approval: March 24, 2008 Keywords: directed, remember, forget, context, REM Copyright 2008, Melissa Lehman
i Table of Contents List of Figures ii Abstract iii Chapter One: Introduction 1 Classical Directed Forgetting Hypotheses 4 The Differential Rehearsal Hypothesis 5 The Contextual Differe ntiation (CD) Hypothesis 8 Context Reinstatement 10 Recall and Recognition 13 Chapter Two: Experiments 14 Experiment 1 Â– Free Recall 14 Method 16 Participants 16 Materials 17 Procedure 17 Results 18 Correct Recall 18 Intrusions 20 Serial Position 21 Discussion 24 Experiments 2 & 3 Â– Recognition: Exclusion versus Inclusion 27 Methods 31 Participants, Materials, and Procedure 31 Exclusion Results 32 Hits 32 False Alarms 33 Inclusion Results 34 Discussion 36 Experiments 4 & 5 37 Methods 37 Participants, Materials, and Procedure 37 Inclusion Results 37 Exclusion Results 38 Hits 38 False Alarms 39
ii Discussion 40 Experiment 6 Â– Delayed Free Recall 40 Method 41 Participants, Materials, and Procedure 41 Delay Task 41 Results 41 Correct Recall 41 Serial Position 42 Intrusions 44 Discussion 45 Chapter Three: A Formal Model of Directed Forgetting 48 A REM Model 48 Representation 48 Buffer Operations 50 Retrieval 51 Free Recall 52 Recognition Â– Inclusion 53 Recognition Â– Exclusion 54 Effects of the Forget Instruction 54 Modeling Additional Data 65 Recognition with Shortened Study Time 65 Delayed Free Recall 65 Chapter Four: General Discussion 72 References 80
iii List of Figures Figure 1. Probability of correct recall and intrusion errors for free recall (Experiment 1). 19 Figure 2. Serial position and output orde r data for for free recall (Experiment 1). 22 Figure 3. Conditional response probabiliti es for free recall (Experiment 1). 23 Figure 4. Exclusion recognition performance (Experiment 2). 32 Figure 5. Exclusion serial position data (Experiment 3). 33 Figure 6. Inclusion recognition performance (Experiment 3). 35 Figure 7. Inclusion serial pos ition data (Experiment 3). 36 Figure 8. Inclusion 4s recogniti on performance (Experiment 4). 38 Figure 9. Exclusion 4s recogniti on performance (Experiment 5). 39 Figure 10. Delayed free recall performance (Experiment 6). 42 Figure 11. Serial position in dela yed free recall (Experiment 6). 43 Figure 12. Conditional response probabil ities for delayed free recall (Experiment 6). 44 Figure 13. Model predictions for corre ct recall and intrusions in free recall. 57 Figure 14. Model predictions for seri al position data in free recall. 58
iv Figure 15. Model predictions for first item output position data in free recall. 59 Figure 16. Model predictions for co nditional response probabilities from free recall. 60 Figure 17. Model predictions for inclusion recognition. 61 Figure 18. Model predictions for se rial position data in inclusion recognition. 62 Figure 19. Model predictions for exclusion recognition. 63 Figure 20. Model predictions for se rial position data in exclusion recognition. 64 Figure 21. Model predictions for in clusion recognition with 4 second study time. 66 Figure 22. Model predictions for ex clusion recognition with 4 second study time. 67 Figure 23. Model predictions for delayed free recall. 68 Figure 24. Model predictions for seri al position data in delayed free recall. 69 Figure 25. Model predictions for first item output position data in delayed free recall. 70 Figure 26. Model predictions for co nditional response probabilities from delayed free recall. 71
v List of Tables Table 1. List 1 Intrusions. 20 Table 2. Parameter Values and Descriptions 1. 56
vi A Global Memory Model of Intentional Forgetting Melissa Lehman ABSTRACT Intentional forgetting is a phenomenon that has been studied by memory researchers since 1968 (Bjork, LaBerge, & Legrand, 1968), however a formal model to explain directed forgetting has not yet been de veloped. In this paper, I will review the literature on directed forget ting and discuss the results six experiments used assess directed forgetting in highly controlled manner. The striking findings are a.) that directed forgetting phenomena are observed for both fr ee recall and recogniti on memory when the list method is utilized, b.) that almost the enti re effect in free recal l is the result of the ability to initially recall the item from the first serial posi tion, and c.) that the costs and benefits are separately affected by an increase in the retention interval. After extensive model analyses, no simple rehearsal or cont ext based model was identified that can handle the full data set. He re I describe a Retrieving Effectively from Memory model (REM; Shiffrin & Steyvers, 1997) that does ac count for the full range of findings by blurring the traditional distinctions between these classical approached to directed forgetting phenomena.
1 Chapter 1: Introduction Forgetting is one of the most frustrating aspects of daily experience. Sometimes we forget information that we would like to remember and other times we cannot help but to remember information that we would ra ther forget. The question addressed by my research has been on the latter source of daily frustration. Specifically, I am interested in the cognitive processes involved in intentional forgetting Forgetting is a hallmark of human memory ; it occurs as the result of unconscious, automatic memory processes. However, resear ch suggests that forgetting can also be the result of conscious attempts to control the a ccessibility of information stored in memory. This form of forgetting is studied in the la boratory using directed forgetting procedures, whereby participants are instructed to forg et some material after studying it (Bjork, LaBerge, & Legrand, 1968; MacLeod, 1998 for a review). Memory is then tested for both the to-be-remembered and to-be-forgotten material. In a free recall task, for instance, participants are asked to generate as many items as possible in any order. What is typical ly found is that to-be-remembered words are remembered better than to-be-forgotten wo rds (Bjork, 1970). On the other hand, other measures of memory show no effect of inten tional forgetting; at least this is what is claimed in the literature (cf. Elmes, Adams, & Roediger, 1970). If I take these findings as a given, even if temporarily, they suggest that forgetting is at times under the control of the participant, although not completely so.
2 There are two methods commonly used to investigate intentio nal forgetting. In the item method words are presented one at a time, w ith a cue to remember or to forget each word. In the list method, two lists are usually studied an d participants are told that they will need to remember both lists (the Â“rememberÂ” condition), or they are told after the first list that they will not need to reme mber that list because they will not be tested on it later Â– and that they should try to re member only the upcoming list (the Â“forgetÂ” condition). For example, the experimenter might tell the participant th at the first list was only for practice, in order to orient him to the task, and that he would not be tested on that list. Of course, contrary to the instructions, memory is tested for both lists, and two effects are found: Participants in the forget condition remember fewer words from the tobe-forgotten list and more words from the to-b e-remembered list than participants in the remember condition. These effects are refe rred to as the Â“costsÂ” and Â“benefitsÂ” of directed forgetting, respectively. Though I will discuss theories that attempt to explain the findings from both methods, the focus of this paper is on the list method. Bjork et al. (1968) devised directed forg etting as a method to eliminate proactive interference (PI). In initial experiments, par ticipants were either placed in a remember condition, which had two lists to be remembered ; a forget condition, which had one list to be forgotten and another to be remember ed; and a no-PI condition, which had only one list. Only the last list from the remember a nd forget conditions was tested, and this list was compared to the no-PI condition; meaning only the benefits of directed forgetting were examined Â– and these were observed. Block (1971) used similar procedures to show that directed forgetti ng eliminates proactive interf erence, but not retroactive interference; instruction to fo rget the second list did not improve memory for the first list
3 (again costs were not examined). Using s lightly different procedures, Epstein (1969) found that when participants were told after learning a list that they would have to recall only the second half of a list, they performed better on the se cond half than participants who were told that they would have to re call the second half, th en the first half. Eventually, researchers focused on the status of the to-be-forgotten material and began testing memory for the to-be forgotten info rmation. They found that, in addition to benefits, there are costs associated with th e instruction to forg et (Bjork, 1970). In a typical recognition test, participants are presented with words, some of which were shown on the study list (tar gets), and some of which were not shown (foils); the task is to decide whether these words were previous ly studied. In contra st to the findings for free recall, the consensus in the directed forget ting literature is that the costs and benefits associated with the list method are not re vealed on a recogniti on task (see MacLeod, 1998 for a review). For instance, recognition tests often show no effect of the forget instruction (Elmes, et al., 1970; Bloc k 1971; Geiselman, Bjork, & Fishman, 1983; Basden, Basden, & Gargano, 1993), and other ti mes the results have been inconsistent (Sahakyan and Delaney, 2005). While there has been over 40 years of research and dozens of experiments investigating intentional forgetting under a wide variety of encoding and testing conditions, there is no consensus on how intentio nal forgetting occurs. In fact, there is no coherent explanation of the effects of even small variations in me thodologies. Its not surprising, therefore, that th e current literature is unorga nized and there is no clear direction for research to proceed. The goal of this research is to establish a theoretically relevant set of benchmark empirical findings concerning the effect of instructions to
4 forget and develop a global model embedded in a rich theoretical framework to explain how directed forgetting occurs and to interpret my findings. Classical Directed Forgetting Hypotheses Before I discuss the hypotheses regarding th e way that directed forgetting occurs, I must discuss the possibility that participan ts are not actually forgetting the words, but they are failing to output the words due to demand characteris tics of the task. Researchers have proposed that participants are not recalling the to-be-forgotten words because they believe they should have forgotte n them after the instruction. In order to eliminate this possibility, some researchers have used money as motivation and still found costs and benefits. Woodward and Bjork ( 1971) first had participants recall the tobe-remembered words, awarding them $.01 for each one they reca lled, and penalizing them $.01 for each to-be-forgotten word that they recalled. Participants recalled about half of the to-be-remembered words, and less than 2% of to-be-forgotten words. After multiple lists were learned, participants were asked to recall both to-be-remembered and to-be-forgotten words, and they were awar ded $.01 for both types of words. Woodward and Bjork found performance similar to that in the first task; even when motivated by money, participants were unable to recall th e to-be-forgotten words. MacLeod (1999) replicated these findings; when offered $.50 for each of the to-be-forgotten words recalled, he saw almost no improvement in me mory for these words. Given that demand characteristics cannot explain the effect of forgetting, I will next discuss hypotheses involving the way that participants are able to forget. Many of the earliest hypotheses were developed by R. A. Bjork and his colleagues (1968). Even though th eir initial research considered only the benefits of
5 directed forgetting, all of their hypotheses have also been applied to explain the costs. For instance, the erasure hypothesis states that participants effectively erase items from memory when given the forget instruction. As a result, the to-be-forgotten words are no longer present in memory, and compared to pa rticipants who have not had this material erased, performance is worse, hence the costs of directed forgetting. Additionally, because these words have been erased from memory, they are no longer available to create proactive inte rference, and memory on the following list is better Â– hence the benefits. Bjork et al. (1968) suggested that the erasur e explanation, while plausible, was not likely. Bjork (1970) tested to-be-forgotten items and found that while memory for these items was worse than for to-be-remembered items, some were still remembered Â– unlikely if erasure is possible. Other research suggests that memories do not exist in an all-or-nothing state, which is implied by the erasure hypothesis (Atkinson, Bower, & Crothers, 1965). Moreover, the erasure hypot hesis cannot predict which items would be erased from memory and which items w ould not be. Lastl y, the hypothesis cannot simultaneously predict why erasure would so metimes occur for free recall but never for recognition. I will not consider the erasur e hypothesis any further for present purposes. The Differential Rehearsal Hypothesis A more viable hypothesis is the differential rehearsal hypothesis (Bjork et al., 1968), which says that participants in the fo rget condition stop rehearsing words from the to-be-forgotten list after the forget instruc tion and devote all further rehearsals to the following list, but this does not occur for participants in the remember condition. Because the words on the to-be-remembered lis t receive comparatively more rehearsals
6 after instruction to forget, they are enc oded better. Because words from the to-beforgotten list are encoded less well, they are a ssumed to decay at a greater rate, leading to the costs of directed forgetting. As with the erasure explanation, these words are no longer available, and thus do not create proactive interference, leading to the benefits. The differential rehearsal hypothesis, at le ast this form, is unlikely to provide a complete explanation of directed forgetting fo r several reasons. First, items encoded at varying levels of initial stre ngth are nevertheless forgotte n at similar rates (e.g., Ebbesen & Wixted, 1991; Slamecka & McElree, 1983). Second, directed forgetting is commonly observed even though the conditions of the experiment are specifically designed to prevent rehearsals. For instance, Bjork et al. (1968) and Block (1970) had participants perform various shadowing tasks during study. On the assumption that participants could not simultaneously selectively rehearse first-li st items and attend to the shadowing task, it is unclear why they observed the effect of in structions to forget. Geiselman, Bjork, and Fishman (1983) investigated in tentional versus incidental learning. They assumed that differential rehearsal would not be engaged when memory was incidentally tested and found the costs and benefits of directed forget ting for both intentionally and incidentally learned words. They argued that while sele ctive rehearsal could explain the directed forgetting effect on the intent ionally learned items, it does not explain the effect on the incidentally learned items. It is important to mention that differe ntial rehearsal might help explain the directed forgetting effect for the item met hod. Whereas for the list method, participants are required to rehearse the words from the lis t because they do not find out until after the initial study list that they will not need to remember them, in the item method,
7 participants can decide during learning whethe r they will rehearse a word Â– because the cue is given after each word is presented. MacLeod (1975) found that when using the item method, the directed forgetting effect pers ists over a one-week delay, evidence he used to support the rehearsal explanation. I wi ll address the issue of delay later in this paper. The erasure and rehearsal explanations are similar in that they both posit that the mechanism behind directed forgetting is a failu re to completely encode to-be forgotten information. Alternatively, the set differentiation hypothesis assumes that the forget instruction has its effect af ter the information has been encoded. Bjork et al (1968) proposed that participants res pond to the forget instruction by differentially coding to-beremembered and to-be-forgotten information, in a way that reduces interference between the two. The set differentiati on hypothesis states that part icipants effectively group the to-be-remembered and to-be-forgotten items separately (Bjork, 1970). When given the forget instruction, they differentiate the to-be-forgotten words by putting them in a separate group from the to-be-remembered word s that follow the forget instruction. The set differentiation hypothesis differs from er asure and rehearsal hypotheses because while the set differentiation hypothesis assumes that items are differentiated during encoding, the items are completely encoded, but into different sets, and during retrieval only the tobe-remembered set is searched. It is important to note that none of these hypotheses are n ecessarily mutually exclusive. For instance, Bjork (1970) assume d a combination of the set differentiation and rehearsal explanations; pa rticipants group the to-be-reme mbered and to-be-forgotten items separately, and then devote all rehearsa l to the to-be-rememb ered group. Even so,
8 the additional complexity associated with a combination of thes e hypotheses in their original form cannot account for the differen tial effects of intenti onal forgetting on free recall and recognition. Geiselman, Bjork, and Fishman (1983) suggested that retrieval inhibition was the process responsible for the directed forget ting effect. Accordi ng to the retrieval inhibition hypothesis, partic ipants effectively group the to-be-forgotten and to-beremembered material separately, and then in hibit the to-be-forgotten set. While these items are still present in memory, they are in hibited during retrieval, thus leading to the costs of directed forgetting. Because they are inhibite d, these items do not create proactive interference, leadi ng to the benefits. Elmes, Adams, and Roediger (1970) suggested that the to-be-forgotten informa tion is Â“suppressedÂ” while to-be-remembered information is Â“selectedÂ”. Again, however the retrieval inhib ition hypothesis cannot explain the effects of inten tional forgetting on free reca ll and recognition. Moreover, neither explanation tells us much about how this process works; when participants differentiate the words into two separate sets how are they able to Â“inhibitÂ” one set and Â“activateÂ” the other? The Contextual Differentiation (CD) Hypothesis Sahakyan and Kelley (2002) proposed a different explanation of directed forgetting based on BjorkÂ’s ( 1970) set-differentiation framew ork. According to their hypothesis, study involves the st orage of information representing the studied items (i.e., item information) and the context in which th e items occur (i.e., context information). Each list is associated with an overlapping but not completely similar set of contextual elements. Sahakyan and Kelley hypothesized th at after receiving a forget instruction,
9 participants engage in a mental context ch ange. This causes the overlap between the contextual elements associated with list 1 and list 2 to decrease. As a result, the contexts of list 1 and list 2 are more c ontextually differentiated in memory than they would be without the forget instruction. In addition, the list 1 context is less similar to the context at test after an instruction to forget due to the change in mental context that occurred. When recalling list 2, there is therefore less interference from the list 1 traces. Hence, this is the source of the benefits of the instruction to forget. The costs are the result of the relative inaccessibility of an effective context cue for list 1 traces, again due to the change in mental context that occurred between the list presentations. The context model of directed forgetting ha s its basis in a large literature on the effects of context change on memory performance. Godden and Baddeley (1975) used an environmental context change to alter memory performance on a recall test. Participants learned a list of words either on land or under water, and then were tested in either the same or a different context. Memory perf ormance was impaired when test context was different from study context. Similar impairme nts have been found in studies that used a mood-context change or a state-context change Macht, Spear, and Levis (1977) showed that when participants were tested on word s while in a different mood state than their mood state during study (i.e. a nxious vs. calm), performance was worse than when mood during study matched mood during test. Similarly, researchers have found that participants who studied while under the infl uence of alcohol or marijuana performed better when they were under th e influence during test than when they were not (Goodwin, Powell, Bremmer, Hoine & Stern, 1969; Ei ch, Weingartner, Stillmin & Gillin, 1975, respectively).
10 Based on the context-cha nge literature, Sahakyan and Kelley (2002) hypothesized that participants given the forget instructi on undergo a mental contex t change. In pilot work, they asked participants to retroactiv ely report on their strate gies when given the forget instruction. They report that particip ants often claimed to Â“think about something elseÂ” Â– a method of changing internal context. In order to test the context-change hypothesis, they designed an experiment wher e half of participants participated in standard directed forgetting conditions, and half participated in a context-change condition. In the context change condition, part icipants were given either the remember or forget instruction, followed by an instructi on to change mental context. Participants were instructed to imagine that they were i nvisible, and to think about what they would do if they would suffer no consequences for th eir actions. There were no differences in performance between the remember-plus-cont ext-change (RCC) and forget-plus-contextchange conditions, so I will discu ss data from only the RCC condition. Participants in the RCC condition performed almost identically to participants in the forget condition Â– showing both costs and be nefits of the context change, compared to participants in the remember condition. Sa hakyan and Kelley took this as support for the context change explanation of directed forgetting. Context Reinstatement Smith (1979) showed that while partic ipants who change rooms between study and test show impairment compared to particip ants who are tested in the same room that they studied in, participants tested in a different room who mentally reinstate the environmental context of the study room do not show this impairment. Participants studied a list of words in a one room, and some of participants were switched to a
11 different room for test. Half of participants who were tested in a new room were given the instructions to write down ten things they remember about the room that they studied in, and to remember their thoughts, feelings, a nd the sensations that they experienced in the study room. Participants in this reinst atement condition performed as well on the task as participants who were tested in the original study room. In a second context change experiment, Sahakyan and Kelley examined the effect of context reinstatement on directed forgetti ng. At the beginning of the experiment, they played music from the movie Star Wars in order to create a dist inct context. They again used standard remember and forget conditi ons, along with the remember plus context change (RCC) condition. After studying the sec ond list, half of participants participated in a context reinstatement procedure Â– they we re instructed to imagine what they were doing immediately before the experiment, a nd describe their thoughts and feelings as they entered the room, along with what they remember noticing about the room or the experiment. After receiving th e context reinstatement instru ctions, participants in the forget and RCC groups showed significantly redu ced costs and benefits compared to the groups that did not receive the reinstatement. While the costs and benefits were not completely eliminated, they were certainly re duced Â– and if there we re a perfect way to mentally reinstate list 1 context, perhaps th ey could be completely eliminated. These findings revealed not only that context reinstatement has similar effects in the directed forgetting paradigm to those in the environmen tal context change paradigm, but also that the to-be-forgotten words were still present in memory (further evidence against the erasure and rehearsal explanations, which s uggest that information is not completely encoded in the first place).
12 Much of the data from previous directed forgetting studies can be better explained by the context change explanation. The Geiselman, Bjork, and Fishman (1983) intentionality manipulation st udy was explained by retrieval inhibition. Perhaps a more clear explanation would be that participants Â“inhibitedÂ” the to-be-forgotten information by utilizing an internal contex t change. After they were gi ven the forget instructions, participants switched their me ntal context (possibly by thin king about something else). As a result, the context of list 1 was dissimilar to the test context Â– and the incidentally learned items from list 1 were pa rt of this dissimilar context. Additionally, much of the data regarding intrusion e rrors supports the context change hypothesis. For example, Bjork (1970) looked at intrusion ra tes in a cued recall task and found that intrusions were very rare ly to-be-forgotten items; they were almost always to-be-remembered items. Bjork ma nipulated the number of to-be-remembered and to-be-forgotten pairs (from 1 to 5 and 0 to 3, respectively). As the number of to-beremembered pairs increased, number of intrus ions from these pairs increased; however, an increase in the number of to-be-forgot ten pairs did not increase the number of intrusions from those pairs. The forget inst ruction leads to differentiation between the contexts of the to-be-forgotten and to-be-reme mbered lists Â– so, even though participants are able to remember some of the to-be-forgotten words, these words barely intrude because participants are better able to differe ntiate between the to-be-remembered and tobe-forgotten information. Because the to-beforgotten pairs occur in a different context, they will not intrude on the to-be-remembered pairs, regardless of the number that are presented in this different context. Other to-be-remembered pairs, however, occur in the
13 same context as the tested item, thus the am ount of interference they create will increase as the number of pairs increases. Recall and Recognition Like all prior hypotheses, the contextchange hypothesis can not readily explain why intentional forgetting a ffects free recall but does not affect recognition. The goal here is to provide a simple account of inten tional forgetting that e xplains the effect of directed forgetting for both task s. I began this research by noting that exta nt literature consistently reports poorly design ed experiments. In the foll owing sections, I describe an improved design and report new findings that test several more specific assumptions about how a change in mental context mi ght allow for intentional forgetting. Importantly, the new model predicts that intentional forget ting should affect recognition as well as free recall. Given that I found that extant free recall experiments were poorly designed, I suspected the same would be true of the prior recognition experiments. Indeed, this was the case and the problems were compounded by poor measurement instruments. As a result, I have conducted several recognition memory experiments using the improved design and the proper measures.
14 Chapter 2: Experiments Experiment 1 Â– Free Recall According to models of free recall (M almberg & Shiffrin, 2005; Raaijmakers & Shiffrin, 1980), context plays a major role during retrieval. When aske d to recall freely a list of recently studied items, for instance, the retrieval cue used to probe memory consists of mentally reinstated contextual el ements. The effectiveness of these cues is a positive function of the similarity between the co ntext in the retrieval cue and contents of memory. Most models of contextual dynami cs assume that the contextual elements available to be encoded change over time (Estes, 1955; Mensink & Raaijmakers, 1989; Howard & Kahana, 2002). Thus, these models make the straightforward prediction that more recent events should be better rememb ered than less recent events. Long-term recency effects are commonly found in me mory literature (e.g., Ebbinghaus, 1885). A recency effect is therefore a crit ical prediction of any model; list 2 ( L2) should be remembered better than list 1 ( L1) all thing being equal. A review of directed forgetting literature shows, however, the opposite is almost always the case. Thus, it is possible that the CD model can be rejected based on prior find ings. However, all things might not be equal. The lack of a recency effect in prior experiments might be due to some experimental confounds that aid encoding of L1 while impairing encoding of L2. For instance, the list method us ually utilizes only two lists; L2 experiences proactive interference from L1, but L1 has no such list before it to crea te interference. In addition, L2 is usually followed by a distractor task, which prevents rehearsal of L2 items, whereas
15 L1 is not (Brown, 1958; Glanzer & Cunitz 1966; Peterson & Pete rson, 1958; but see Bjork & Whitten, 1974). This might encourage participants to continue rehearsing words from L1 while they are learning L2, to the detriment of L2 words. To control for these confounds, I utilized a threelist design (cf. Jang & Huber, 2008; Sahakyan, 2004), with a distractor task after each list. The forget instruction comes after L2 and participants are only tested on lists 2 and 3. This ensures th at each list is precede d by another list and followed by a distractor task. A second prediction concerns intrusion erro rs. Because context at test is more similar to the context of L3 than to the context of L2, the number of intrusions from L3 while trying to recall L2 should be greater than the nu mber of intrusions from L2 when trying to recall L3. Another prediction concerning intr usion errors is th at intrusion rates will be lower in the forget condition than in the remember condition. Because the context change that occurs with the forget instructi on makes the lists more distinct, participants should be better able to determine whether a retrieved word came from the wrong list and thus fail to output that word. This predic tion is consistent with BjorkÂ’s (1970) findings that intrusions almost always come from to -be-remembered rather than to-be-forgotten material. Finally, the numb er of intrusions of an L1 item will be greater when trying to recall from L2 than when trying to recall from L3. The context used to probe memory for L2 will be more similar than L3 will be to L1. Additionally, the model predicts fewer L1 intrusions after the forget instruction because th e context at test will be less similar to the context of L1, regardless of whether L2 or L3 is to be recalled (cf. Sahakyan, 2004). However, intrusions errors are usually quite rare. Thus, I might not be able to gather meaningful data to test these predictions us ing a free recall procedure. Intrusion errors
16 are much more common when a recognition procedure is used. I will return to address this methodological issue when I discuss the recognition experiments that are to follow. These predictions can be contrasted w ith those of the differential rehearsal hypothesis. Usually words studied at the begi nning of a list are remembered better than words studied at later serial positions when me mory is tested after some filled delay (i.e., a primacy effect ). According to many models, the primacy effect is observed because early-list items are rehearsed longer than later-list items (Atkinson & Shiffrin, 1968). Additionally, words at the end of a list will be rehearsed in short-term memory until test time. According to the same models, this produces the recency effect. As I mentioned earlier, however, a distractor task eliminates the recency effect, by preventing participants from rehearsing words from the end of th e list (Glanzer & Kun itz, 1966). If the differential rehearsal explanation for directed forgetting is correct (Bjork et al., 1968), participants in the remember condition covertly rehearse L2 items while beginning to study L3, but this should not occur in the forg et condition of the experiment because presumably there is no reason to continue studying the L2 items. If so, I should observe no, or perhaps an attenuated, primacy effect on L3 in the remember condition. Moreover, if participants in the remember condition ar e continuing to rehearse words from the end of L2 while they are learning L3, then I should see a recency effect for L2. Method Participants. Participants were 180 undergradu ate psychology students at the University of South Florida who participated in exchange for course credit. Data for twelve participants were not used because th ey were unable to recall any words from any lists leaving 168 particip ants (42 per condition).
17 Materials. Experiments were all run using Au thorware software, which allows for presentation of visual information and input of user responses. The entire experiment was completed on a computer in an individual participant room. For each participant, 48 words were randomly chosen from the Franci s and Kucera (1982) norms and divided into three lists of 16 words. Procedure. At the beginning of the experime nt, participants were shown an information page that told them about the supposed purpose of the study. They were told that the experimenters wanted to see how well people could not only remember information but also remember where that in formation came from. Participants were informed that they would see three lists of wo rds, and that they would be tested on only one of the lists, but they would not be told wh ich list until later in the experiment, so they needed to remember all of the lists. The instructions were as follows: At the beginning of this experiment, you will study three lists of words. The words will appear on the screen one at a time for a few seconds each. Your task is to remember these words for a later memory test. Importantly, I will only ask you to rememb er the words from one of the lists, which will be chosen randomly, but you will not be told which list until later in the experiment. In between each list there will be short math task. This is involves adding digits in your head and entering the tota l into the computer. Once you have done so, the next list of words will be presented. Once they understood the inst ructions, they continued on to begin the study lists. Participants were given a warning before each list that the study list was about to begin.
18 Lists were shown one word at a time, with each word appearing on the screen in black on a white background for 8 seconds. The lists consisted of 16 words. After each list, participants participated in a math distra ctor task, where they completed two-digit addition problems. The distractor task lasted 30 seconds and participants were instructed to complete as many problems as they could in this amount of time. Participants in the remember condition we re shown each list and distractor task followed by the test. In the forget condition, participants were shown the first two lists and distractor tasks. They were then s hown the forget instruction, followed by study of the third list and a third di stractor task. The forget instruction was as follows: Next you are going to receive the third st udy list. This is the list that you will be asked to recall, so you do not n eed to worry about the first two lists. After all three study lists (a nd distractor tasks), partic ipants were given a free recall test lasting 90 seconds. They were told to enter onto the sc reen all of the words that they could remember from the specified li st. Half of particip ants in each condition were tested on L2 and half were tested on L3. Participants from the forget condition who were tested on L2 (the Â“forgetÂ”) list were told that I wa nt them to recall from this list even though I had previously told them that they wonÂ’t need to remember it. After being tested on the specified list, participants were te sted on the other list (e ither list 2 or list 3), however this data was only used to determ ine whether any participants failed to recall any words from either list Â– in which case their data was thrown away. Results Correct Recall. The statistical analyses are conf ined to the data obtained from L2 and L3. The results of a two-way ANOVA show th at for correct recall, there was a main
19 effect of list, F (1,164) = 68.84, MSE = .021, p < .001; for both remember and forget conditions, probability of recall was significantly greater for L3. This confirms the prediction that more recent lis ts will be better remembered than less recent lists. There was no main effect of instruction, but th ere was a significant List x Instruction interaction, F (1,164) = 19.66, p < .001. As shown in Figure 1 recall was better for the remember condition than the forget condition on L2, but the opposite is true on L3. According to planned comparisons, all re sults shown here are significant to a .05 criterion. Figure 1. Probability of correct recal l and intrusion errors for free recall (Experiment 1). List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Remember Forget Correct recall Intrusions Note: The intrusions in this graph refer to intr usions that came from either list 2 or list 3. When recalling from list 2, any list 3 item that was output is referred to as an intrusion and vice versa.
20 Intrusions. As expected, intrusion rates were very low. These are shown in Figure 1. For intrusions from L2 when recalling L3 and vice versa, there were no significant results; however there were some interesting trends. Participants were more likely to have intrusions from L3 while being tested on L2 than they were to have intrusions from L2 while being tested on L3. Further, the probability of either type of intrusion is lower for participants in the fo rget condition than in the remember condition. For intrusions that came from L1, there is a significant main effect of List, F (1,164) = 17.83, MSE = .002, p < .001; participants were more likely to have intrusions from L1 while they were recalling L2 than when they were recalling L3 for both remember and forget conditions, and again in trusion rates were lower for participants in the forget condition. These are shown in Table 1 Table 1. List 1 intrusions Remember Forget L2 L3 L2 L3 Free Recall 0.0510.0160.0380.01 Delay 0.09380.03270.06820.0556 False-alarm rates for unstudied foils Remember Forget L2 L3 L2 L3 Exclusion 8s 0.1120.0830.1430.073 Exclusion 4s 0.15840.07780.21510.0611 False-alarm rates for inclusion Remember Forget Inclusion 8s 0.07920.0771 Inclusion 4s 0.10030.0683 Note In inclusion, L2 and L3 were tested together, thus there are only false-alarm rates for remember and forget conditions.
21 Thus, the model correctly pred icts that intrusions from L3 while being tested on L2 are greater than intrusions from L2 while being tested on L3, intrusion rates are lower in the forget condition than in the rememb er condition, and intrusions from L1 are more likely when participants are recalling from L2 than when recalling from L3. Serial Position. In addition to looking at correc t recall and intrusion errors, I examined serial position data. This allowed us to explore in deta il the contribution of rehearsal to the directed forge tting effect. The left panel of Figure 2 shows the serial position functions obtained in my experi ment. There is a primacy effect for L3 in both the forget and remember conditions, and it is smaller in the remember condition than in the forget condition. There was a signifi cant main effect of Serial Position, F (15,164) = 5.59, MSE = .125, p < .001. The List x Instruction x Serial Position interaction was not reliable. The smaller primacy effect for L3 in the remember condition suggests that participants might not have given as many rehearsals to words at the beginning of L3 as in the forget condition because they are still rehearsing words from L2. I also examined another aspect of se rial position to further understand the rehearsal component of the dire cted forgetting effect; the fi rst item output during recall. There was a significant main effect of Serial Position, F (15,161) = 10.04, MSE = .057 p < .001, and this was moderated by a significan t List x Instruction x Serial Position interaction, F (15, 161) = 3.29, p < .001 for first item recalled at test. The right panel of Figure 2 shows that the first item recalled from L2 was most likely to be the first word on the list only in the re member condition, whereas on L3, the first item recalled was more likely to be the first word on the list for the forget condition than the remember condition.
22 This indicates that the instruction to forget had a large impact on which word would be recalled first, and L2 and L3 were impacted in opposite directions. Figure 2 Serial position data for free recall (Experiment 1). Serial Position First Item Recalled Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Remember Forget Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Note: For the sake of clarity, the 16 item lis t was compiled into bins. For serial position, each bin spanned two serial positions. For instance, bin n contains the data from serial positions 2n-1 and 2n. For first item output, bin 1 repres ents the first item on the list list 2 list 3
23 (since this is where differences are seen) and all other serial positions are grouped by three. In addition to looking at serial position a nd first item output patterns to evaluate the contribution of rehearsal, I also exam ined the conditional response probability for each condition. Conditional response probabi lity (CRP; Kahana, 1996) refers to the probability of recalling an item from a given se rial position after successful recall of an item from a nearby or distant serial positi on. The distance in se rial position from one item recalled to that of the next item recalled is referred to as lag and lag can be forward or backward. For example, if a pa rticipant recalls an item from the 5th serial position, and next recalls an item from the 6th serial position, this would be represented by a lag of +1. If the participant next re calls an item from the 2nd serial position, this would be represented by a lag of -4. Conditional response probabilities allow us to see the degree to which participants are recalling successi ve items on a list succe ssively Â– presumably this indicates that they are using previous it ems as cues with which to probe memory. As shown in Figure 3, CRP curves differ slightly between conditions, F (1,87) = 2.107, MSE = .092 p < .001. On L2, participants in the forget condition are more likely to move in the backward direction than partic ipants in the remember condition.
24 Figure 3 Conditional response probabilities for free recall (Experiment 1). List 2Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Remember Forget List 3Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Note. Conditional response probabilities re fer to the probability of recalling an item a given distance away from the previous item in serial position. Successive recall of nearby items is more likely than recall of distant items, a nd forward movement is more likely than backward. Discussion I simultaneously observed the cost and benef its associated with directed forgetting and a recency effect. Additionally, while intrusion rates were extremely low, the patterns of intrusions seen in this experiment ar e consistent with a context differentiation hypothesis. Serial position effects, on the other ha nd, suggest that a context differentiation hypothesis is not enough to e xplain the difference between conditions. A difference in the amount of primacy on L3 suggest that participants ar e differentially rehearsing the
25 material in remember and forget conditions. While the primacy effects were consistent with the differential rehearsa l hypothesis, however other as pects of the serial position curves were not. For instance, if the differe ntial rehearsal hypothesis is correct, then the costs and benefits should have only been obs erved for the initial serial positions. However, the cost and benefits were observe d for almost the entire length of serial position curves. In addition, I fail ed to find any recency effect on L2 for either the remember or forget conditions. This suggest s that particip ants in neither the forget condition nor in the remember condition we re rehearsing words from the end of L2 while they are learning L3. These findings suggested, however, that it was possible that while participants were rehearsing words from L2 while studying L3, these words did not come from the end of L2. That is, it is possible that the distractor task was effective in preventing rehearsal during the interval between the list s, but at the beginning of list L3 other L2 items were retrieved from memory and rehearsed. To evaluate this new hypothesis, I sought to determine which items from L2 would be most likely to be retrieved after the distractor task and co-rehearsed at the beginning of L3. It is likely that the word that is getting the most rehearsals during study will be the fi rst word output during recall (Rundus, 1971). The first-item output analysis revealed a different pattern between remember and forget conditions, and suggested that although partic ipants in the remember condition probably were not rehearsing words from the end of L2 while they were learning L3, they may have been rehearsing words from the beginning of L2. Additionally, when recalling L3, they were not very likely to first output the first item from the list Â– suggesting that this item is
26 not getting as many rehearsals as it would if L2 was not being rehearsed during study of L3. The serial position data supports a differe ntial rehearsal expl anation. However, the observed serial position data are also consis tent with the context model if the context cues at test are more strongly associated with the items at the beginning of each list. This would produce an advantage for first items on th e list when the first attempt at recall was made. This advantage should diminish for L3 in the remember condition due to the additional competition from L2 items. In contrast, the first recall advantage for L2 items should be greater following the instruction to remember, since there is less contextual drift and thus the appropriate context cues are more readily available. The CRP curves in this experiment are consistent with thos e typically found in free recall experiments Â– participants are most likely to successively recall from nearby serial positions, and recall is more likely to m ove in a forward than a backward direction. These are referred to as contiguity effects, a nd according to some models they are due to the fact that items from nearby serial positions are associated with a more similar set of contextual elements than those from distan t serial positions (Howard & Kahana, 2002). The CRP findings are also cons istent with my serial position data. Participants in the remember condition are more likely than pa rticipants in the forget condition to recall the item in the first serial position on L2. The CRP analysis showed that participants in the forget condition were more likely to move backward on L2 than participants in the remember condition. Since participants in the remember condition are more likely to recall the item first item on L2, they can only move forward, thus will have a lower number of backward recalls.
27 The CRP functions and the relative lack of recency for L2 in the forget condition is a challenge for the differential rehearsal model to explain. At the same time however, the first-item output findings and the lack of primacy for L3 in the remember condition suggest that differential rehearsal may be play ing a part in the dire cted forgetting effect. Of course, as a REM is a descendent of Atkinson and Shiffrin (1968) modal model, assumptions that produce differential effects of rehearsal an important component of the model. The REM based model I have deve loped assumes that rehearsal affects the storage of item information and context informa tion. I will discuss this model in detail after I present the results from the remaining experiments. Experiments 2 and 3 Â– Recogniti on: Exclusion versus Inclusion It is widely believed that directed fo rgetting does not affect recognition memory (MacLeod, 1998), but I am skeptical for severa l reasons. Both recognition and free recall are episodic memory tasks, and there are very few variables that a ffect one but not the other. In addition, the recognition memory experiments that have been reported often used measures and procedures that are less th en optimal, and most of these studies did not report all of their recogn ition data. For example, Block (1971) reported dÂ’ This tells us that sensitivity was unaffected, but it does not tell us anything about bias, which may have been affected by the forget instruction. Other times small differences in recognition performance were observed, but they were not st atistically significant. Note that context effects on recognition can be very small (c f. Murnane, Phelps, & Malmberg, 1999), and therefore it is possible that the statisti cal tests used in these experiments are underpowered. For instance, it is common to use a sample size of 40+ in each condition of a directed forgetting e xperiment (cf. Sahakyan and Kelley, 2002), but in recognition
28 experiments, samples have been as small as 16 participants in each condition (cf. Basden, Basden, & Gargano, 1993). Many of the methodological issues were probably the result of the fact that recognition memory has never been investigated in a systematic fashion. In fact, many (perhaps most) of the experiments that ha ve been reported simply threw in a few recognition memory trials at th e end of a free recall experiment For instance, Basden et al. (1993) compared to-be-forgot ten words to to-be-remembered words for participants in the forget condition, but did not have a remember condition to which to compare these words. Often, the recognition test followed a free recall test (eg. Elmes et al., 1970). This is problematic because it could add noise to the data, meaning more variation and less ability to see an effect. Perhaps most importantly, nobody has develo ped a cogent model that can predict why free recall and recognition might be differen tially affected by direct ed forgetting. In fact, all models predict th at recognition should be aff ected by directed forgetting, including the model that I have developed. Interestingly, Sahakyan and Kelley (2002) use a failure to find effects of contex t change on recognition memory (Godden and Baddeley, 1980) as support for the context change hypothesis of directed forgetting. There are, however, problems in the literatur e on context change a nd recognition. As in the directed forgetting recogniti on literature, researchers in the field of context change and recognition often fail to report all of their data. Th ey report that there is no difference in sensitivity ( dÂ’ ); however this does not tell us whether there was an effect of the context change. Indeed, context depe ndent recognition has been consistently observed when the appropriate experiments have been conducted (cf. Murnane et al.,
29 1999). In any case, the context change account of directed forgetting does predict an effect in recognition, and data show ing that context change does not have an effect on recognition actually disc onfirms this account. Given the shortcomings of th e prior literature, it is wo rth considering the different ways in which recognition memory can be tested, particularly since the list method requires multiple study lists. Under these conditions, recognition experiments can use either an inclusion test or an exclusion test (Jacoby, 1991; Winograd, 1968). In an inclusion test, the task is simply to say Â“y esÂ” if a word was studi ed on any list during the experiment. Hence, the context that differen tiates the study lists is not required in order to perform the task. Indeed, it is logical to assume that the context used to probe memory would tend to consist of those context feat ures that the study lis ts have in common. Compare this to the free recall procedures us ed in the direct forgetting literature where participants must use a context cue for a particular list, the one specified by the experimenter. The different context cues required to perform free recall a nd inclusion tasks might explain the different effects of directed forgetting that have been reported. In any case, an inclusion test is not the optimal te st for examining context effects on memory. Interestingly, all of the recognition memory experiments in the directed forgetting literature have used an inclusion procedure. Thus, if I accept a CD account for directed forgetting, then it is reasonable to expect small or perhaps ev en null effects of directed forgetting on recognition performance. While so me may consider an inclusion task to be a purely based on a judgment of familiarity of a test item, there are still contextual elements that will affect memory on this task. As context is used in the retrieval process,
30 a more recent list should be better remembered than earlier lists, because context at test will still be more similar to that of the last list (as it was in the recall experiment). Additionally, because the forget instruction di fferentiates context at encoding, the context change that it elicits should not differ betw een recall and recognition tasks. The context change makes the context at test less similar to that of the to-be-forgotten list, increasing the difference between memory for L2 and L3 in the forget condition. Note that because the performance of an inclusion task ma y not rely on contextual elements that differentiate the lists, the effects of the forg et instruction may be small. Nevertheless, they should still be observed. Based on the CD modelÂ’s predictions, I again expect to see costs and benefits, al ong with recency of L3. In an exclusion test, on the other hand, the participantÂ’s task is to say Â“yesÂ” only if a word was studied on a single specified list. In this case, the participant must use context in the retrieval cue that differentiate s the study lists in orde r to accurately perform the exclusion task. Thus, if the context m odel is accurate, then I would expect to see robust effects of directed forgetting on exclus ion task performance. That is, because the exclusion test is a context-base d task, I expect to see effects similar to those in free recall, including the costs and benefits of directed forgetting on hit rates and the recency of L3 in the remember condition. Additionally, fa lse alarm rates should be similar to the intrusion rates for the free recall experiment; there should be more L3 false alarms when a participant is attempting to recognize from L2 than there will be L2 false alarms when a participant is attempting to recognize from L3, and intrusion rates should be lower in the forget condition. In order to facilitate m odeling, the following experiment utilized a
31 design exactly like that used in the prior fr ee recall experiment except that recognition memory is tested using an exclusion procedure. Methods Participants, Materials, and Procedure. In the exclusion experiment, participants were 148 (37 in each condition) undergraduate psychology students at the University of South Florida who participated in exchange for course credit. For each participant, 64 words were randomly chosen from the Franci s and Kucera (1982) norms to create four lists of 16 words. The study procedure was identical to that of Experiment 1. For the test, participants were told which list they would need to recognize words from (either list 2 or list 3). They were told to respond Â“yes Â” if the word shown was on the specified list, and to respond Â“noÂ” if the word shown was from a different list or if it was a new word. The test list consisted of all of the words from lists 2 and 3, and 16 new words, in a random order. In the inclusion experime nt, participants were 60 (30 in each condition) undergraduate psychology students at the Univer sity of South Florida who participated in exchange for course credit. The design, mate rial, and procedure was the same as in the exclusion experiment, except part icipants were told that if they had seen that word on either list (2 or 3) then they should respond by clicking Â“yes Â” and if the word was a new word they were to respond by clicking Â“no.Â”
32 Figure 4 Exclusion recognition performance (Experiment 2). List 23 P(Old) 0.2 0.4 0.6 0.8 1.0 Remember Forget Foils Targets Note. In this experiment, participants were to only endorse items that were studied on a specific list. Thus, some items should have been rejected even though they were studied because they were studied on the to-beexcluded list and not studied on the to-beendorsed list. This graph shows hit rates fo r list 2 and list 3 in the remember and the forget conditions (targets), al ong with false-alarm rates for items that were studied on the to-be-excluded list (foils). For example, if the participant was instructed to positively endorse only items on list 2, any list 3 items th at were positively endor sed counted as list 3 foils. Exclusion Results Hits. For the Exclusion condition, participan ts were asked to recognize either from L2 or L3, so the data was analyzed as a two-way ANOVA with both List and Instruction as between-subject factors. The hit rates were greater for L3 than L2, F (1,144) = 22.06, MSE = .026, p < .001. While there was no significa nt main effect of Instruction,
33 there was a significant List x Instruction interaction, F = 13.54, p < .001. As shown in Figure 4, I found a crossover interact ion, just as in Experiment 1, with hit rates in the remember condition greater than in the forget condition for L2, and the opposite for L3. Planned comparisons confirm the simple effects. Thus in terms of hit rates, I found the costs and benefits of direct ed forgetting, and a recency effect. As displayed in Figure 5 there were no significant e ffects of serial position, F (15,144) = 1.05, MSE = .225, p = .399. Figure 5 Exclusion serial position data (Experiment 3). List 2Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 Remember Forget List 3Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 Note. For the sake of clarity, the 16 ite m list was compiled into 8 bins spanning two serial positions. For instance, bin n contains the data fr om serial positions 2n-1 and 2n. False Alarms. First let us consider the false-alarm rates for those test items that were not studied. These are shown in Table 1 Analysis of the false alarm rates for new words shows no significant main or interacti on effects. This sugge sts that all things being equal, recognition perfor mance (whether an item was st udied or not) is captured
34 solely by the differences that were observed in hit rates. The crossover interaction that was observed in hit rates, ther efore, indicates that there ar e costs and benefit associated with the instruction to forget for recognition memory just as is the case for free recall. The exclusion instructions were to res pond negatively to items that were studied but were not studied on the target list. The ra te at which participants failed to do so can also be considered a false-alarm rate. There was significant main ef fect of instruction on the false-alarm rates, F (1,144) = 5.14, MSE = .049, p = .03. As shown in Figure 4, falsealarm rates are uniformly lower for particip ants in the forget condition than in the remember condition. This is consistent with the context model inso far as an acceleration of the change of context between the two lists as the result of the instruction to forget should make it easier to reject items from the wrong, non-target list. Additionally, while the effect of List is not reliable, participants responded Â“yesÂ” to L3 items when being tested on L2 more often than they responded Â“yesÂ” to L2 items while being tested on L3. This trend is consistent with the assumption that the context used to probe memory is more similar to L3 than to L2. Thus, more L3 items are mistakenly associated with the L2 context when memory is pr obed and the target list is L2 and vice versa when L3 is the target list. Inclusion Results The inclusion version of the recogniti on experiment asked participants to recognize words that had been studied on both lists 2 and 3 Â– thus it was analyzed in a mixed-model ANOVA with List as a within-par ticipants variable and Instruction as a between-subjects variable. There was a significant main effect of List, F (1,58) = 35.43, MSE = .018, p <.001; hit rates were higher for L3 than for L2.
35 Figure 6 Inclusion recognition pe rformance (Experiment 3). List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Remember Forget Just as in recall, there was no main effect of Instruction, but there was a significant List x Inst ruction interaction, F (1,58) = 5.271, p = .025. As shown in Figure 6, hit rates were higher for the remember condition on L2 but there was a minimal difference in recognition on L3. Planned comparisons revealed that performance was significantly higher for L2 in the remember condition, but the difference between conditions was not significant for L3 ( p = .47). There were no differences in probability of responding Â“yesÂ” to new words between reme mber and forget conditions, as shown in Table 1 As shown in Figure 7, there was not a significant eff ect of serial position in the inclusion experiment, F (15,58) = 1.43, MSE = .147, p = .123.
36 Figure 7 Inclusion serial posit ion data (Experiment 3). List 2Bin 12345678 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Remember Forget List 3Bin 12345678 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Note. For the sake of clarity, the 16 ite m list was compiled into 8 bins spanning two serial positions. For instance, bin n contains the data fr om serial positions 2n-1 and 2n. Discussion The results of Experiment 2 are clearly in consistent with the conventional wisdom of the field. I observed clear costs and benefits for recogni tion memory as a result of the instruction to forget. Moreover, Experiment 3 shows that there are costs and benefits associated with the instruction to forget ev en for an inclusion memory task. The costs appear to be larger than the benefits. Howe ver, the inclusion task is easier than the exclusion task, and thus hit rates are relatively high, especially for L3. Thus, I was concerned that the relatively small benefits mi ght be due to some participants performing at ceiling. This concern motivat ed the next two experiments.
37 Experiments 4 and 5 Because of the potential of ceiling eff ect masking the benefits of directed forgetting in Experiment 3, Experiment 4 is a replication of Experime nt 3 with a reduced study time. In order to replicate my findings an d to allow for a further test of the model, Experiment 5 is a replication of Experiment 2 (Exclusion) also with a reduced study time. Methods Participants, Materials, and Procedure In Experiment 4 an inclusion procedure was used. Participants were 86 (43 in each condition) undergraduate psychology students at the University of South Fl orida who participated in exchange for course credit. The procedure was identical to that of Expe riment 3, but with a 4 second study time. In Experiment 5, an exclusion procedure wa s used. Participants were 172 (43 in each condition) undergraduate psychology students at the University of South Florida who participated in exchange for course cred it. The procedure was identical to that of Experiment 2, but with a 4 second study time. Inclusion Results As in Experiment 3, the data was anal yzed in a mixed-model ANOVA with List as a within-participants variable and Instru ction as a between-subjec ts variable. There was a significant main effect of List, F (1,82) = 45.49, MSE = .019, p <.001; hit rates was higher for L3 than for L2. There was no main effect of Instruction, but there was a significant List x Inst ruction interaction, F (1,82) = 7.21, MSE = .019, p =.009. As shown in Figure 8 hit rates were better for partic ipants in the remember condition on L2 and the forget condition on L3. Planned comparisons revealed that all differences shown here were significant. As predicted, I saw recency of L3 and both the costs and benefits of
38 directed forgetting. As in Experiment 3, I sa w no difference in false alarm rates between the remember and forget conditions (see Table 1) Because there were no significant serial position effects in Experiments 2 and 3, serial position effects were not examined for Experiments 4 and 5. Figure 8 Inclusion 4s recognition performance (Experiment 4). List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Remember Forget Exclusion Results Hits. As in Experiment 2, the data was an alyzed as a two-way ANOVA with both List and Instruction as between-subject f actors. The hit rates were greater for L3 than L2, F (1,168) = 15.92, MSE = .028, p < .001. While there was no si gnificant main effect of Instruction, there is a significan t List x Instruct ion interaction, F = 6.71, p < .001. As shown in Figure 9 I found a crossover interaction, w ith hit rates in the remember
39 condition greater than in the forget condition for L2, and the opposite for L3. Planned comparisons confirm the simple effects. In te rms of hit rates, I repl icated the pattern of Experiment 2. Figure 9 Exclusion 4s recognition performance (Experiment 5). List 23 P(Old) 0.0 0.2 0.4 0.6 0.8 1.0 Remember Forget Foils Targets Note. In this experiment, participants were to only endorse items that were studied on a specific list. Thus, some items should have been rejected even though they were studied because they were studied on the to-beexcluded list and not studied on the to-beendorsed list. This graph shows hit rates fo r list 2 and list 3 in the remember and the forget conditions (targets), al ong with false-alarm rates for items that were studied on the to-be-excluded list (foils). False Alarms False alarm patterns for those te st items that were not studied (unstudied foils) are similar to those in E xperiment 2; however the differences between lists are now significant, F (1,168) = 21.12, MSE = .029, p < .001. These are displayed in Table 1 In the exclusion experiments, the rate at which participants responded positively to items that were studied but not on the target list (foils) is also a false-alarm rate. In the
40 exclusion experiment with a shortened study time, some differences emerge compared with the longer study time. There is no longer a main effect of inst ruction, as there was in the long study time version, but there is now a significant List x In struction interaction, F (1,168) = 7.55, MSE = .047, p = .007. As shown in Figure 9, the advantage of lower false-alarm rates for the forget condition is only apparent on L3; there is no difference in false-alarm rates on L2. Discussion In both inclusion and exclusion with re duced study time, both costs and benefits were found. The shortened study time was succes sful in pulling participants away from ceiling and allowing significant be nefits of directed forgetti ng to emerge. Contrary to previous findings in recognition findings, we found an effect of directed forgetting in both free recall and recognition. Findings from the recognition experiments are consistent with our context cha nge model of directed forgetting. Experiment 6 Â– Delayed Free Recall If the costs and benefits involved in di rected forgetting are due to a context change, then increasing the delay between study and test should change the context, such that it is less similar to the context of L2 or L3. Thus, I expect to see a decrease in the recency effect that I observed in Experi ment 1. Moreover, those features that discriminate L2 and L3 should be more difficult to reinst ate after a relatively long delay. If this is the case, then both the costs and the benefits of directed forgetting should be eliminated according to the present model. The results obtained fr om this experiment will help to understand the nature of direct ed forgetting. One question the data will
41 address is whether the effect of directed forgetting is to in stigate a permanent change in the state of a memory trace, which is implied by many different hypotheses. Method Participants, Materials, and Procedure Participants were 176 (44 in each condition) undergraduate psychology students at the University of South Florida who participated in exchange for course credit. The procedure was identical to that of Experiment 1 (free recall) experiment, excep t with an increased lag between study and test. After study of the third list (and completi on of the third distractor task), participants engaged in a 5 minute delay task, designed to prevent rehearsal, after which the test appeared exactly as it did in Experiment 1. Delay task After completion of the third distr actor task, participants were told that they would next see a short video and th en they would be tested on the video. In order to encourage attending to the video a nd prevent rehearsal of words from the lists, participants were told that they must pass the quiz in order to conti nue to the next phase of the experiment (although this was not enfor ced). They then watc hed a four and a half minute informative video about how contact lenses are made, followed by a five question quiz that took approximately 30 seconds. Afte r the quiz, they went on to the free recall test, which took place exactly as in Experiment 1. Results Correct Recall. The results of a two-way ANOVA show that for correct recall, there was a main effect of List, F (1,172) = 10.05, MSE = .024, p < .001, and a main effect of Instruction, F (1,172) = 6.27, p = 013; probability of recall was greater for L3 than for L2, and greater for the forget condition than the remember condition. Both main effects
42 are qualified by a significant List x Instruction interaction, F (1,172) = 6.58, p = .011. As shown in the left-panel of Figure 10, there was no difference in recall for L2, but recall was higher in the forget conditi on than the remember condition on L3. Figure 10 Delayed free recall performance (Experiment 6). List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Remember Forget List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Note. The left panel shows probability of r ecall for correct items. The right panel shows probability of intrusions that came from either list 2 or list 3. When recalling from list 2, any list 3 item that was output is referred to as an intrusion and vice versa. Serial Position. In delayed free recall, significan t interaction effects were found for serial position, F (15,172) = 3.290, MSE = .057, p < .001. As shown in Figure 11 serial position curves did not differ for L2; however the primacy effect is significantly larger for L3 in the forget condition than in the remember condition. While there was a significant advantage for the beginning of th e list in first item output, there was not a significant interaction between fi rst item output and instruction, F (15,172) = 1.491, MSE = .125, p = .10.
43 Figure 11 Serial position in delaye d free recall (Experiment 6). Serial Position First Item Recalled Data Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Remember Forget Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Note. For the sake of clarity, the 16 item lis t was compiled into bins. For serial position, each bin spanned two serial positions. For instance, bin n contains the data from serial positions 2n-1 and 2n. For first item output, bin 1 repres ents the first item on the list (since this is where differences are seen) and all other serial positions are grouped by three. list 2 list 3
44 Unlike in immediate free recall, there were no significant differences in CRP between lists, F (1,90) = 1.073, MSE = .077, p = .301. As shown in Figure 12 on both the L2 and L3, probability of a forward lag of 1 is gr eater for the remember than the forget conditions. Figure 12. Conditional response probabilities for delayed free recall (Experiment 6). List 2Lag -5-4-3-2-1012345 Conditional Response Probability 0.0 0.1 0.2 0.3 0.4 Remember Forget List 3Lag -5-4-3-2-1012345 Conditional Response Probability 0.0 0.1 0.2 0.3 0.4 Note. Conditional response probabilities re fer to the probability of recalling an item a given distance away from the previous item in serial position. Successive recall of nearby items is more likely than recall of distant items, a nd forward movement is more likely than backward. Intrusions. Compared with free recall without the five-minute delay, recall after the five minute delay also leads to a different pattern of intrus ions. As shown in the right panel of Figure 10 intrusions from any list are higher in this experiment compared with
45 the no-delay free recall. A higher level of intrusions suggests that the context that differentiates the lists is not be ing used to distinguish the lists after the delay, leading to confusion as to which list the words came from. For intrusions from L2 when recalling L3 and vice versa, there was a significant main effect of List, F (1,172) = 5.38, MSE = .013, p = .022, a main effect of Instruction, F (1,172) = 5.10, p = .025, and a List x Instruction interaction, F (1,172) = 7.04, p < .001. The pattern of intrusions ex actly replicates the correct recall pattern. There was no difference between remember and forget conditions for L2 intrusions when recalling L3, but there were significantly more intrusions from L3 when recalling L2 in the forget condition than in the remember condition. Further, there was no difference in recall between L2 and L3 for the remember condition. For intrusions that came from L1, there is a significant main effect of List, F (1,172) = 9.78, MSE = .006, p = .002. As expected, participants were more likely to have intrusions from L1 while they were recalling L2 than when they were recalling L3 for both remember and forget conditions. Add itionally, there was a significant List x Instruction interaction, F (1,172) = 4.23, p = .04. The difference in L1 intrusions between L2 and L3 was greater in the remember conditi on than in the forget condition. L1 intrusions are displayed in Table 1 Discussion The delayed free recall experiment suppor ted the context-change hypothesis. First, it eliminated th e recency effect of L3 seen in the immediate free recall experiment. Because of the delay between study and test, the context during test is no longer a very good match to the L3 context, thus its recency advant age disappeared. In addition, the
46 costs were eliminated, as expect ed, but not the benefits of di rected forgetting. Costs are diminished because a large change in context eliminates the advantage of L2 in the remember condition, and hence the text contex t is not very similar to either list. Additionally, when comparing th ese findings to those in from Experiment 1, one can see that recall is better on L2 in the forget condition after th e delay. This suggests that a reduction in L3 recall (also apparent after the delay) allows for less interference from L3 items, thus more sampling of L2. While it might be expected that the dela y would also eliminate the benefits, the results can be explained on the assumptions th at the delay does not negatively affect the ability to reinstate the most recent L3 context as much as it does the more distant L2 context. It does appear in the lower right panel of Figure 11 however, that subjects have a more difficult time initially retrieving the first item on L3 after the delay. After the delay, the first-item recall function is much shallower when compared to the same function from Experiment 1. Thus, subjects we re more likely to initiate retrieval in the later serial positions. This was even more so the case for L2 where subjects often initiated retrieval by outputti ng from the end of the list. According to Sahakyan and DelaneyÂ’s (2003) differential encoding hypothesis, we would expect the elimination of the co sts but not the benefits. However, that hypothesis is not supported when the full set of data are considered. Sahakyan and Delaney hypothesized that while the costs are due to a context change, the benefits are due to a change to a better encoding strategy between lists th at occurs only in the forget condition. If so, we would expect to obser ve benefits for all serial positions on L3, but we
47 only observe strong benefits for items on the first half of L3 in much the same we observed benefits after only short delay in Experiment 1. In addition to correct recall, the intrusion rates seen in the delayed free recall experiment are also consistent with both a context change and a differential encoding hypothesis. First, the intrus ion rates were higher in dela yed free recall compared to immediate free recall, suggesting that the contex t that participants use to differentiate the lists is not as readily available after a dela y; thus when an item from the wrong list is sampled, participants are less able to judge that this cam e from the wrong list based on the different contexts of the two lists. Th e intrusion rates for items that came from L2 or L3 exactly mirrored the correct recall patterns of data Â– with significantly more intrusions from L3 when trying to recall from L2 in the forget condition compared to all other conditions. While in free recall, we expected to see an overa ll lower intrusion rate for the forget condition, this is not the case in the delay experiment. After the five minute delay, the context has changed such that participan ts are no longer as good at using context to differentiate lists and judge whether an item came from the wrong list. The higher output of L3 items in the forget condition suggests th at these items are more accessible due to stronger encoding; since particip ants are not using th e context that differe ntiates the lists, these highly activated items appear as both correctly recalled items and as intrusions.
48 Chapter Three: A Formal Model of Directed Forgetting I will now discuss a formal model of the c ontext change that occurs with directed forgetting. The model must be able to account for findings in both free recall and recognition. It must be able to handle not only the costs and be nefits of directed forgetting and the recency of L3, but also the serial position curves, first-item output, and CRP functions. The first-item output findings are particularly important because most of the directed forgetting effect in free recall a ppears to be driven by differences in the firstitem output patterns. Finally, the model must also be able to account for recognition with shortened study time and fo r delayed free recall. A REM Model I have developed a model in the REM fr amework that accounts for all of the directed forgetting data mentioned above in terms of the context ch ange explanation by extending the REM free recall and context encoding model (Malmberg and Shiffrin, 2005) and the REM recognition models (Malmberg, Holden, & Shiffrin, 2004; Xu & Malmberg, 2007; Shiffrin and Steyvers, 1998). The results are a single set of wellspecified assumptions that account for the e ffect of directed fo rgetting on free recall, inclusion recognition, and exclusion for both lis t method and the item method procedures. Representation According to REM, general knowledge of items is stored in lexical/semantic memory traces and information about past even ts is stored in epis odic memory traces. Lexical/semantic traces are acquired over a li fetime. They contain information about
49 how words are spelled and pronounced and what they mean. In addition, they contain information about the contexts or situations in which they have been encountered. As such, they are accurate, complete, and generalizab le to the contexts in which they usually occur. Two concatenated vectors of f eatures represent these traces. One vector represents the item and the other represen ts the contexts in which it has been encountered. These vectors are generated acco rding to a geometric distribution with the base rate parameter, g : ,..., 1 ) 1 ( } [1j g g j V Pj. (1) When a word is studied, the wi item features of its lexical semantic trace are copied to form a new episodic trace that re presents this occurrence. In addition, wc features of the current context are stored. Ep isodic encoding is an incomplete and errorprone process. During the storage process, a feature may be copied correctly, it may be copied incorrectly, or it may fail to get copied at all. The pr obability of storing a feature given a certain unit of time ( t ) is represented by the xu parameter. Given that a feature is stored, it is stored corre ctly with a probability c An item will be stored incorrectly with a probability 1-c in which case a feature will be randomly chosen according to the geometric distribution. The absence of a stored feature is represented by the value zero. When items are studied, context is stored in episodic traces in the same way. For the sake of simplicity, I will assume that co ntext features change between lists with a probability of but not within lists. Thus, all it ems within a list will share the same context information. When a context featur e value is changed it is randomly sampled from the geometric distribution. I further assu me that context featur es change after the
50 final study list. The context changes in the sa me manner between the final list and test. The features representing the item itself, however, will be different for each item. Buffer Operations As a descendent of the modal model (A tkinson and Shiffrin, 1968), the interaction of control processes and structural aspects of memory are used to model serial position data. Control processes operate on items lo cated in a limited capacity rehearsal buffer during encoding. For present purposes, I c hose a buffer capacity of two items, although the larger capacities would also work. Upon the presentation of the first item on a list, its lexical/semantic item features enter th e buffer and two things happen. First, to attempts to copy the items features in an episodic trace are made. The probability of storing an item feature is iu. Second, t1 attempts to copy the current cont ext features in an episodic trace are made. The probability of storing a context feature is cu. I assume that attention is focused on the item itself rather than context, so item information will be better encoded than context information, represented by a greater u* value for item information (* iu>* cu). Upon the presentation of the second ite m on the list, its lexical/semantic item features take the remaining sl ot in the buffer, and now three things happen. The item and context features are stored in the same way as before. However, in addition, some of the items features of the first item are stored in another concatenated vector (cf. Kimball, Smith & Kahana, 2007). This represents the assumption that an episodic association between the two items in the buffer is stored (this loosely corresponds to strengthening an inter-item association in SAM). The probability of storing the associative items features is au. I assume that attention is primarily fo cused on the present item; information about
51 the older item in the buffer will be encode d worse than the current-item information, represented by a greater u* value for item information (* iu>* au). That is, I assume that participants will tend to focus their attenti on on the most novel items in the rehearsal buffer. As new items are added to the buffer, this process repeats, with the oldest item being dropped with a probability Retrieval The first step of the retrieval process is similar across all te st conditions (recall, recognition-inclusion, recognition-exclusion). An activated subset is created, which consists of only the items with the strongest association to the curr ent context. From a participantÂ’s perspective, only items encounter ed recently, that is during the experiment, are relevant to the task. In order to access only these releva nt items, the retrieval task uses only items in this subset. In order to create the active subset, the current context cue is matched against the episodic images stored in memory. The matching process involves calculating a likelihood ratio for each trace, which takes into account bot h features that match and features that do not match. Matching f eatures increase and mismatching features decrease the likelihood ration; cases where no f eatures are stored do not contribute to the likelihood ratio either way. Likelihood ratios are calculated accord ing to the following equation: ijm ijn i i i n jg g g g c c c 1 1 1) 1 ( ) 1 ( ) 1 ( ) 1 (, (2)
52 where g is the environmental base rate for the occurrence of features, i is a feature value ranging from 1 to infinity, nij is the number of mismatching context features for an item (regardless of th eir value), and nijm is the number of times feature i matched the retrieval cue with value j. The Â“activatedÂ” subset of memory will consist of a certain percentage of all traces in memory, represented by the parameter. The items that get into the subset are those with th e highest likelihood ratios. Free Recall The free recall task begins with the cr eation of the cue with which to probe memory. The initial cue consis ts of only context features. I further assume that the context cue is a combination of the current te st context and reinstat ed list context, in order to allow one to access th e intended list. The proportion of reinstated list context features is represented by the parameter. Free recall operates in REM as a memory search process, with cycles of sampling a nd recovery (Malmberg & Shiffrin, 2005). For simplicity, I assume that if an item is sample d, it is recovered with a probability of 1.0. The cue is matched against all traces in the su bset in an attempt to sample an item from the given list. Likelihood ratios for all im ages are calculated according to Equation 2. The probability of sampling image, Ii, given the context retrieval cue, Q, is as follows k i iQ I P ) | ( (3) If an item is sampled and it comes from the correct list, it is output with a probability of 1.0. If, however, an item is sa mpled and it comes from an incorrect list, I assume that the participant undergoes a monitoring process. The probability of outputting an incorrect item is a function of the overlap in context between lists
53 (represented by the parameter). If the context change that occurs between lists is large, then there is little overlap between lists and participants will be better able to judge whether a sampled item came from the correct li st or not. If the context change between lists is small, there will be much overlap betw een the lists, and it will be much harder for participants to judge wh ether a sampled item came from the intended list. If an item is output, the next cue used to probe memory will consist of both context and item information. Again, the c ontext portion of the cue consists of both current context features and c ontext features associated with the given list. The item portion of the cue consists of the item vector fr om the last item recalled. Thus it is most likely that co-rehearsed items, which share the current itemÂ’s information, will be sampled next. If no item is output, then the orig inal context cue is used for the next probe of memory. The sample a nd recovery process repeats times. Recognition Â– Inclusion In the inclusion task, a participantÂ’s ta sk is to positively endorse any studied items. For this reason, a simple global matc hing process is used. In REM, a decision about whether an item is judged as Â“oldÂ” is made based on the likeli hood ratios calculated for all items in the comparison set. In this ca se, the cue (a test item) is compared to all items in the activated subset. The Â“odds Â” are calculated according to the following equation: n j jn11, (4) and if the odds exceed a specified criterion, the item is judged as old, otherwise it is judged as new. In the absence of any instru ctions that would lead to a bias to respond
54 differently, an old decision is made if the odds exceed a criterion of 1.0, thus this is the decision criterion that I use here. Recognition Â– Exclusion In the exclusion task, a participantÂ’s task is to positively endorse only items that came from a given study list. Foils consist of both new ite ms and items from the other list, so participants must be able to distingu ish between studied items from the correct list and studied items from the othe r list. A global matching proce ss, as that in the inclusion task, is first used followed by a monitoring task, as in the free recall task. After an item is identified as a studied item based on the global match (according to Equation 4), a participant again makes an output decision that is dependent on the overlap in context between the two lists. Large ove rlap in context means that it is harder to distinguish between the two lists and the fals e alarm rate will be increased. Effects of the Forget Instruction The forget instruction will have multiple effects in these tasks. First, it will increase the rate of context ch ange between lists, so that the two lists share less context features (less overlap). Second, it will alter the encoding of the first item on the third list in the forget condition. The first item on a ny list is encoded more strongly than other items because the rehearsal buffer is not full dur ing encoding of this item, and it has more opportunities to be linked to the context of th e list. I assume that the opportunity for linking this item to context is greater for the third list in the forget condition, because this list is encoded after the forget instruction, when no items fr om the previous list are being rehearsed. Finally, the forget instruction will decrease the probability of reinstating features for use in the context cue used in the recall process. Because the forget
55 instruction increases the context change that occurs between lists, it should be harder to reinstate context features afte r this instruction. As menti oned previously, the probability of an intrusion error will be dependent on the context change that occurs between lists, thus the probability of outputting an item from the wrong list will be lower in the forget condition. I attempted to find a reasonable set of pa rameters to account for my more than 300 data points. I did not attempt to find a Â“best-fittingÂ” set of parameters, and I am focused more on accounting for the overall patterns of the observed data, and less concerned with formal model comparisons. Th at said, I did vary a number of parameters in order to determine if the quantitative pr edictions were in the Â“right ballparkÂ”. Descriptions of each parameter are listed in Table 2. Many parameter values are common to all experimental conditions. In addition, there are 11 free parameters, but the majority of these are scaling parameters th at do not differ between conditions; only four parameters differ between remember and forg et conditions and these are carry most of water for the model. The parameters that wi ll differ from those listed in the table after the forget instruction are as follows: t1 = 12; = .8; 2 = .15; = .5. Data and model predictions for Experiments 1 through 3 are presented in Figures 13-20. Overall there is a st rong correspondence between the model and data. The model captures the costs and benefits of directed forgetting for free recall, exclusion recognition, and inclusion recognition. It also accounts fo r the serial position, first-item output, and CRP data for free recall. While a more complicated model could probably do a better job than the current model, there does not seem to be much to be purchased by the additional complexity.
56 Table 2. Parameter Values and Descriptions Parameter Value Description g .4 Environmental base ra te (standard value) wi 8 Number of item features wc 8 Number of context features c .8 Probability of correct ly storing a feature u*i .5 Probability of storing an item feature u*c .2 Probability of stor ing a context feature u*cr .1 Probability of copying a co-rehearsed item's feature t1 6* Number of storage attemp ts for first item on a list t0 2 Number of storage attempts for all other items on a list 20 Number of sampling attempts .2* Probability of change for c ontext features between lists .75 Probability of dropping the oldest item in the buffer 2 .2* Probability of reinstating context features on list 2 3 .8 Probability of reinstating context features for list 3 .8 Size of activated subset of items .6* Probability of outputting an intrusion Note. Parameter values with asterisks are thos e that differ in the forget condition. For the forget condition, t1 = 12; = .8; 2 = .15; = .5.
57 Figure 13. Model predictions for correct re call and intrusions in free recall. Data Model List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Remember Forget Correct recall Intrusions List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 Correct recall Intrusions
58 Figure 14. Model predictions for serial position data in free recall. Data Model Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Remember Forget Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Note. For the sake of clarity, the 16 ite m list was compiled into 8 bins spanning two serial positions. For instance, bin n contains the data fr om serial positions 2n-1 and 2n. list 2 list 3
59 Figure 15. Model predictions for first item output position data in free recall. Data Model Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 123456 P(First Output) 0.0 0.1 0.2 0.3 0.4 0.5 Note. For the sake of clarity, the 16 item list was compiled into bins. For first item output, bin 1 represents the first item on the list (s ince this is where differences are seen) and all other serial positions are grouped by three. list 2 list 3
60 Figure 16. Model predictions for conditional response probabilities from free recall. Data Model Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Remember Forget Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Lag -5-4-3-2-1012345 Conditional Resp onse Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 list 2 List 3
61 Figure 17. Model predictions fo r inclusion recognition. Data Model List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Remember Forget List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0
62 Figure 18. Model predictions for serial pos ition data in inclusion recognition. Data Model Bin 12345678 Hit Rate 0.6 0.8 1.0 Remember Forget Bin 12345678 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Bin 12345678 Hit Rate 0.6 0.8 1.0 Bin 12345678 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Note. For the sake of clarity, the 16 ite m list was compiled into 8 bins spanning two serial positions. For instance, bin n contains the data fr om serial positions 2n-1 and 2n. list 2 list 3
63 Figure 19. Model predictions fo r exclusion recognition. Data Model List 23 P(Old) 0.2 0.4 0.6 0.8 1.0 Remember Forget Foils Targets List 23 P(Old) 0.2 0.4 0.6 0.8 1.0 Foils Targets
64 Figure 20. Model predictions for serial pos ition data in exclusion recognition. Data Model Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 Remember Forget Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 Bin 12345678 Hit Rate 0.4 0.6 0.8 1.0 list 2 list 3
65 Modeling Additional Data While not explicitly required of me, I also explored the ability of the model to account for the data from three additional e xperiments. Overall the model did well, especially when you note that the predictions derived for Experiment 6 simply fell out of the model derived for the earlie r experiments. In that se nse, Experiment 6 provided an a priori test of the model. Recognition with shortened study time The model parameters are identical to those in the previous recognition experiments aside from a reduced t value. Because study time is reduced by half, the t values for these two conditions were reduced by half. No other changes were made to the model. The new parameter values are: t1 = 3; t0 = 1; t1(forget) = 6. Data and model predictions for Experiments 3 and 4 are presented in Figures 21-25. Delayed Free Recall The model parameters are identical to t hose in the previous free recall experiment, except for an increased context ch ange that occurs af ter the last list. In addition, because time has passed, it may be harder to recover the c ontents of a trace even after that trace is sampled. For this reason, recovery probabi lities are also reduced by half. The new parameter value is: = .8. Data and model predictions for Experiment 6 are presented in Figure 26.
66 Figure 21. Model predictions for inclusion recognition with 4 second study time. Data Model List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0 Remember Forget List 23 Hit Rate 0.5 0.6 0.7 0.8 0.9 1.0
67 Figure 22. Model predictions for exclusion recognition with 4 second study time. Data Model List 23 P(Old) 0.0 0.2 0.4 0.6 0.8 1.0 Remember Forget Foils Targets List 23 P(Old) 0.0 0.2 0.4 0.6 0.8 1.0 Foils Targets
68 Figure 23. Model predictions fo r delayed free recall. Data Model List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Remember Forget List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Foil List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Intrusions List 23 P(Recall) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Intrusions
69 Figure 24. Model predictions for serial pos ition data in delayed free recall. Data Model Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Remember Forget Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Bin 12345678 P(Recall) 0.0 0.1 0.2 0.3 0.4 0.5 Note. For the sake of clarity, the 16 ite m list was compiled into 8 bins spanning two serial positions. For instance, bin n contains the data fr om serial positions 2n-1 and 2n. list 2 list 3
70 Figure 25. Model predictions for first item out put position data in delayed free recall. Data Model Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Remember Forget Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Bin 12345678 P(First Output) 0.00 0.05 0.10 0.15 0.20 0.25 Note. For the sake of clarity, the 16 item list was compiled into bins. For first item output, bin 1 represents the first item on the list (s ince this is where differences are seen) and all other serial positions are grouped by three. list 2 list 3
71 Figure 26. Model predictions for conditional re sponse probabilities from delayed free recall. Data Model Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Remember Forget Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Lag -5-4-3-2-1012345 Conditional Response Probability 0.00 0.05 0.10 0.15 0.20 0.25 0.30 list 2 List 3
72 Chapter Four: General Discussion The primary achievements of this project are both empirical and theoretical. By carefully considering the designs of prior experiments in th e literature, I was able to develop hypotheses about the sources of ma ny of the inconsistency observed there. When a set of carefully designed experiment s were used to systematically explore directed forgetting, I observed reliable effect s of directed forgetti ng for both recognition and recall. In addition, I was ab le to observe recency effects in free recall that to date had eluded prior investigations. With a coherent set of observati ons in place, I was then able to explore a wide variety of models with the REM framework to account for them. Here I will discuss the successes of the model already discussed and the failures of several models that lead to the current model. The free recall model accurately predicte d the patterns of da ta for both correct recall and intrusions. While the intrusion ra tes were higher overall in the model, the pattern of intrusions matched that of the data. The differences between remember and forget conditions in serial position, first it em output, and CRP were sometimes smaller in the model compared to those in the data, however the overall patterns were consistent. Aside from a few data points, model predic tions were also consistent with the data for recognition inclusion and exclusion. For hit rates in inclusion, the model predicted both costs and benefits; the data showed costs, and the appearance of benefits that were not significant (perhaps due to a ceiling effect). Serial pos ition predictions matched data for all lists except for L2 in the forget condition, in whic h the model predicted a lower hit
73 rate for the first item on the list compared to the rest of the items on the list; this pattern was not apparent in the data. Both hit rates and false alarm rates seen in exclusion were predicted quite accurately by the model; the model matc hed the data both qualitatively and quantitatively. Results for model predictions in serial position were similar to those in inclusion Â– the model predicted the data well in all conditions except for L2 in the forget condition, in which the model agai n predicted a lower hit rate fo r the first item on the list. For inclusion and exclusion with re duced study time, the model was again accurate in predicting the data; the only diffe rence predicted in the model that was not apparent in the data was a difference in the number of intrusions coming from L2 during a test of L3 between remember and forget conditions. The model predicted (as in free recall and exclusion with longer study time) an ove rall lower intrusion rate for the forget condition, but this was not apparent in the data. Finally, the model predictions were also consistent with the data in delayed free recall. Data patterns were accu rately predicted by the mode l for correct recall, serial position, first-item output, and CRP. Again, th e degree of the effect may have been smaller, but was quantitatively predicted by th e model. A slight difference was seen for intrusions; however the pattern of intrusion rates does not diff er between the data in the model Â– intrusion rates for the remember conditi on are just lower overall in the data than in the model. The model was successful in accounting for a significant amount of data given very few free parameters. Between the Re member and Forget conditions, only four parameters were changed, in order to account for 310 data points. In addition, by altering
74 a few additional parameters, I was also ab le account for data from my recognition experiments with reduced study time, a nd my delayed free recall experiment. There are a few critical aspects of the model contributing to the difference between remember and forget conditions. Fi rst, an increased c ontext change between lists in the forget conditions creates le ss overlap in contextual features between L2 and L3. This contributes to both the costs and benef its of directed forgetting; the costs occur because the context at test shar es less common features with L2 and the benefits occur due to less competition from L2 items when retrieving from L3. The change in context that occurs with the forget instruction also leads to a difficulty in reinstating L2 context features at the time of recall, also contribu ting to the costs. Fi nally, the decrease in overlap between the contexts of the two lists makes intrusion rates lower for the forget condition. Because contexts share less common features, the lists are more distinct, making participants better able to determine that a word came from an incorrect list. In addition to creating a great er change in context, the forget instruction also has an effect on the rehearsal component of the m odel. After the forget instruction, the first item on L3 is better encoded compared to the re member condition. Th is contributes not only to the benefits in the free recall experiment but also to th e long lasting benefits in the delayed free recall experiment. Even after a delay in which context is again changed, the benefits persist, suggesting that L3 items are better encoded. Th e serial position data from delayed free recall indicate that the initial items on L3 in the forget condition are significantly more likely to be re called than the initial items on L3 in the remember condition, supporting the hypothesis th at the first item is better encoded than other items.
75 My model of directed forgetting utilizes both context and rehearsal components. The free recall serial position and first-item out put data discussed in this paper suggest that a combination of contex t change and rehearsal is c ontributing to the directed forgetting effect, and the current model provid es the best predictions for these data. Further empirical work is necessary in order to properly evaluate these hypotheses. For example, experiments designed to eliminate rehearsal (either by us ing simultaneous tasks during encoding or incidentally encoded lists) within the current 3 list + distractor task design may help shed some light on the issue of rehearsal. Future work on the model may also be needed in order to determine whether both context and rehearsal components are necessar y. It seems unlikely that a pure rehearsal model could account for the data, given that a rehearsal model with no context change would not predict a recency effect, as ther e is no mechanism in a rehearsal model to produce recency. It may be that a combina tion context + rehearsal model that differs from the current one could account for the data; context may change over time, but without an increased context change after the forget instruction, and only a change in rehearsal contributing to the directed forgetting effect. A context-only model may be a more reas onable model, howev er various contextonly models were attempted, all of which were unable to predict the current patterns of the data. All of these models used the same basic process for directed forgetting as the current model Â– an increased context change between lists given the forget instruction. The first model developed was a context-only model which had context features that changed at different rates some experime ntal context features stayed constant throughout all three lists, some changed more quickly so that there was less overlap in
76 these features between lists, and some changed very rapidly so that they were different even for items in the same list. Thus, there were some features that were shared between lists and others that made the lists more distin ct. The forget instruction increased the rate of change for all context feat ures. Other variations in which the forget instruction only increased the rate of context change for certain features were also attempted, with similar results. For retrieval, the rapidly changing featur es were not used in the sampling process (in an attempt increase sampling from L2 or from the beginning of L3). In a slightly different version of this model, these feat ures were changed and random feature values were used instead. In either version, recen cy was present but I was unable to get higher sampling of L2 in any case (meaning even when sampling was intended for L2, L3 items were still more likely to be sampled); at most, I could produce equal recall for L2 and L3. In addition, I was unable to get an advant age for items at the beginning of the list (primacy), a vital aspect of the model. The second model used a vector that was divided into individual list components Â– one part of the vector was dedicated to L1, one part was dedicated to L2, and one part was dedicated to L3. When encoding items from a give n list, the sections of the vector that were dedicated to other lists would be left blank (with a few features being encoded erroneously). Two versions of this model were attempted Â– one where context changed between and within lists, and one where co ntext changed between lists but not within. This model produced the costs, due to the co ntext change making the test context less similar to that of L2, but there was no mechanism to cr eate the benefits, and thus they
77 were not observed. This model also fa iled to produce primacy, as there was no mechanism to give the advantage to ite ms in a specific position on the list. A second version of this model was a comb ination context + rehearsal model. In this version, co-rehearsal of items in the bu ffer led to storage of item information from co-rehearsed items, and this information was used as part of the retrieval cue. Additionally, the first item on the list was give n more rehearsals thus leading to better encoding. This eliminated the problem of no primacy (and also increased the probability of recalling successively studied items, which produced the CRP curves), but still did not produce the benefits of directed forgetting. The next version of the model returned to the use of a single c ontext vector, rather than one that was divided into list portions. During recall, the contex t of a given list was reinstated to use as a cue to recall from that list. This allowed for recall from a specific list but did not produce a serial-position curve. Another version of this model included special context features that were associated with only the first item on the list to be stored in lieu of co-rehearsed item features (since this item is alone in the buffer at the beginning of a study list). During recall, a sp ecial cue containing only these features was reinstated. This produced primacy, but ag ain I was unable to get higher sampling of L2 when it was the intended recall list. Another manipulation of the model used the same Â“specialÂ” cue at the beginning of recall but then used a rec overed itemÂ’s stored context as the context cue for sampling of the next item, a similar process to that used in Howard and KahanaÂ’s (2002) Temporal Context Model (TCM). This model produced a primacy effect, howev er it also created recall that was too good Â– once an item from a given list was sampled, it was too easy to
78 stay in that list. The final version of this model solved al l of these probl ems by returning to the reinstated list cue (rather than a sp ecial cue for the first item on a list). Context information for the first item on a list was simp ly encoded better than that of other items on the list, and co-rehearsed item information wa s used as part of the cue after an item was recovered. This solved the previous problems but still did not allow us to get the first-item output patterns seen in the data. From this model, the current manipulations were made to produce a working model. While none of the earlier models were ab le to fully account for all of the data, a combination of one of these models combined with the current assumptions may be necessary. A variation of the current model in which context changes within a list (in addition to between lists) may be better ab le to account for the data. While this manipulation was present in ea rlier models that were not successful in predicting the directed forgetting data, addi ng this assumption into the current model may provide a better account of the current data and may also better predict future data. If I implement the assumption that contex t changes within a list, it will also be important to consider the way th at context changes within a list. Traditionally, models of context change within a li st (Mensink & Raaijmakers, 1989) assume that context fluctuates randomly throughout the list. In my model, this would be represented by the same type of change that occurs between li sts occurring within lists at a slower rate. Howard and KahanaÂ’s (2002) Temporal Contex t Model (TCM) assumes that rather than context changing randomly, context drifts base d on contextual states elicited by studied items.
79 My model was quite successful in accoun ting for a variety of data given very few free parameters. It will be useful to examine other ways in which this model can be manipulated to better fit the data (for exam ple by adding in a within -list context change similar to that in TCM). In addition to explaining direct ed forgetting data, a contextchange model such as this could be used to explain other context-dependent phenomena. Given the success of the current model, it will also be possible to generate predictions concerning the effect s of directed forgetting on other specific memory tasks, which of course can be empiri cally tested. I am particular ly interested in extending the model to explain the effect of directed forgetting on memory performance measured using the item-method. As previously discu ssed, there is no reason to believe that the item-method would necessitate a context change component to the model, and I predict a directed forgetting effect using only a re hearsal manipulation in the item-method. A rehearsal model for the item method would eliminate all assump tions about contextchange and instead utilize a manipulation of number of rehearsals each item receives. This effect would be achieved by changing the value of the t parameter between to-beremembered and to-be-forgotten words.
80 References Atkinson, R.C., G.H. Bower and E.J. Crothers, An Introduction to Mathematical Learning Theory, New York: Wiley, 1965. Atkinson, R.C. & Shiffrin, R.M. (1968). Hu man memory: A proposed system and its control processes. In Spence, K.W. & Spence, J.T. (Eds.), The Psychology of Learning and Motivation (pp 89-195), New York: Academic Press. Basden, B. H., Basden, D. R., & Gargano, G. J. (1993). Directed forgetting in implicit and explicit memory tests: A comparison of methods. Journal of Experimental Psychology: Learning, Memo ry, and Cognition, 19, 603Â–616. Bjork (1970). Positive forgetting: The nonint erference of items intentionally forgotten. Journal of Verbal Learning and Verbal Behavior, 9, 255-268. Bjork, R.A., LaBerge, D., & Legrand, R. (1968). The modification of short-term memory through instructions to forget. Psychonomic Science, 10, 55-56. Bjork, R.A., & Whitten, W.B. (1974). Recency-sen sitive retrieval proc esses in long-term free recall. Cognitive Psychology, 6, 173-189. Block, R. A. (1971). Effects of instructi ons to forget in short-term memory. Journal of Experimental Psychology, 89 1Â–9. Brown, J. (1958). Some tests of the decay theory of immediate memory. The Quarterly Journal of Experi mental Psychology, 10, 12-21. Ebbinghaus, H. (1885). Memory: A contribution to experimental psychology. New York: Teachers College, Columbia University.
81 Eich, J. E., Weingartner, H., Stillman, R. C., & Gillin, J C. (1975). State-dependent accessibility of retrieval cues in th e retention of a categorized list. Journal of Verbal Learning & Verbal Behavior, 14(4), 408-417. Elmes, F. J., Adams, C., & Roediger, H. L. (1970). Cued forgetting in short-term memory: Response selection. Journal of Experiment al Psychology, 86, 103Â–107. Epstein, W. (1969). Poststimulus output specif ication and differenti al retrieval from short-term memory. Journal of Experime ntal Psychology, 82, 168-174. Estes, W. K. (1955). Statistical theory of spontaneous recovery and regression. Psychological Review, 62, 145Â–154. Francis, W. N., & Kucera, H. (1982). Frequency analysis of E nglish usage: Lexicon and grammar. Boston: Houghton Mifflin. Geiselman, R. E., Bjork, R. A., & Fishman, D. (1983). Disrupted retrieval in directed forgetting: A link with posthypnotic amnesia. Journal of Experimental Psychology: General, 112, 58Â–72. Glanzer, M. & Cunitz, A. R. (1966). Tw o storage mechanisms in free recall. Journal of Verbal Learning & Verbal Behavior, 5(4), 351-360. Godden, D., & Baddeley, A. D. (1975). Cont ext-dependent memory in two natural environments: On land and under water. British Journal of Psychology, 66, 325Â– 331. Godden, D., & Baddeley, A. D. (1980). When does context influence recognition memory? British Journal of Psychology, 71, 99Â–104. Goodwin, D. W., Powell, B., Bremer, D., Ho ine, H., & Stern, J. (1969). Alcohol and recall: State dependent effects in man. Science, 163, 1358Â–1360.
82 Jacoby, L. L (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Journal of Memory and Language, 30(5), 513-541. Jang, Y. & Huber, D. E. (2008). Context retr ieval and context change in free recall: Recalling from long-term memory drives list isolation. Journal of Experimental Psychology: Learning Memory and Cognition, 34, 112-127 Kahana, M. J. (1996). Associative re trieval processes in free recall. Memory & Cognition, 24, 103Â–109. Howard, M.W. & Kahana, M.J. (2002). A distri buted representation of temporal context. Journal of Mathematical Psychology, 46, 269-299. Kimball, D.R., Smith, T.A., & Kahana, M.J. (2007). The fSAM model of false recall. Psychological Review, 114(4) 954-993. Macht, M. L., Spear, N. E., & Levis, D. J. (1977). State-dependent retention in humans induced by alterations in affective state. Bulletin of the Psychonomic Society, 10, 415Â–418. MacLeod, C.M. (1975). Long-term recognition a nd recall following directed forgetting. Journal of Experimental Psyc hology: Human Learning and Memory, 1, 271Â–279. MacLeod, C.M. (1989). Directed forgetting aff ects both direct and indirect tests of memory. Journal of Experimental Ps ychology: Learning, Memory, and Cognition, 15, 13Â–21. MacLeod, C. M. (1998). Directed forgetting. In J. M. Golding & C. M. MacLeod (Eds.), Intentional forgetting: In terdisciplinary approaches (pp. 1Â–57). Mahwah, NJ: Erlbaum.
83 MacLeod, C. M. (1999). The item and list methods of directed forgetting: Test differences and the role of demand characteristics. Psychonomic Bulletin & Review, 6(1), 123-129. Malmberg, K. J., Holden, J. E., Shiffri n, R. M. (2004). Modeling the Effects of Repetitions, Similarity, and Normative Word Frequency on Old-New Recognition and Judgments of Frequency. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30(2), 319-331. Malmberg, K. J. & Shiffrin, R. M. ( 2005). The 'One-Shot' Hypothesis for Context Storage. Journal of Experimental Psychology : Learning, Memory, and Cognition, 31(2), 322-336. Mensink, G. J. M., & Raaijmakers, J. G. W. (1988). A model for interference and forgetting. Psychological Review, 95, 434Â–455. Mensink, G. J. M., & Raaijmakers, J. G. W. (1989). A model for c ontextual fluctuation. Journal of Mathematical Psychology, 33, 172Â–186. Murnane, K., Phelps, M. P., & Malmberg, K. (1999). Context-dependent recognition memory: The ICE theory. Journal of Experimental Psychology: General, 128, 403Â–415. Peterson, L. R., & Peterson, M. J. (1959). Short term retention of individual verbal items. Journal of Experi mental Psychology, 58, 193-198. Raaijmakers, J. G. W., & Shiffrin, R. M. (1981). Search of associative memory. Psychological Review, 88, 93Â–134. Rundus, D. (1971). Analysis of rehe arsal processes in free recall. Journal of Experimental Psychology, 89(1), 63-77.
84 Sahakyan, L. (2004). The destructive eff ects of the Â“forge tÂ” instructions. Psychonomic Bulletin & Review, 11, 555Â–559. Sahakyan, L., & Delaney, P. F. (2005). Direct ed Forgetting in Incidental Learning and Recognition Testing: Support for a Two-Factor Account. Journal of Experimental Psychology: Learning, Me mory, and Cognition, 31(4), 789-801. Sahakyan, L., & Kelley, C. M. (2002). A contex tual change account of the directed forgetting effect. Journal of Experimental Ps ychology: Learning, Memory, and Cognition, 28, 1064Â–1072. Shiffrin, R. M., & Steyvers, M. (199 7). A model for recognition memory: REMÂ— retrieving effectively from memory. Psychonomic Bulletin & Review, 4, 145Â–166. Shiffrin, R. M., & Steyvers, M. (1998). The effectiveness of retrieval from memory. In M. Oaksford & N. Chater (Eds.), Rational models of cognition (pp. 73Â–95). London: Oxford University Press. Slamecka, N.J. & McElree, B. (1983). Normal forgetting of verbal lis ts as a function of their degree of learning. Journal of Experi mental Psychology: Learning, Memory, and Cognition, 9(3), 384-397. Smith, S. M. (1979). Remembering in and out of context. Journal of Experimental Psychology: Human Learning and Memory, 5, 460Â–471. Winograd, E. (1968). List differentiati on, recall, and category similarity. Journal of Experimental Psychology, 78(3), 510-515. Wixted, J. T. & Stretch, V. (2004). In defens e of the signal-detec tion interpretation of Remember/Know judgments. Psychonomic Bulletin & Review, 11, 616-641.
85 Woodward, A.E., Jr., & Bjork, R.A. (1971). Fo rgetting and remembering in free recall: Intentional and unintentional. Journal of Experimental Psychology, 89, 109-116. Xu, J. & Malmberg, K. J.(in press). Modeling the effects of verbaland non-verbal pair strength on associ ative recognition, Memory & Cognition.