Neuropsychologia 44 (2006) 2806–2813

The mechanisms of errorless learning Nicole D. Anderson a,b,c,∗, Fergus I.M. Craik c,d a Kunin-Lunenfeld Applied Research Unit, Baycrest, 3560 Bathurst Street, Toronto, Ont. M6A 2E1, Canada b Department of Medicine (Psychiatry), University of Toronto, Canada c Department of Psychology, University of Toronto, Canada d Rotman Research Institute, Baycrest, Canada Received 10 November 2005; received in revised form 13 April 2006; accepted 22 May 2006 Available online 9 August 2006

Abstract Errorless learning enhances relative to errorful, trial-and-error learning, but the extent to which this advantage relies on implicit or processes is not clear. Previous attempts to determine the mnemonic mechanisms of errorless learning have relied on contrasts between patient groups or between tasks, but both approaches are problematic. In this study, healthy younger and older adults were engaged in errorless or errorful learning using a process dissociation procedure that provides separate estimates of explicit recollection and implicit familiarity within-subjects and within-task (Hay & Jacoby, 1996). Consistent with much prior research, we found an age-related decrement in recollection, but age-invariance in familiarity. In the young adults, errorless learning reduced the misleading familiarity of prior errors, but this benefit was offset by the non-elaborative nature of the errorless learning condition that also reduced recollection. In the older adults, who are less able to oppose familiarity- based errors because of their lower recollection, errorless learning only reduced the misleading impact of previous errors. Our results support Badde- ley and Wilson’s (1994) position that the errorless learning effect is mediated by processes: individuals with reduced explicit mem- ory benefit from errorless learning because errorless learning bypasses the need to engage explicit error elimination processes. We do not recommend standard errorless learning for individuals with intact explicit memory, such as students trying to learn information in preparation for an examination. © 2006 Elsevier Ltd. All rights reserved.

Keywords: Errorless learning; Aging; Memory; Explicit memory; Implicit memory; Process dissociation procedure

Errorless learning is a technique wherein individuals are pre- same general pattern was found for the older control partici- vented from making errors when initially learning information. pants, but the data were limited by ceiling effects in both control This technique, originally devised within the animal learning groups. Similar benefits of errorless learning in the amnestic field (Terrace, 1963), was first applied to the rehabilitation of patient KC using a different paradigm were reported by Hayman, memory impairments by Baddeley and Wilson (1994). Their Macdonald, and Tulving (1993). study compared the effects of errorful and errorless learning in The benefit of errorless learning for memory-impaired indi- a group of amnestic individuals and younger and older healthy viduals has since been replicated many times, in people with control participants. In errorful learning, participants were told, brain injury (Hayman et al., 1993; Komatsu, Mimura, Kato, for example, “I am thinking of a five-letter word that begins with Wakamatsu, & Kashima, 2000; McKenna & Gerhand, 2002; QU”. Participants then generated up to three errors before being Parkin, Hunkin, & Squires, 1998; Tailby & Haslam, 2003; told the correct word that they were to remember. In the errorless Wilson, Baddeley, Evans, & Shiel, 1994), Alzheimer’s disease condition, participants were told, for example, “I am thinking (Clare, Wilson, Carter, Roth, & Hodges, 2002), and schizophre- of a five-letter word that begins with QU and it is QUOTE”. nia (O’Carroll, Russell, Lawrie, & Johnstone, 1999). Moreover, Baddeley and Wilson found that in the amnestic participants, the technique has been applied broadly, for example to train word subsequent learning and memory was more successful if the ini- processing skills (Hunkin, Squires, Aldrich, & Parkin, 1998), tial learning experience was errorless rather than errorful. The proper names (Parkin et al., 1998), and face-name associations (Clare et al., 2002; Kalla, Downes, & Van den Broek, 2001). The cognitive mechanisms of the errorless learning effect are ∗ Corresponding author. Tel.: +1 416 785 2500x3366; fax: +1 416 785 4295. debated. Some argue for a role of implicit memory. For example, E-mail address: [email protected] (N.D. Anderson). Baddeley and Wilson (1994) proposed that implicit memory is

0028-3932/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.neuropsychologia.2006.05.026 N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 2807 unable to eliminate strong incorrect competing responses, and could be mediated either by recollection (R) or by familiarity hence amnestic individuals, who rely mainly on implicit mem- (F), as both would lead to a correct response (thus R or F). How- ory, benefit from the prevention of errors during learning (see ever, when participants studied atypical word pairs, they would also Evans et al., 2000). Others, however, argue that explicit provide the (incorrect) typical response word only if recollec- memory is responsible for the errorless learning effect. Hunkin, tion for the immediately preceding study list had failed and they Squires, Parkin, and Tidy (1998) compared indirect word frag- were relying on the familiarity-based memory habit developed ment completion and direct cued performance after error- in the first phase of the study (F alone). Hence, an estimate of less or errorful learning in a group of amnestic brain-injured recollection (R) is derived from the probability of a correct typ- individuals. They found an advantage for errorless learning in ical response on typical trials (R or F) minus the probability , a task that arguably relies heavily on explicit memory, of an erroneous typical response on atypical trials (F alone). An but no difference between the learning conditions for word frag- estimate of familiarity (F) is then computed by dividing the prob- ment completion, a task that relies mainly on implicit memory. ability of an incorrect typical response to atypical word pairs by Hunkin et al. concluded that explicit memory, however limited in 1-R (Jacoby, 1991). brain-injured populations, benefits from the prevention of errors Hay and Jacoby (1996, 1999) reported that manipulating the during learning (see also Tailby & Haslam, 2003, for a similar strength of the habit by changing the percentage of typical trials approach and conclusion). during training affected familiarity but not recollection, whereas These previous attempts to determine whether the errorless normal aging, faster presentation rates, and shorter response learning effect is mediated by implicit or explicit memory suffer deadlines affected recollection but not familiarity. These disso- because they address the question indirectly. As pointed out by ciations support the view that recollection and familiarity make many (e.g., Evans et al., 2000; Hunkin, Squires, Parkin, et al., independent contributions to memory performance within a task. 1998), even patients with severe typically have some The Hay and Jacoby (1996, 1999) paradigm is an error- degree of residual explicit memory. Similarly, neither direct ful learning paradigm, in that participants were made to guess nor indirect tests of memory are process-pure (Jacoby, Toth, during the training phase, and given that the atypical pair was & Yonelinas, 1993)—there is involvement of implicit memory presented on some of the trials, errors were common. We there- on direct tests (e.g., “‘dog’ just feels right so I am going to say fore replicated their paradigm as an errorful learning condition it” in a free recall test), and involvement of explicit memory in but also added a second errorless learning condition in which the indirect tests, an effect that has been called “explicit contami- word pairs were shown in full on each trial, thereby eliminating nation”. Hence, using either subject groups or memory tests to errors. Because any differences between the errorful and error- disentangle the contributions of explicit and implicit memory to less conditions could be attributed to participants not attending in errorless learning is problematic. the errorless condition (because they were not required to make Our goal was to explore the mnemonic mechanisms of error- any response), we added a third errorless condition in which the less learning more directly. To this end, we used the habit word pairs were also shown in full on each trial, but participants paradigm developed by Hay and Jacoby (1996). This paradigm were asked to read each response word out loud.1 uses a process dissociation procedure (Jacoby, 1991) that pits We tested healthy younger and healthy older adults with implicit and explicit memory against each other within a task.We the aims of replicating the common pattern of an age-related agree with Jacoby (1991) that the terms “explicit” and “implicit”, decrement in recollection but age invariance of familiarity (see meant to relate to memory processes, have become too entangled Jacoby, Jennings, & Hay, 1996), and exploring the mnemonic with particular memory tests (e.g., recall and word stem com- mechanisms of errorless learning. Recollection always leads pletion, respectively), and hence prefer the terms “recollection” to a correct response, but the accuracy of that are and “familiarity” to refer to the intentional and unintentional mediated by familiarity depends on the situation. In the current (respectively) memory processes that are involved during both experiment, reliance on familiarity with the typical responses explicit and implicit memory. established in the training phase leads to correct responses for In Hay and Jacoby’s (1999) paradigm, familiarity-based typical pairs, but it also leads to errors for atypical pairs. In memory habits were created in a training phase of the study, and general, when recollection and familiarity are in conflict, per- then was explored for a list of items that were formance will improve to the extent that one uses recollection to either congruent or incongruent with the habits. More specifi- oppose the misleading effects of familiarity. If errorless learning cally, in the training phase participants were presented with a eliminates the implicit influence of prior errors on current learn- word and a stem of an associated word (e.g., “knee-b n ”) and ing and memory, as Baddeley and Wilson (1994) contended, participants guessed what the second word was before it was then in an opposition paradigm as used in this study estimates presented. These pairs were made either “typical” by being pre- of familiarity should be lower following errorless than errorful sented on 75% of the trials (e.g., “knee-bend”) or “atypical” by learning (i.e., fewer erroneous typical responses should be emit- being presented on 25% of the trials (e.g., “knee-bone”). In the ted). If errorless learning facilitates explicit memory, as Hunkin, subsequent episodic memory phase, participants were presented Squires, Aldrich, et al. (1998) and Hunkin, Squires, Parkin, et al. with short lists of word pairs to study that were either typical or atypical, and were then given the first word from each pair and asked to recall the second word. When participants studied 1 We thank an anonymous reviewer on an earlier version of this manuscript typical word pairs, correct recall of the typical response word for this suggestion. 2808 N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813

Table 1 Participant demographic data Group Condition Age Education Mill Hill Digit symbol

M S.D. M S.D. M S.D. M S.D.

Errorful (n = 24) 21.0 3.3 14.8 2.2 18.5 3.3 93.0 12.0 Young Errorless (n = 24) 21.1 3.2 14.9 2.6 17.2 4.5 90.2 10.2 Errorless-speak (n = 12) 20.9 2.7 15.2 2.3 18.8 3.9 96.2 15.0 Errorful (n = 26) 73.6 6.0 15.2 3.3 21.6 5.0 58.4 12.2 Old Errorless (n = 26) 75.9 5.5 14.7 3.6 20.6 5.2 57.4 9.4 Errorless-speak (n = 12) 74.3 5.1 16.5 2.7 23.6 4.6 61.7 15.4

(1998) argued, then estimates of recollection should be higher In the training phase, participants were presented with a stimulus-word fragment following errorless than errorful learning. pair (e.g., knee-b n ), and were instructed to guess the response word. Partici- There is reason to believe, however, that errorless learn- pants had 2 s to respond (verbally) and then the correct response word for that trial was presented for 1 s, followed by a blank screen inter-stimulus interval ing may not be beneficial for memory in all cases (cf. Evans (ISI) for 500 ms. The Errorless condition proceeded in a similar way, except et al., 2000). Errorful learning requires participants to gener- that the complete stimulus-response pair was shown in full for 3 s on each trial, ate target memoranda when given an associated cue, whereas followed by a blank screen ISI for 500 ms. The Errorless-speak condition was errorless learning is entirely passive. We know from much pre- identical to the errorless condition, except that participants were instructed to vious research with healthy adults that elaborative processing read the response words out loud. The response words from one of the sets (A or B) were shown on 75% and generating to-be-remembered material improve later mem- of the trials (“typical” pairs), and the response words from the other set were ory performance (Craik & Tulving, 1975; Slamecka & Graf, shown on the remaining 25% of the trials (“atypical” pairs). Set assignment to 1978) and are associated with large increases in recollection typical and atypical pairs was counterbalanced across participants. Participants and smaller but reliable increases in familiarity (for a review were instructed to pay to the responses that were presented, that more see Yonelinas, 2002). Hence, recollection and familiarity may than one response word would appear for each stimulus word, and that some response words would appear more often than others. The examiner recorded in fact be higher following errorful than errorless learning. responses. The training phase consisted of three consecutive blocks of 128 trials, with each stimulus word presented eight times in each block with six typical 1. Method and two atypical response words. The order of the items within each block was random, with the restriction that the same stimulus word was not presented more 1.1. Participants than three times in a row. The entire training phase lasted approximately 25 min. After training, participants completed 16 successive study-test lists divided Sixty younger adults and 64 healthy older adults volunteered to participate. into two halves of eight lists. Each study list comprised eight word pairs from the Twenty-four younger adults and 26 older adults were assigned to each of the training phase—six typical and two atypical, maintaining the response proba- errorful and errorless conditions, and an additional 12 participants in each age bilities from the training phase. Participants in the errorful condition were again group participated in the errorless-speak condition (see procedure). Participants presented with the stimulus-word fragment pair and were instructed to guess the were recruited from volunteer research participant pools or responded to flyers response word within 2 s and then the correct response word for that trial was posted at the university, and were paid for their participation or received course shown for 1 s. Participants in the Errorless condition were shown the complete credit. Age and years of formal education did not differ between learning con- word pair for 3 s; the study phase proceeded the same way in the errorless-speak ditions within age group and the education level did not differ between younger condition, except participants were asked to read the response words aloud. After and older adults (all p > 0.28, see Table 1). Consistent with prior research com- each study list, a random number between 30 and 100 appeared on the screen paring younger and older adults, vocabulary scores were higher in the older than for 1 s, followed by a blank screen for 6.5 s, and participants were instructed to younger group, t(122) = 4.34, p < 0.001, whereas digit symbol performance was count backwards by threes from the presented number until the test list began. better in the younger than older group, t(122) = 15.94, p < 0.001. Vocabulary Each test list consisted of eight stimulus-word fragment pairs corresponding to and digit symbol scores did not differ between conditions for either age group studied items and participants were instructed to supply the response word from (all ps > 0.23). This study was approved by the Research Ethics Boards at the the immediately preceding study list. Participants were warned that each test list University of Toronto and Baycrest. would also contain pairs from the training phase that were not presented in the study list (two unstudied pairs were presented in each test list),2 and for these items, participants were to supply the first word that came to mind. Participants 1.2. Materials, design, and procedure in both conditions were given 3 s to respond and the examiner again recorded the responses. Test pairs were followed by a 550 ms ISI. Each typical pair was The materials were from Hay and Jacoby (1996), and consisted of 16 nouns, tested six times (three times in each half of the test phase), each atypical pair each paired with two associates that completed the same word fragment (e.g., was tested twice (once in each half of the test phase), and each unstudied pair bend/bone for knee-b n and farm/yard for barn- ar ). One associate for each appeared twice (once in each half of the test phase). No stimulus word was used homograph was assigned to Set A, and the other to Set B. more than once within a study-test list. The digit symbol task was administered Participants were tested individually using an IBM-compatible laptop (15 in. in between the two halves of the test phase. The entire study-test phase lasted screen) running Presentation software Version0.76. Words were presented in the approximately 25 min. middle of a white screen in black lowercase Times New Roman letters, with each letter approximately 15 × 10 mm in size. Participants were seated approximately 75 cm from the screen. Participants first completed the Mill Hill vocabulary test 2 Each pair was meant to appear as an unstudied pair twice in the study-test and then the memory task. phase, once in each half, but due to a programming error the two unstudied pairs The training phase and study-test phase procedures are illustrated in Fig. 1. were selected randomly (without replacement for that list) from the items that The errorful condition proceeded much as described by Hay and Jacoby (1999). were not in the current study list. N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 2809

Fig. 1. Schematic description of the training and study-test phase procedures. In the training phase and study phase for the errorful group, participants saw a stimulus word and an associated word fragment and have 2 s to guess what the target word was, and then the correct target was shown beside the cue word for 1 s. In the training phase and study phase for the errorless groups, participants were shown the correct full word pair for that trial for 3 s. In the test phases in all conditions, participants saw a cue word and an associated word fragment and had 3 s to recall the response word that was presented in the immediately preceding study list or provide the first word that came to mind.

2. Results

An alpha level of 0.05 was used for all statistical tests, and significant effects were decomposed using Sidak post hoc com- parisons. 2.1. Training data

The training data were analyzed using 2 × 3 repeated mea- sures ANOVAs, with Age group between-subjects and Block within-subjects. Note that training data are available only from the participants in the errorful condition, as participants in the errorless condition made no response during this phase of the study. Participants provided either the typical or the atypical response on the majority of trials, particularly by the second Fig. 2. Mean probability of providing a typical (white bars) or atypical (hatched block (see Fig. 2), but did occasionally provide another response bars) response in each block of the training phase (±1 S.E.M.). 2810 N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 or failed to respond. The probability of providing either the typ- ical or the atypical response was higher in the younger than older adults, F(1, 48) = 17.86, and increased across blocks, F(2, 96) = 162.90, particularly for the older adults, F(2, 96) = 11.98. The probability of providing a typical response did not dif- fer overall between younger and older adults, F(1, 48) < 1, but increased across blocks, F(2, 96) = 35.50, and this increase was particularly marked for the older adults from Blocks 1 to 2 (age group × Block interaction), F(2, 96) = 11.37. The probability of providing an atypical response was greater in the younger than older adults, F(1, 48) = 13.80, and increased from blocks 1 to 2, F(2, 96) = 4.17, marginally more so for younger than older adults (age group × Block interaction), F(2, 96) = 3.15, p = 0.05.

2.2. Study-test data

The probability of providing a typical response in the study- test phase, given that the test item was a typical studied pair, an atypical studied pair, or unstudied pair is shown in Fig. 3. These data were first analyzed using a 2 × 3 × 3 repeated mea- sures ANOVA, with Age group and learning condition between- subjects factors and test item type a within-subjects factor. All effects were significant at p < 0.05. To help decompose these interactions, separate 2 × 3 ANOVAs were conducted within each pair type. The probability of providing a (correct) typical response for typical study pairs (Fig. 3; panel A) was higher in younger than older adults, F(1, 118) = 28.83, and higher in the errorful than in the two errorless conditions (which did not differ from each other), F(2, 118) = 28.35, with no interaction between age group and learning condition, F(2, 118) < 1. The probability of pro- viding an (erroneous) typical response for atypical study pairs (Fig. 3; panel B) was higher in older than younger adults, F(1, 118) = 48.77, and there was an interaction between age group and learning condition, F(2, 118) = 4.76, as younger adults pro- vided comparable proportions of typical responses in the three learning conditions, F(2, 57) = 1.12 (n.s.), whereas older adults provided more erroneous typical responses in the errorful than in the errorless conditions, F(2, 61) = 4.50. The probability of providing a typical response for unstudied pairs (Fig. 3; panel C) was only affected by the learning condition, being higher follow- ing errorful than errorless learning, whether or not the response Fig. 3. Mean probability of providing a typical response to typical word pairs (panel A), atypical word pairs (panel B) and unstudied word pairs (panel C) in words were read aloud, F(2, 118) = 18.74 (other Fs < 1). Partic- the study-test phase (±1 S.E.M.). ipants provided either the typical or the atypical response on the majority of the trials (M = 0.97–0.99); hence, the comparable analyses of atypical responses provided mirror image results. Table 2 2.3. Recollection and familiarity Estimates of recollection and familiarity Age group Condition Recollection Familiarity Estimates of recollection and familiarity are shown in Table 2 M S.D. M S.D. and were analyzed in separate 2 × 3 ANOVAs, with age group and learning condition between-subjects factors. Recollection Errorful 0.62 0.19 0.70 0.20 was higher in younger than older adults, F(1, 118) = 77.09, and Young Errorless 0.49 0.16 0.64 0.09 Errorless-speak 0.41 0.19 0.62 0.13 was higher after errorful than errorless learning overall, F(2, 118) = 4.30, but the effects of Age group and Learning condi- Errorful 0.23 0.18 0.76 0.11 tion interacted, F(2, 118) = 3.37. This is because recollection Old Errorless 0.24 0.15 0.63 0.09 Errorless-speak 0.20 0.10 0.60 0.16 was higher after errorful than errorless learning for the younger N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 2811 adults, F(2, 57) = 6.34, but not for older adults, F(2, 61) < 1. the older adults were relying more on automatic, familiarity- Familiarity was greater following errorful than errorless learn- based processes during errorful learning. That is, although the ing, F(2, 118) = 9.84, but did not differ between Age groups in probabilities of providing a typical response during the training either condition. phase were similar for younger and older adults, the mnemonic mechanisms may have differed for the two age groups: errorful 3. Discussion training appears to have built stronger memory habits and recol- lections in younger adults, whereas it built only stronger memory Our goal was to explore the independent contributions of habits in the older adults. This idea that errorful, or feedback- explicit and implicit memory processes to errorless learning based, learning can occur either via implicit or explicit processes in healthy younger and older adults, using a process dissoci- is consistent with other research (e. g., Shohamy et al., 2004). ation procedure (Hay & Jacoby, 1996, 1999). We found that As described in the introduction, the memorial advantage of estimates of (explicit) recollection were reduced in the older errorless learning has been shown in many studies (see Kessels group relative to their younger counterparts, whereas estimates & De Haan, 2003, for a review). In the current study, error- of (implicit) familiarity did not differ between the two age less learning was actually less beneficial than errorful learning groups, consistent with much previous research (e.g., Hay & when correctly recalling highly probable items (i.e., typical word Jacoby, 1999; Jennings & Jacoby, 1997). The current study pairs). However, for the older adults errorless learning did reduce also revealed a dissociation between the mnemonic mechanisms the number of errors made when recalling less probable items underlying errorless learning in younger and older adults. In (i.e., atypical word pairs). In a paradigm such as that used in the younger adults, errorless learning reduced both recollection and current study, recollection and familiarity are placed in opposi- familiarity, whereas in older adults, it only reduced familiarity. tion with one another, and hence the optimal strategy would be to Hence for both groups, errorless learning reduced the mislead- rely on recollection to oppose the misleading influence of famil- ing, automatic influence of prior errors. This was also seen in the iarity. The older adults showed the typical age-related reduction responses to unstudied pairs. The training phase was designed in recollection relative to the younger adults, and hence were to create a memory habit such that participants would guess less adept at using recollection to suppress incorrect familiarity- the typical response on 75% of trials. In the errorful condition based responses (cf. Baddeley & Wilson, 1995; Jennings & this was clearly achieved, as younger and older participants pro- Jacoby, 1997). The reduction in the strength of activation of the vided typical responses to unstudied pairs on 76 and 77% of typical responses caused by errorless learning hence reduced trials, respectively.3 However, errorless learning lead to a weaker their likelihood of being emitted, leading to lower estimates of memory habit (64 and 63% for the younger and older adults, familiarity following errorless than errorful learning conditions. respectively). The reductions in familiarity and habit strength It should also be pointed out that the current paradigm dif- as a function of errorless learning are consistent with Baddeley fers from prior errorless learning paradigms in that it involved and Wilson’s (1994) claim that errorless learning works through probabilistic learning. On some trials a particular response was implicit memory. Moreover, the effects of errorless learning on correct, whereas on other trials a different response was correct. familiarity and habit strength cannot be explained by a failure We used this paradigm because it fostered a large proportion of to attend to the word pairs during training: participants in the errors during learning. It will be important in future studies to errorless-speak condition were required to read the response explore whether the current findings hold in paradigms in which words aloud, yet their performance did not differ from that of the correct response is invariant across learning trials. participants in the standard errorless learning condition. It would be interesting to use a remember-know paradigm For the younger adults, however, errorless learning was asso- (Gardiner & Richardson-Klavehn, 2000; Tulving, 1985)to ciated with an additional reduction in recollection. We know explore the effects of errorless learning on recollection and from other research that the more a person is engaged in elabora- familiarity. In this paradigm, a list of items could be presented tive, semantic processing of memoranda, the better the memory at study, in one or multiple study trials under conditions of (e.g., Craik & Tulving, 1975; Slamecka & Graf, 1978). Indeed, errorless or errorful learning, and then in a recognition task, Tailby and Haslam (2003) recently demonstrated that the error- participants would indicate whether each item is new, remem- less learning advantage is amplified if participants engaged in bered from the study list along with specific details from the more active strategies. Similarly, Rodriguez-Fornells, encoding context, or whether they simply know the item was Kofidis, and Munte¨ (2004) reported that recognition hit rates on the study list but cannot recollect the actual encoding event. were higher for words studied in an errorful than errorless learn- Estimates of recollection and familiarity can be derived from this ing condition, and also suggested that the requirement to guess paradigm (Jacoby, Yonelinas, & Jennings, 1997). Hence, learn- the target information leads to deeper levels of processing. The ing would be non-probabilistic, given that the “correct” answer fact that generating a semantic associate in the errorful condition would not change from trial to trial. Nevertheless, given the cur- increased recollection only for the younger adults suggests that rent results, we would expect the errorless learning advantage to be mediated by familiarity, rather than recollection. Moreover, in a remember-know paradigm, recollection and familiarity are 3 This match between actual and observed probabilities of a typical response not in opposition but rather work in concert supporting correct is all the more interesting given the fact that the allocation of word pairs to responses. There would be no penalty for relying on familiar- unstudied items was not balanced as intended (see footnote 2). ity in this case, because either process would lead to correct 2812 N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 responses for studied items, and hence, errorless learning may learning reduces the implicit activation of these errors. How- well increase estimates of familiarity. ever, errorless learning also reduced recollection in the younger Kessels, Te Boekhorst, and Postma (2005) recently reported adults. Standard errorless learning methods may not be ideal for another process dissociation study of errorless and errorful learn- individuals with intact explicit memory processes (e.g., in edu- ing in younger and older adults in a task. Their cation settings to help students prepare for exams) because these study involved a single learning trial, in which participants were methods under-utilize self-initiated, elaborative processes. The either shown (errorless learning) or guessed (errorful learning) key to successful memory rehabilitation for all groups may be to the location of objects in a scene. This was followed by an design encoding environments that are error-free (thereby reduc- Inclusion memory task in which participants were to place the ing the misleading familiarity of errors), but nonetheless require object in the target location and an Exclusion memory task in people to engage in elaborative processing (thereby increasing which participants were to place the object in an incorrect loca- recollection). tion. Recollection or familiarity would lead participants to place objects in the target location on the Inclusion task, but only famil- iarity, in the absence of recollection, would result in erroneous Acknowledgments placement of objects in the target location on the exclusion task. Kessels et al. (2005) reported lower recollection in older than We are grateful to Samy Arita, Cherylyn Dickson, and Sharyn younger adults, but no age-related differences in familiarity, the Krueger for data collection, to Kyla MacKay for data checking, same age-related pattern that was found in the current study. to Jimmy Shen for computer programming, to Janine M. Hay However, in their study errorless learning affected recollection for sharing her stimuli, and to Larry L. Jacoby for his advice on but not familiarity, and did so only for the younger adults such the design of this experiment. This work was supported by an that recollection estimates were higher in errorless than error- operating grant (RGPIN 238361-03) from the Natural Sciences ful learning. There are a few potential explanations why the and Engineering Council of Canada awarded to the first author. current results and those of Kessels et al. diverge. First, some have argued that the different task instructions for Inclusion and References Exclusion confuse participants and can lead to different levels of response bias (Curran & Hintzman, 1995; Graf & Komatsu, Baddeley, A., & Wilson, B. A. (1994). When implicit learning fails: Amnesia 1994). One advantage of the current paradigm over the inclu- and the problem of error elimination. Neuropsychologia, 32(1), 53–68. sion/exclusion paradigm is that estimates of recollection and Clare, L., Wilson, B. A., Carter, G., Roth, I., & Hodges, J. R. (2002). Relearn- familiarity are obtained within a single task and using a single ing face-name associations in early Alzheimer’s disease. Neuropsychology, 16(4), 538–547. set of instructions. Second, it is unclear how many errors were Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of actually made during errorful learning in Kessels et al.’s study: if words in episodic memory. Journal of Experimental Psychology: General, subject selected the correct object location, the next trial was pre- 104(3), 268–294. sented. In our study, younger and older adults both made mean Curran, T., & Hintzman, D. L. (1995). Violations of the independence assump- totals of 178 erroneous guesses during learning (out of 384 tri- tion in process dissociation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 531–547. als). Third, participants made fewer errors on the exclusion task Evans, J. J., Wilson, B. A., Schuri, U., Andrade, J., Baddeley, A., Bruna, O., et al. in the study by Kessels et al. than in our comparable condi- (2000). A comparison of “errorless” and “trial-and-error” learning methods tion (i.e., typical responses to atypical study pairs): roughly 12 for teaching individuals with acquired memory deficits. Neuropsychological and 16% for the younger and older adults in their study versus Rehabilitation, 10(1), 67–101. 29 and 60% in the current study. Low error rates pose prob- Gardiner, J. M., & Richardson-Klavehn, A. (2000). Remembering and knowing. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. lems for estimating familiarity within participants, and indeed 229–244). Oxford: Oxford University Press. Kessels et al. were unable to estimate familiarity for nine of their Graf, P., & Komatsu, S. (1994). Process dissociation procedure: Handle with participants, a problem that was not encountered in the current caution!. European Journal of Cognitive Psychology, 6, 113–129. study. The underestimation of familiarity estimates in their study Hay, J. F., & Jacoby, L. L. (1996). Separating habit and recollection: Memory would have limited their chances of observing an effect of error- slips, process dissociations, and probability matching. Journal of Experi- mental Psychology: Learning, Memory, and Cognition, 22, 1323–1335. less learning on implicit memory processes. Nevertheless, there Hay, J. F., & Jacoby, L. L. (1999). Separating habit and recollection in young and remains the possibility that in some circumstances, the errorless older adults: Effects of elaborative processing and distinctivness. Psychology learning effect is not only mediated through familiarity, as was and Aging, 14(1), 122–134. found in the current study, but also increases conscious recollec- Hayman, C. A. G., Macdonald, C. A., & Tulving, E. (1993). The role of repetition tion, as reported by Kessels et al. This will have to be explored and associative interference in new semantic learning in amnesia: A case experiment. Journal of Cognitive Neuroscience, 5(4), 375–389. in future studies. Hunkin, N. M., Squires, E. J., Aldrich, F. K., & Parkin, A. J. (1998). Errorless In summary, the current study demonstrated an age-related learning and the acquisition of word processing skills. Neuropsychological difference in the mnemonic mechanisms of errorless learning. Rehabilitation, 8(4), 433–449. For both age groups, errorless learning reduced the misleading Hunkin, N. M., Squires, E. J., Parkin, A. J., & Tidy, J. A. (1998). Are the bene- effects of prior errors. Errorless learning is effective for individ- fits of errorless learning dependent on implicit memory? Neuropsychologia, 36(1), 25–36. uals with explicit memory deficits (e.g., older adults, people with Jacoby, L. L. (1991). A process dissociation framework: Separating automatic amnesia or Alzheimer’s disease) because their ability to oppose from intentional uses of memory. Journal of Memory and , 30, the misleading influence of prior errors is reduced, and errorless 513–541. N.D. Anderson, F.I.M. Craik / Neuropsychologia 44 (2006) 2806–2813 2813

Jacoby, L. L., Jennings, J. M., & Hay, J. F. (1996). Dissociating automatic and McKenna, P., & Gerhand, S. (2002). Preserved semantic learning in an amnesic consciously controlled processes: Implications for diagnosis and rehabilita- patient. Cortex, 38, 37–58. tion of memory deficits. In D. Herrmann, C. McEvoy, C. Hertzog, P. Hertel, O’Carroll, R. E., Russell, H. H., Lawrie, S. M., & Johnstone, E. C. (1999). & M. K. Johnson (Eds.), Basic and applied memory research: Theory in Errorless learning and the cognitive rehabilitation of memory-impaired context: Vol. 1, (pp. 161–193). Mahwah, NJ: Erlbaum. schizophrenic patients. Psychological Medicine, 29, 105–112. Jacoby, L. L., Toth, J. P., & Yonelinas, A. P. (1993). Separating conscious Parkin, A. J., Hunkin, N. M., & Squires, E. J. (1998). Unlearning John Major: and unconscious influences of memory: Measuring recollection. Journal of The use of errorless learning in the reacquisition of proper names fellowing Experimental Psychology: Learning, Memory, and Cognition, 122, 139–154. herpes simplex encephalitis. Cognitive Neuropsychology, 15(4), 361–375. Jacoby, L. L., Yonelinas, A. P., & Jennings, J. M. (1997). The relation between Rodriguez-Fornells, A., Kofidis, C., & Munte,¨ T. F. (2004). An electrophysio- conscious and unconscious (automatic) influences: A declaration of inde- logical study of errorless learning. Cognitive Brain Research, 19, 160–173. pendence. In J. Cohen & J. W. Schooler (Eds.), Scientific approaches to Shohamy, D., Myers, C. E., Grossman, S., Sage, J., Gluck, M. A., & Poldrack, R. consciousness (pp. 13–47). Mahwah, NJ: Erlbaum. A. (2004). Cortico-striatal contributions to feedback-based learning: Con- Jennings, J. M., & Jacoby, L. L. (1997). An opposition procedure for detecting verging data from neuroimaging and neuropsychology. Brain, 127, 851– age-related deficits in recollection: Telling effects of repetition. Psychology 859. and Aging, 12(2), 352–361. Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a Kalla, T., Downes, J. J., & Van den Broek, M. (2001). The pre-exposure tech- phenomenon. Journal of Experimental Psychology: Human Learning and nique: Enhancing the effects of errorless learning in the acquisition of Memory, 4(6), 592–604. face-name associations. Neuropsychological Rehabilitation, 11(1), 1–16. Tailby, R., & Haslam, C. (2003). An investigation of errorless learning in Kessels, R. P. C., & De Haan, E. H. F. (2003). Implicit learning in memory reha- memory-impaired patients: Improving the technique and clarifying theory. bilitation: A meta-analysis on errorless learning and vanishing cues methods. Neuropsychologia, 41, 1230–1240. Journal of Clinical and Experimental Neuropsychology, 25(6), 805–814. Terrace, H. S. (1963). Errorless discrimination learning in the pigeon: Effects Kessels, R. P. C., Te Boekhorst, S., & Postma, A. (2005). The contribution of chlorpromazine and imipramine. Science, 140, 318–319. of implicit and explicit memory to the effects of errorless learning: A Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26(1), comparison between young and older adults. Journal of the International 1–12. Neuropsychological Society, 11, 144–151. Wilson, B. A., Baddeley, A., Evans, J., & Shiel, A. (1994). Errorless learning Komatsu, S., Mimura, M., Kato, M., Wakamatsu, N., & Kashima, H. (2000). in the rehabilitation of memory impaired people. Neuropsychological Reha- Errorless and effortful processes involved in the learning of face-name bilitation, 4(3), 307–326. associations by patients with alcoholic Korsakoff’s syndrome. Neuropsy- Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of chological Rehabilitation, 10(2), 113–132. 30 years of research. Journal of Memory and Language, 46, 441–517.