- Email: [email protected]
Recognition memory for pseudowords
Journal of Memory and Language 50 (2004) 259–267
Journal of Memory and Language www.elsevier.com/locate/jml
Recognition memory for pseudowordsq Robert L. Greene* Department of Psychology, Case Western Reserve University, 10900 Euclid Avenue, Cleveland, OH 44106-7123, USA Received 30 June 2003; revision received 24 October 2003
Abstract Participants are more likely to give positive responses on a recognition test to pseudowords (pronounceable nonwords) than words. A series of experiments suggests that this difference reflects the greater overall familiarity of pseudowords than of words. Pseudowords receive higher ratings of similarity to a studied list than do words. Pseudowords receive more positive recognition responses than words even when pseudowords are remembered at least as well as words. This pseudoword effect is found on forced-choice recognition and on frequency judgments but not in associative recognition. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Familiarity; Frequency; Mirror effect; Pseudowords; Recognition; Similarity
The mirror effect (Glanzer & Adams, 1985) is the finding that, in recognition experiments involving two classes of stimuli, the class with the higher hit rate tends to have the lower false-alarm rate. This effect has been most often demonstrated through comparisons of lowfrequency and high-frequency words. The mirror effect is commonly portrayed as one of the regularities of recognition memory (Glanzer, Adams, Iverson, & Kim, 1993). Indeed, the mirror effect has been listed (along with the list-strength effect) as one of the two phenomena ‘‘that are at the heart of testing and evaluating the models’’ of recognition memory (Ratcliff & McKoon, 2000, p. 575). Despite the influence of the mirror effect, it is possible to find exceptions to this pattern. One of the best-documented exceptions involves comparisons of words with pseudowords (pronounceable nonwords). The standard finding here is that pseudowords (or, alternatively, very low-frequency words that can be assumed to be
q This research was presented at the 2002 meeting of the Psychonomic Society in Kansas City, MO. * Corresponding author. Fax: 1-216-368-4891. E-mail address: [email protected].
unknown by all participants) tend to have both higher hit rates and higher false-alarm rates than do words (Hintzman & Curran, 1997; Hockley & Niewiadomski, 2001; Whittlesea & Williams, 2000, 2001; Wixted, 1992). This finding, that overall positive-response rates are higher for pseudowords than for words, will be known here as the pseudoword effect. Several related explanations have been proposed for this pseudoword effect. This class of related explanations will be referred to here as the overcompensation account. According to this class of explanations, participants know that pseudowords may be less memorable than words. On a recognition test, participants try to equate for this memorability difference, either by adopting a lower criterion for pseudowords than for words (Hockley & Niewiadomski, 2001; Stretch & Wixted, 1998) or by rescaling the familiarity values of the pseudowords (Hintzman & Curran, 1997). However, participants systematically overcompensate, so that pseudowords become far more likely to receive positive responses than do words. Experiments 1–3 test this overcompensation explanation, while Experiments 4–8 address an alternative account, namely, that pseudowords receive more positive responses than words because they are higher in familiarity.
0749-596X/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jml.2003.12.001
260
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Experiment 1 One version (Hockley & Niewiadomski, 2001; Stretch & Wixted, 1998) of the overcompensation approach attributes the pseudoword effect to the use of a lower response criterion for pseudowords than for words. If this version were true, then one may not expect to find the pseudoword effect in forced-choice recognition. It is often assumed that no response criteria are needed on forced-choice tests (e.g., Hicks & Marsh, 1998; Lockhart, 2000; Macmillan, 1993; Murdock, 1982a). However, Wixted (1992, Experiment 1) reported results that suggest that a pseudoword effect may be found in forced-choice recognition; he found a tendency for subjects to select rare words more often on this test than words of higher frequency. Experiment 1 was an attempt to replicate this pattern using words and pseudowords. Method Participants Sixteen students from introductory-psychology classes participated to fulfill a course requirement. Materials In the experiments reported here, the materials used were the 60 pseudowords (regular nonwords) and 60 words used by Whittlesea and Williams (listed in Appendix A of their 2000 paper). A different random ordering of stimuli and random assignment of stimuli to experimental conditions was presented to each participant. Procedure Participants were tested individually. They were asked to study a list for an unspecified memory test. A list of 30 words and 30 pseudowords intermixed was shown one item at a time on a computer screen at a 1 s rate. Immediately after presentation of the list, a self-paced forced-choice recognition test was administered on the computer. This test consisted of 60 test pairs, each containing one word and one pseudoword. Participants were told that each test pair contained one item that had been presented on the study list and one that had not been. They were asked to select the studied item in each test pair and were informed that words and pseudowords were equally likely to be the correct response. Results The overall proportions of items given positive responses are shown in Table 1 for words and pseudowords for all experiments reported here. Unless otherwise noted, a significance level of .05 is used for all statistical tests.
Table 1 Proportions of positive responses given to words and pseudowords Experiment 1 2 3 5 6
7 8
(Forced-Choice) (Repeated pseudowords) (Directed forgetting of words) (1 s Presentation rate) (3 s Presentation rate) (Overall recognition) (Remember responses) (Know responses) (Guess responses) (Frequency judgment) (Associative recognition)
Words .38 .41 .29 .36 .44 .38 .21 .13 .04 .33 .46
Pseudowords .62 .62 .61 .61 .58 .60 .19 .33 .08 .49 .46
A pseudoword effect in forced-choice recognition was found, as all 16 participants selected the pseudoword as the old item more often than the word (p < :01 by sign test). As a result of this tendency, participants were more accurate on test pairs when the pseudoword was the correct answer (.85) than when the word was the correct choice (.62), F ð1; 15Þ ¼ 21:30, MSe ¼ 18:40.
Experiment 2 The standard overcompensation account of the pseudoword effect attributes the lower criterion used for pseudowords to participantsÕ beliefs that words are much more memorable than pseudowords. According to this approach, if participants were put in a situation where they know that words are at least as memorable as pseudowords, no such pseudoword effect should be found. In Experiment 2, this was accomplished by presenting all pseudowords twice and all words once. Method Participants Twenty students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants were told that they would be studying a mixed list of words and pseudowords (pronounceable nonwords) for an unspecified study list. They were told that the experimenter wanted to equate memory for the words and the pseudowords and that all words would be presented once and all pseudowords would be presented twice. They then saw the study list of 90 events (30 words presented once with 30 pseudowords presented twice) at a 1 s rate. The two occurrences of a repeated item were always separated by 10–15 intervening items.
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Immediately after completion of the study list, participants were reminded that all pseudowords were presented twice and all words once and were asked whether they thought this was effective in approximately equating memory for the stimulus classes; all participants agreed with this claim. A self-paced yes–no recognition test on 120 items, shown individually on a computer screen, was then given. Results Overall proportions of positive responses are shown in Table 1. Recognition accuracy was approximately equated for the two classes, with mean A0 values being .80 and .80 for words and pseudowords, F < 1:0. Despite this equating of overall accuracy and participantsÕ seeming acknowledgment of this fact, a pseudoword effect was still found, as is clear from the overall proportions of positive responses given in Table 1. Pseudowords exhibited higher hit rates (.82) and falsealarm rates (.42) than did words (.61 and. 21, respectively). An analysis of variance on numbers of positive responses was conducted using stimulus class (word/ pseudoword) and study status (old/new) as factors. This analysis found significant main effects of both stimulus class, F ð1; 19Þ ¼ 47:03, MSe ¼ 16:88, and study status, F ð1; 19Þ ¼ 229:62, MSe ¼ 12:75, but no significant interaction between them, F < 1:0.1
Experiment 3 Experiment 3 was a conceptual replication of Experiment 2 in which the pseudoword effect was examined in another situation where the typical memory advantage for words was eliminated. Here it was done by using the directed-forgetting procedure (see MacLeod, 1998, for a review). Before presentation of the list, participants were instructed to remember the pseudowords and to forget the words.
1
The logic of this research program has been to use a constant set of stimuli while varying the procedure across experiments. However, there is a need to establish some generality across stimuli. For this purpose, Experiment 2 was replicated using a set of 20 new participants from the same source as the other experiments and using an identical procedure to that of Experiment 2. A new set of stimuli was created, consisting of 30 words with frequencies between 10 and 40 occurrences per million in the Kucera and Francis (1967) corpus and 30 rare words that did not occur in the Kucera and Francis corpus but are included in the Oxford English Dictionary. A pseudoword effect was again obtained, with the hit rates being .80 and .65 for pseudowords and words respectively; the false-alarm rates were .40 and .22. Statistical analyses led to the same conclusions as Experiment 2.
261
Method Participants Twenty introductory-psychology students participated to fulfill a course requirement. Procedure Participants were told that they would be seeing an intermixed list of 30 words and 30 pseudowords at a 1 s rate and were told that they must try to remember the pseudowords and to ignore the words. To help them do this, the word FORGET was printed above each word as it was shown, and the word REMEMBER was shown above each pseudoword. Immediately after seeing the list, participants were asked whether they thought they remembered the pseudowords from the list better than the words, and all agreed with this assertion. They then received a 120-item yes–no recognition test. Results Overall proportions of positive responses are shown in Table 1. Recognition accuracy was better for the pseudowords (mean A0 ¼ :76) than for the words (mean A0 ¼ :69), F ð1; 19Þ ¼ 5:33, MSe ¼ 0:01. Despite this reversal in accuracy, a pseudoword effect was found, with pseudowords showing both higher hit rates (.77) and false-alarm rates (.44) than words (.39 and. 19, respectively). An analysis of variance on numbers of positive responses found both significant effects of stimulus type (word/pseudoword), F ð1; 19Þ ¼ 47:36, MSe ¼ 37:71, and study status (old/new), F ð1; 19Þ ¼ 148:88, MSe ¼ 8:38. There was also a significant interaction, F ð1; 19Þ ¼ 13:41, MSe ¼ 5:38, indicating that the difference between the stimulus types was greater on old items than on new items; this reflects the pseudoword advantage in accuracy. The results of the first three experiments are inconsistent with an overcompensation account of the pseudoword effect, which would assume that participants use a lower criterion (or rescaled familiarity values) for pseudowords because they believe them to be less familiar than words. The effect was found when a forced-choice test was used (Experiment 1) and when the word advantage in recognition accuracy was either eliminated (Experiment 2) or reversed (Experiment 3). An alternative could be that pseudowords truly tend to be more familiar than words (see Wixted, 1992, p. 689, for a similar suggestion). It is typically assumed that familiarity is determined by the similarity of a test stimulus to a study list (e.g., Gillund & Shiffrin, 1984; Hintzman, 1988; Murdock, 1982b; Ratcliff & McKoon, 2000). All words contain strong semantic features that make each one unique and distinctive. Although words may be studied more effectively than pseudowords, they
262
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
are less similar to other items on the study list, reducing their overall sense of familiarity. In contrast, pseudowords, while not completely devoid of meaningful associations, presumably have much less semantic content. Their similarity to the study list would be based largely on their structural features (e.g., letters and phonemes). Insofar as these structural features overlap greatly between items, an assessment of familiarity based chiefly upon structural features could yield higher values than one in which meaning plays a substantial role. The occurrence of the pseudoword effect may indeed depend upon structural overlap between pseudowords and words. When the pseudowords are constructed so that none of them resemble any common words in the English language, no pseudoword effect is found (Groninger, 1976; Whittlesea & Williams, 2000). Similarly, nonwords that consist merely of strings of consonants and therefore have little resemblance to English words do not show the pseudoword effect (Greene, 1996). Rather than relying on intuitions, it was important to collect evidence that pseudowords truly may seem more familiar (i.e., more similar to a study list) than words. In Experiment 4, such similarity was assessed directly.
Experiment 5 If the pseudoword effect truly results from greater familiarity of pseudowords than words, a manipulation that lessens participantsÕ reliance on familiarity should reduce the effect. One such manipulation should be presentation rate: as a presentation rate is slowed, participants may have a greater opportunity to perform the elaborative and associative processing that would allow them to base their later recognition decisions on recollection of the original episodes. Some support for the notion that slower presentation rates enhance recollection more than familiarity comes from studies employing the Remember/Know procedure, where study duration has been shown to have a greater effect on the probability that participants will claim that they remember an old item than on the probability that they will merely know it had been presented (Dewhurst & Anderson, 1999; Hirshman, Fisher, Henthorn, Arndt, & Passannante, 2002; Hirshman & Henzler, 1998). Method Participants Twenty-four students from introductory-psychology classes participated to fulfill a course requirement.
Experiment 4 Method Participants Sixteen students from introductory-psychology classes participated to fulfill a course requirement. Procedure All participants saw a 60-item study list (30 words and 30 pseudowords intermixed) in anticipation of an unspecified memory test. Immediately after presentation of the list, participants were given a similarity-rating test. They were shown 30 new words and 30 new pseudowords intermixed. They were told that none of these items had appeared on the study list. They were asked to rate on a seven-point scale (with seven being maximum) the similarity of each test stimulus to the study list. Participants did not request, and were not offered, information as to how they should conceive of similarity for these ratings. Results Fifteen of the 16 participants gave higher mean ratings to pseudowords than to words. There was a significant difference between the mean rating given to pseudowords (4.56) and the mean rating given to words (3.31), F ð1; 15Þ ¼ 125:00, MSe ¼ 0:10.
Procedure Presentation rate of the study list (1 vs. 3 s) was manipulated between participants. Immediately after presentation of the 60-item study list (30 words and 30 pseudowords intermixed), a yes–no recognition test on 120 items was given. Results Overall proportions of positive responses are shown in Table 1. For the 3-s presentation rate group, hit rates were .77 and .73 for pseudowords and words respectively; false-alarm rates were .39 and .15. For the 1-s group, hit rates were .81 and .61, and false-alarm rates were .41 and .12 for pseudowords and words. An analysis of variance on number of positive responses was conducted using the between-subject factor of presentation rate and the within-subject factors of stimulus type (word/pseudoword) and study status (old/ new). Stimulus type had a significant main effect, F ð1; 22Þ ¼ 68:10, MSe ¼ 11:57, as did study status, F ð1; 22Þ ¼ 298:56, MSe ¼ 15:34. Critically, there was a significant interaction between stimulus type and presentation rate, F ð1; 22Þ ¼ 4:80, MSe ¼ 11:57, indicating that the pseudoword effect was reduced with the slower presentation rate. There was also a significant interaction between stimulus type and study status, F ð1; 22Þ ¼ 16:94, MSe ¼ 7:31, indicating that the
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
pseudoword effect was greater on false alarms than on hits. No other interactions were significant.
Experiment 6 If the pseudoword effect reflects familiarity differences between stimuli, then one should expect to see differences between words and pseudowords in the subjective experiences of remembering. A common way of studying subjective experiences in recognition has been to ask participants to discriminate between two different states, a Remember state where the prior occurrence of an item could be specifically remembered and a Know state where the participant knows an item was on the list without recollecting the experience (Tulving, 1985). Gardiner and Java (1990, Experiment 2) reported that pseudowords received more Know responses than did words (see also Rajaram, Hamilton, & Bolton, 2002); that study is replicated here using the stimuli of Whittlesea and Williams (2000). Following the recommended procedure of Gardiner and RichardsonKlavehn (2000), all participants were asked to classify their positive recognition responses as Remember, Know, or Guess judgments. Method Participants Twenty-four students from introductory-psychology classes participated to fulfill a course requirement. Procedure The study list consisted of 30 words and 30 pseudowords intermixed shown at a 1 s rate. Participants were then given a 120-item yes–no recognition test. For each positive recognition response, participants were asked to classify their subjective recognition experience as a Remember experience, a Know experience, or a Guess (with instructions based on the sample instructions provided in Gardiner & Richardson-Klavehn, 2000). That is, if the participant felt that the recognition decision was accompanied by a recollection of the experience, then a Remember response would be made. If the item appeared familiar but no recollective experience occurred, then a Know response would be made. The participant was also allowed to classify responses as Guesses. Results Overall proportions of positive responses are shown in Table 1. The pseudoword effect was evident in overall recognition and in Know responses and Guess responses but not in Remember responses. In overall recognition, the hit rates were .76 for pseudowords and .58 for words;
263
the corresponding false-alarm rates were .43 and .18. In overall recognition, there were significant effects of study status (old/new), F ð1; 23Þ ¼ 242:32, MSe ¼ 12:17, and stimulus type, F ð1; 23Þ ¼ 86:54, MSe ¼ 11:87, as well as a significant interaction between these factors, F ð1; 23Þ ¼ 4:38, MSe ¼ 7:45. Participants were asked to classify recognition decisions based upon recollection as Remember responses. There was no pseudoword effect on these items. The proportion of old items that received Remember responses were .30 and .37 for pseudowords and words respectively; the corresponding false-alarm rates were .08 and .05. An analysis of variance was carried out on the number of Remember responses with the factors of study status and stimulus type. The main effect of study status was significant, F ð1; 23Þ ¼ 110:69, MSe ¼ 14:17. The interaction between study status and stimulus type was also significant, F ð1; 23Þ ¼ 16:31, MSe ¼ 3:13, reflecting the advantage for words over pseudowords in accuracy. However, the main effect of stimulus type was not significant, F ð1; 23Þ ¼ 1:19, MSe ¼ 3:51. Items that were familiar but were not recollected received Know responses. The proportion of old items receiving Know responses were .38 and .17 for pseudowords and words respectively; the corresponding falsealarm rates were .27 and .08. An analysis of variance on the number of items receiving Know responses found significant main effects of study status, F ð1; 23Þ ¼ 36:49, MSe ¼ 6:08, and stimulus type, F ð1; 23Þ ¼ 85:66, MSe ¼ 10:09. The interaction was not significant, F ð1; 23Þ ¼ 1:01, MSe ¼ 4:12.2 Participants were asked to make a Guess response only when they made a positive recognition response in the absence of both recollection and familiarity. As a result, this option was not used often. However, a pseudoword effect was found on Guess responses. The proportion of old items that received Guess responses were. 08 and. 05 for pseudowords and words respectively; the corresponding false-alarm rates were. 08 and. 04. There was a significant effect of stimulus type, F ð1; 23Þ ¼ 11:94, MSe ¼ 1:85. Neither the main effect of study status nor the interaction were significant (in both cases, F < 1:0). The presence of a pseudoword effect on Guess responses is consistent with the claim by Yonelinas (2002) that these responses are familiarity-based and involve a subjective experience similar to that found with Know responses.
2
Yonelinas (2002) advocated measuring familiarity as the probability of a Know response given that a Remember response was not given (i.e., K=½1 R). When this measure is used, the probability of a positive response based on familiarity was .18 for words and .42 for pseudowords. The pseudoword effect was still significant using this measure, F ð1; 23Þ ¼ 100:34, MSe ¼ 0:02.
264
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Experiment 7 If the pseudoword effect reflects familiarity differences, then it should be found on tests other than recognition that also involve assessments of familiarity. One such task is frequency judgment (Hintzman, 1988, 2001; Hintzman & Hartry, 1990). In this experiment, participants received a study list of items presented once or twice. The frequency-judgment test required participants to remember how often a test stimulus had occurred on the list. Method Participants Sixteen students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants were asked to study a list for an unspecified memory test. The study list consisted of 180 presentations (60 items presented twice and 60 items presented once). A different random assignment of stimuli to frequencies was obtained for each participant, with the constraint that equal numbers of words and pseudowords must be shown at each frequency. Occurrences of a repeated item were separated by 10–15 presentations. Immediately after the study list, participants were given a test in which they were shown the 120 unique stimuli shown on the study list. They were asked to give each stimulus a response of 1 or 2 to indicate the number of times it had been shown on the study list. Results Proportions of positive responses (i.e., frequency judgments of 2) are shown in Table 1. A pseudoword effect was found, with participants giving responses of 2 to .60 of twice-presented pseudowords, .51 of twicepresented words, .38 of once-presented pseudowords, and .16 of once-presented words. The main effects of stimulus type, F ð1; 15Þ ¼ 18:90, MSe ¼ 16:44, frequency, F ð1; 15Þ ¼ 137:83, MSe ¼ 8:45, and the interaction, F ð1; 15Þ ¼ 62:02, MSe ¼ 5:55, were all significant.
Experiment 8 In associative recognition, participants study a list of pairs. On a subsequent test, they are asked to give positive responses to intact pairs (i.e., pairs containing two items that had been presented together on the study list) and negative responses to rearranged pairs. Associative recognition is often viewed as a test relying primarily on recollection, rather than familiarity (e.g., Bain & Humphreys, 1988; Cameron & Hockley, 2000; Clark &
Gronlund, 1996; Westerman, 2000). If this interpretation is true and if the pseudoword effect is due to the greater familiarity of pseudowords than words, then no pseudoword effect should be found in associative recognition. Method Participants Twenty students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants studied a list of 60 pairs, shown at a 2 s rate. Each pair consisted of either two words or two pseudowords. Assignment of stimuli to pairs was done randomly for each participant. Immediately after the study list, participants were given an associative recognition test. This test consisted of 60 pairs, half of which were intact from the study phase and half of which were rearranged. Participants were asked to give positive responses to intact pairs and negative responses to rearranged pairs. Results As is evident from the proportions shown in Table 1, no pseudoword effect was found. Stimulus type did affect accuracy. Hit rates were .57 for pseudoword pairs and .78 for word pairs. False-alarm rates were .35 for pseudoword pairs and .14 for word pairs. An analysis of variance on the number of positive responses found a significant effect of study status, F ð1; 19Þ ¼ 213:17, MSe ¼ 3:93, and a significant interaction between study status and stimulus type, F ð1; 19Þ ¼ 33:54, MSe ¼ 5:64. The main effect of stimulus type was not significant, F < 1:0. In short, a mirror-effect pattern emerged, with word pairs having higher hit rates and lower false-alarm rates than pseudoword pairs. This replicates previous claims that mirror effects may sometimes be found in associative recognition (Greene, 1996; Hockley, 1994). In this case, the mirror pattern found may well be a result of recollective processes. People could use each item of a test pair as a cue to retrieve the mate it had been paired with on the study list. Such a recollection strategy, when successful, would allow participants to give positive responses to intact pairs and negative responses to rearranged pairs. If it is assumed that this recollection strategy would be more effective with words than with pseudowords (as the overall difference in accuracy suggests), a mirror-effect pattern would result.
General discussion The pseudoword effect, the tendency to give more positive responses to pseudowords than to words, is a
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
reliable effect. The results presented here appear to be inconsistent with an account that attributes the effect to participantsÕ use of a lower criterion (or recoding process) for pseudowords as a result of their perceived difficulty. The pseudoword effect was found in forcedchoice recognition (Experiment 1). In addition, it was found even when the word advantage in accuracy was eliminated through repetition of pseudowords (Experiment 2) or reversed through directed forgetting of words (Experiment 3). The conclusion that the pseudoword effect does not reflect the use of different response criteria for words and pseudowords is consistent with recent evidence demonstrating that participants are ‘‘remarkably reluctant’’ to move response criteria back and forth on a single list (Morrell, Gaitan, & Wixted, 2002, p. 1107). An account that attributes the pseudoword effect to familiarity differences between the two stimulus classes received more support. New pseudowords are rated as more similar to a study list than are new words (Experiment 4). The pseudoword effect was reduced when a slower presentation rate was used (Experiment 5). The pseudoword effect was found only when recognition was accompanied by a Know experience, not a Remember experience (Experiment 6). The pseudoword effect was found on frequency judgment (Experiment 7), another task relying on familiarity, but not on associative recognition (Experiment 8), a task demanding recollection. The discrepancy-misattribution account The results reported here suggest that recognition differences between words and pseudowords reflect the greater familiarity of the latter than the former. The results of Experiment 4 suggest one possible explanation for this familiarity difference, namely, that pseudowords are more similar to the study list as a whole than are words. However, there is an important alternative explanation for these familiarity differences. Whittlesea and Williams (2000) see familiarity as being determined not only by the similarity of test stimuli to a list but also by the fluency with which test stimuli are processed (see Lindsay & Kelley, 1996; Whittlesea, Jacoby, & Girard, 1990). Fluent processing of a stimulus leads to a heightened sense of familiarity. Critical to the application of this account to the pseudoword effect is the assumption ‘‘that when people are aware of a reason why processing of a test stimulus might be especially fluent, they do not experience a feeling of familiarity’’ (Whittlesea & Williams, 2000, p. 548). Words are processed fluently. However, because participants expect to be able to perceive common words easily, they do not attribute this fluent processing to prior presentation on the study list. In contrast, participants are assumed to believe that nonwords should not be perceived fluently. Although these pseudowords are not processed as fluently as
265
words (when measured by pronunciation time), Whittlesea and Williams assume that they are processed more fluently than participants would expect them to be. Whittlesea and Williams argue that participants attribute this discrepancy between expected fluency and actual fluency for pseudowords to their presentation on the study list. Because this account attributes the pseudoword effect to differences in familiarity, it would naturally be consistent with the results of Experiment 1 and Experiments 5–8. Moreover, to the extent that attributions for discrepancies between expected and actual fluencies occur unconsciously, this account would also be consistent with the results of Experiments 2 and 3. Therefore, this approach is a possible way for explaining familiarity differences between words and pseudowords. Implications for Theories of the Mirror Effect The pseudoword effect represents an empirical challenge to the generalization that the mirror effect is a regularity of stimulus comparisons in recognition memory. The mirror effect was notable because it was difficult for standard global-matching models of recognition (e.g., Gillund & Shiffrin, 1984; Hintzman, 1988; Murdock, 1982b) to offer a principled explanation of this pattern. Indeed, models that have been termed ‘‘new generation global memory models’’ (Ratcliff & McKoon, 2000, p. 575) have been proposed specifically to account for mirror effects (McClelland & Chappell, 1999; Shiffrin & Steyvers, 1997) by the use of likelihood ratios in the decision process (see also Glanzer et al., 1993). Such an approach seems necessary if the mirror effect truly is an underlying regularity of recognition memory. However, the need for such an approach is less obvious if the mirror pattern is not an inevitable result of recognition judgments. The mirror-effect pattern, whereby a condition with higher hit rates will have lower false-alarm rates, may be found in between-subject or between-list manipulation of experimental conditions. This does not challenge any plausible theory of memory insofar as it is easy to account for such a pattern by suggesting that participants establish different criteria in different conditions; that is, if two lists differ markedly in strength and participants utilize different response criteria for the lists, it is possible for one list to have higher hit rates and lower falsealarm rates than the other (Stretch & Wixted, 1998). However, the mirror effect is more difficult to explain when different stimulus classes are compared on a single list. Given that participants seem disinclined to shift response criteria back and forth on a single list (Morrell et al., 2002; Stretch & Wixted, 1998) and that the mirror effect can be found even when word frequency is varied continuously (Glanzer, Kim, & Adams, 1998), a simple criterion-shift explanation would not be sufficient for
266
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
such experiments. However, if it turns out that relatively few stimulus comparisons regularly display the mirroreffect pattern in such designs, then it may be more profitable to explain the mirror effect not in terms of general principles of recognition memory but rather as a result of specific properties of the stimuli used in such experiments. An example of such an approach would be theories that attribute the mirror effect in word-frequency experiments to greater recollection of low-frequency targets than high-frequency targets and greater familiarity of high-frequency lures than low-frequency lures (Joordens & Hockley, 2000; Reder et al., 2000). Because such theories attribute mirror effects to properties of the specific stimuli typically used in experiments comparing high-frequency and low-frequency words, rather than to decision processes always used in recognition, they could explain why mirror effects are not found in experiments such as the ones reported here. In any event, theories of the mirror effect cannot merely offer accounts of its occurrence but must also offer principled explanations of times when it fails to be found, as with the pseudoword effect.
References Bain, J. D., & Humphreys, M. S. (1988). Relational context: Independent cues, meanings, and configurations. In G. Davis & D. M. Thomson (Eds.), Memory in context: Context in memory (pp. 97–137). London: Wiley. Cameron, T. E., & Hockley, W. E. (2000). The revelation effect for item and associative recognition: Familiarity versus recollection. Memory & Cognition, 28, 176–183. Clark, S. E., & Gronlund, S. D. (1996). Global matching models of recognition memory: How the models match the data. Psychonomic Bulletin & Review, 3, 37–60. Dewhurst, S. A., & Anderson, S. J. (1999). Effects of exact and category repetition in true and false recognition memory. Memory & Cognition, 27, 664–673. Gardiner, J. M., & Java, R. I. (1990). Recollective experience in word and nonword recognition. Memory & Cognition, 18, 23–30. Gardiner, J. M., & Richardson-Klavehn, A. (2000). Remembering and knowing. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 229–244). New York: Oxford University Press. Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall. Psychological Review, 91, 1– 67. Glanzer, M., & Adams, J. K. (1985). The mirror effect in recognition memory. Memory & Cognition, 13, 8–20. Glanzer, M., Adams, J. K., Iverson, G. J., & Kim, K. (1993). The regularities of recognition memory. Psychological Review, 100, 546–567. Glanzer, M., Kim, K., & Adams, J. K. (1998). Response distribution as an explanation of the mirror effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 633–644.
Greene, R. L. (1996). Mirror effect in order and associative information: Role of response strategies. Journal of Experimental Psychology: Learning, Memory and Cognition, 22, 687–695. Groninger, L. D. (1976). Predicting recognition during storage: The capacity of the memory system to evaluate itself. Bulletin of the Psychonomic Society, 7, 425–428. Hicks, J. L., & Marsh, R. L. (1998). A decrement-to-familiarity interpretation of the revelation effect from forced-choice tests of recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 1105– 1120. Hintzman, D. L. (1988). Judgments of frequency and recognition memory in a multiple-trace memory model. Psychological Review, 95, 528–551. Hintzman, D. L. (2001). Similarity, global matching, and judgments of frequency. Memory & Cognition, 29, 547–556. Hintzman, D. L., & Curran, T. (1997). Comparing retrieval dynamics in recognition memory and lexical decision. Journal of Experimental Psychology: General, 126, 228–247. Hintzman, D. L., & Hartry, A. L. (1990). Commensurability in memory for frequency. Journal of Memory and Language, 29, 501–523. Hirshman, E., Fisher, J., Henthorn, T., Arndt, J., & Passannante, A. (2002). Midazolam amnesia and dual-process models of the word-frequency mirror effect. Journal of Memory and Language, 47, 499–516. Hirshman, E., & Henzler, A. (1998). The role of decision processes in conscious recollection. Psychological Science, 9, 61–65. Hockley, W. E. (1994). Reflections of the mirror effect for item and associative recognition. Memory & Cognition, 22, 713– 722. Hockley, W. E., & Niewiadomski, M. W. (2001). Interrupting recognition memory: Tests of a criterion-change account of the revelation effect. Memory & Cognition, 29, 1176–1184. Joordens, S., & Hockley, W. E. (2000). Recollection and familiarity through the looking glass: When old does not mirror new. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 1534–1555. Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Lindsay, D. S., & Kelley, C. M. (1996). Creating illusions of familiarity in a cued recall remember/know paradigm. Journal of Memory and Language, 35, 197–211. Lockhart, R. S. (2000). Methods of memory research. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 45–57). New York: Oxford University Press. MacLeod, C. M. (1998). Directed forgetting. In J. M. Golding & C. M. MacLeod (Eds.), Intentional forgetting: Interdisciplinary approaches (pp. 1–57). Mahwah, NJ: Erlbaum. Macmillan, N. A. (1993). Signal detection theory as data analysis method and psychological decision model. In G. Keren & C. Lewis (Eds.), A handbook for data analysis in the behavioral sciences: Methodological issues (pp. 21–57). Hillsdale, NJ: Erlbaum. McClelland, J. L., & Chappell, M. (1999). Familiarity breeds differentiation: A Bayesian approach to the effects of experience in recognition memory. Psychological Review, 105, 724–760.
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267 Morrell, H. E. R., Gaitan, S., & Wixted, J. T. (2002). On the nature of the decision axis in signal-detection-based models of recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28, 1095–1110. Murdock, B. B. (1982a). Recognition memory. In C. R. Puff (Ed.), Handbook of research methods in human memory and cognition (pp. 1–26). New York: Academic Press. Murdock, B. B. (1982b). A theory for the storage and retrieval of item and associative information. Psychological Review, 104, 839–862. Rajaram, S., Hamilton, M., & Bolton, A. (2002). Distinguishing states of awareness from confidence during retrieval: Evidence from amnesia. Cognitive, Affective & Behavioral Neuroscience, 2, 227–235. Ratcliff, R., & McKoon, G. (2000). Memory models. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 571–581). New York: Oxford University Press. Reder, L. M., Nhouyvanisvong, A., Schunn, C. D., Ayers, M. S., Angstadt, P., & Hiraki, K. (2000). A mechanistic account of the mirror effect for word frequency: A computational model of remember-know judgments in a continuous recognition paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 294–320. Shiffrin, R. M., & Steyvers, M. (1997). A model for recognition memory: REM: Remembering effectively from memory. Psychonomic Bulletin & Review, 4, 145–160.
267
Stretch, V., & Wixted, J. T. (1998). On the difference between strength-based and frequency-based mirror effects in recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 1379–1396. Tulving, E. (1985). Memory and consciousness. Canadian Psychologist, 26, 1–12. Westerman, D. L. (2000). Recollection-based recognition eliminates the revelation effect in memory. Memory & Cognition, 28, 167–175. Whittlesea, B. W. A., Jacoby, L. L., & Girard, K. (1990). Illusions of immediate memory: Evidence for an attributional basis of familiarity and perceptual quality. Journal of Memory and Language, 29, 716–732. Whittlesea, B. W. A., & Williams, L. D. (2000). The source of feelings of familiarity: The discrepancy-attribution hypothesis. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 547–565. Whittlesea, B. W. A., & Williams, L. D. (2001). The discrepancy-attribution hypothesis: I. The heuristic basis for feelings of familiarity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 3–13. Wixted, J. T. (1992). Subjective memorability and the mirror effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 681–690. Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of 30 years of research. Journal of Memory and Language, 46, 441–517.
Journal of Memory and Language www.elsevier.com/locate/jml
Recognition memory for pseudowordsq Robert L. Greene* Department of Psychology, Case Western Reserve University, 10900 Euclid Avenue, Cleveland, OH 44106-7123, USA Received 30 June 2003; revision received 24 October 2003
Abstract Participants are more likely to give positive responses on a recognition test to pseudowords (pronounceable nonwords) than words. A series of experiments suggests that this difference reflects the greater overall familiarity of pseudowords than of words. Pseudowords receive higher ratings of similarity to a studied list than do words. Pseudowords receive more positive recognition responses than words even when pseudowords are remembered at least as well as words. This pseudoword effect is found on forced-choice recognition and on frequency judgments but not in associative recognition. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Familiarity; Frequency; Mirror effect; Pseudowords; Recognition; Similarity
The mirror effect (Glanzer & Adams, 1985) is the finding that, in recognition experiments involving two classes of stimuli, the class with the higher hit rate tends to have the lower false-alarm rate. This effect has been most often demonstrated through comparisons of lowfrequency and high-frequency words. The mirror effect is commonly portrayed as one of the regularities of recognition memory (Glanzer, Adams, Iverson, & Kim, 1993). Indeed, the mirror effect has been listed (along with the list-strength effect) as one of the two phenomena ‘‘that are at the heart of testing and evaluating the models’’ of recognition memory (Ratcliff & McKoon, 2000, p. 575). Despite the influence of the mirror effect, it is possible to find exceptions to this pattern. One of the best-documented exceptions involves comparisons of words with pseudowords (pronounceable nonwords). The standard finding here is that pseudowords (or, alternatively, very low-frequency words that can be assumed to be
q This research was presented at the 2002 meeting of the Psychonomic Society in Kansas City, MO. * Corresponding author. Fax: 1-216-368-4891. E-mail address: [email protected].
unknown by all participants) tend to have both higher hit rates and higher false-alarm rates than do words (Hintzman & Curran, 1997; Hockley & Niewiadomski, 2001; Whittlesea & Williams, 2000, 2001; Wixted, 1992). This finding, that overall positive-response rates are higher for pseudowords than for words, will be known here as the pseudoword effect. Several related explanations have been proposed for this pseudoword effect. This class of related explanations will be referred to here as the overcompensation account. According to this class of explanations, participants know that pseudowords may be less memorable than words. On a recognition test, participants try to equate for this memorability difference, either by adopting a lower criterion for pseudowords than for words (Hockley & Niewiadomski, 2001; Stretch & Wixted, 1998) or by rescaling the familiarity values of the pseudowords (Hintzman & Curran, 1997). However, participants systematically overcompensate, so that pseudowords become far more likely to receive positive responses than do words. Experiments 1–3 test this overcompensation explanation, while Experiments 4–8 address an alternative account, namely, that pseudowords receive more positive responses than words because they are higher in familiarity.
0749-596X/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jml.2003.12.001
260
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Experiment 1 One version (Hockley & Niewiadomski, 2001; Stretch & Wixted, 1998) of the overcompensation approach attributes the pseudoword effect to the use of a lower response criterion for pseudowords than for words. If this version were true, then one may not expect to find the pseudoword effect in forced-choice recognition. It is often assumed that no response criteria are needed on forced-choice tests (e.g., Hicks & Marsh, 1998; Lockhart, 2000; Macmillan, 1993; Murdock, 1982a). However, Wixted (1992, Experiment 1) reported results that suggest that a pseudoword effect may be found in forced-choice recognition; he found a tendency for subjects to select rare words more often on this test than words of higher frequency. Experiment 1 was an attempt to replicate this pattern using words and pseudowords. Method Participants Sixteen students from introductory-psychology classes participated to fulfill a course requirement. Materials In the experiments reported here, the materials used were the 60 pseudowords (regular nonwords) and 60 words used by Whittlesea and Williams (listed in Appendix A of their 2000 paper). A different random ordering of stimuli and random assignment of stimuli to experimental conditions was presented to each participant. Procedure Participants were tested individually. They were asked to study a list for an unspecified memory test. A list of 30 words and 30 pseudowords intermixed was shown one item at a time on a computer screen at a 1 s rate. Immediately after presentation of the list, a self-paced forced-choice recognition test was administered on the computer. This test consisted of 60 test pairs, each containing one word and one pseudoword. Participants were told that each test pair contained one item that had been presented on the study list and one that had not been. They were asked to select the studied item in each test pair and were informed that words and pseudowords were equally likely to be the correct response. Results The overall proportions of items given positive responses are shown in Table 1 for words and pseudowords for all experiments reported here. Unless otherwise noted, a significance level of .05 is used for all statistical tests.
Table 1 Proportions of positive responses given to words and pseudowords Experiment 1 2 3 5 6
7 8
(Forced-Choice) (Repeated pseudowords) (Directed forgetting of words) (1 s Presentation rate) (3 s Presentation rate) (Overall recognition) (Remember responses) (Know responses) (Guess responses) (Frequency judgment) (Associative recognition)
Words .38 .41 .29 .36 .44 .38 .21 .13 .04 .33 .46
Pseudowords .62 .62 .61 .61 .58 .60 .19 .33 .08 .49 .46
A pseudoword effect in forced-choice recognition was found, as all 16 participants selected the pseudoword as the old item more often than the word (p < :01 by sign test). As a result of this tendency, participants were more accurate on test pairs when the pseudoword was the correct answer (.85) than when the word was the correct choice (.62), F ð1; 15Þ ¼ 21:30, MSe ¼ 18:40.
Experiment 2 The standard overcompensation account of the pseudoword effect attributes the lower criterion used for pseudowords to participantsÕ beliefs that words are much more memorable than pseudowords. According to this approach, if participants were put in a situation where they know that words are at least as memorable as pseudowords, no such pseudoword effect should be found. In Experiment 2, this was accomplished by presenting all pseudowords twice and all words once. Method Participants Twenty students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants were told that they would be studying a mixed list of words and pseudowords (pronounceable nonwords) for an unspecified study list. They were told that the experimenter wanted to equate memory for the words and the pseudowords and that all words would be presented once and all pseudowords would be presented twice. They then saw the study list of 90 events (30 words presented once with 30 pseudowords presented twice) at a 1 s rate. The two occurrences of a repeated item were always separated by 10–15 intervening items.
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Immediately after completion of the study list, participants were reminded that all pseudowords were presented twice and all words once and were asked whether they thought this was effective in approximately equating memory for the stimulus classes; all participants agreed with this claim. A self-paced yes–no recognition test on 120 items, shown individually on a computer screen, was then given. Results Overall proportions of positive responses are shown in Table 1. Recognition accuracy was approximately equated for the two classes, with mean A0 values being .80 and .80 for words and pseudowords, F < 1:0. Despite this equating of overall accuracy and participantsÕ seeming acknowledgment of this fact, a pseudoword effect was still found, as is clear from the overall proportions of positive responses given in Table 1. Pseudowords exhibited higher hit rates (.82) and falsealarm rates (.42) than did words (.61 and. 21, respectively). An analysis of variance on numbers of positive responses was conducted using stimulus class (word/ pseudoword) and study status (old/new) as factors. This analysis found significant main effects of both stimulus class, F ð1; 19Þ ¼ 47:03, MSe ¼ 16:88, and study status, F ð1; 19Þ ¼ 229:62, MSe ¼ 12:75, but no significant interaction between them, F < 1:0.1
Experiment 3 Experiment 3 was a conceptual replication of Experiment 2 in which the pseudoword effect was examined in another situation where the typical memory advantage for words was eliminated. Here it was done by using the directed-forgetting procedure (see MacLeod, 1998, for a review). Before presentation of the list, participants were instructed to remember the pseudowords and to forget the words.
1
The logic of this research program has been to use a constant set of stimuli while varying the procedure across experiments. However, there is a need to establish some generality across stimuli. For this purpose, Experiment 2 was replicated using a set of 20 new participants from the same source as the other experiments and using an identical procedure to that of Experiment 2. A new set of stimuli was created, consisting of 30 words with frequencies between 10 and 40 occurrences per million in the Kucera and Francis (1967) corpus and 30 rare words that did not occur in the Kucera and Francis corpus but are included in the Oxford English Dictionary. A pseudoword effect was again obtained, with the hit rates being .80 and .65 for pseudowords and words respectively; the false-alarm rates were .40 and .22. Statistical analyses led to the same conclusions as Experiment 2.
261
Method Participants Twenty introductory-psychology students participated to fulfill a course requirement. Procedure Participants were told that they would be seeing an intermixed list of 30 words and 30 pseudowords at a 1 s rate and were told that they must try to remember the pseudowords and to ignore the words. To help them do this, the word FORGET was printed above each word as it was shown, and the word REMEMBER was shown above each pseudoword. Immediately after seeing the list, participants were asked whether they thought they remembered the pseudowords from the list better than the words, and all agreed with this assertion. They then received a 120-item yes–no recognition test. Results Overall proportions of positive responses are shown in Table 1. Recognition accuracy was better for the pseudowords (mean A0 ¼ :76) than for the words (mean A0 ¼ :69), F ð1; 19Þ ¼ 5:33, MSe ¼ 0:01. Despite this reversal in accuracy, a pseudoword effect was found, with pseudowords showing both higher hit rates (.77) and false-alarm rates (.44) than words (.39 and. 19, respectively). An analysis of variance on numbers of positive responses found both significant effects of stimulus type (word/pseudoword), F ð1; 19Þ ¼ 47:36, MSe ¼ 37:71, and study status (old/new), F ð1; 19Þ ¼ 148:88, MSe ¼ 8:38. There was also a significant interaction, F ð1; 19Þ ¼ 13:41, MSe ¼ 5:38, indicating that the difference between the stimulus types was greater on old items than on new items; this reflects the pseudoword advantage in accuracy. The results of the first three experiments are inconsistent with an overcompensation account of the pseudoword effect, which would assume that participants use a lower criterion (or rescaled familiarity values) for pseudowords because they believe them to be less familiar than words. The effect was found when a forced-choice test was used (Experiment 1) and when the word advantage in recognition accuracy was either eliminated (Experiment 2) or reversed (Experiment 3). An alternative could be that pseudowords truly tend to be more familiar than words (see Wixted, 1992, p. 689, for a similar suggestion). It is typically assumed that familiarity is determined by the similarity of a test stimulus to a study list (e.g., Gillund & Shiffrin, 1984; Hintzman, 1988; Murdock, 1982b; Ratcliff & McKoon, 2000). All words contain strong semantic features that make each one unique and distinctive. Although words may be studied more effectively than pseudowords, they
262
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
are less similar to other items on the study list, reducing their overall sense of familiarity. In contrast, pseudowords, while not completely devoid of meaningful associations, presumably have much less semantic content. Their similarity to the study list would be based largely on their structural features (e.g., letters and phonemes). Insofar as these structural features overlap greatly between items, an assessment of familiarity based chiefly upon structural features could yield higher values than one in which meaning plays a substantial role. The occurrence of the pseudoword effect may indeed depend upon structural overlap between pseudowords and words. When the pseudowords are constructed so that none of them resemble any common words in the English language, no pseudoword effect is found (Groninger, 1976; Whittlesea & Williams, 2000). Similarly, nonwords that consist merely of strings of consonants and therefore have little resemblance to English words do not show the pseudoword effect (Greene, 1996). Rather than relying on intuitions, it was important to collect evidence that pseudowords truly may seem more familiar (i.e., more similar to a study list) than words. In Experiment 4, such similarity was assessed directly.
Experiment 5 If the pseudoword effect truly results from greater familiarity of pseudowords than words, a manipulation that lessens participantsÕ reliance on familiarity should reduce the effect. One such manipulation should be presentation rate: as a presentation rate is slowed, participants may have a greater opportunity to perform the elaborative and associative processing that would allow them to base their later recognition decisions on recollection of the original episodes. Some support for the notion that slower presentation rates enhance recollection more than familiarity comes from studies employing the Remember/Know procedure, where study duration has been shown to have a greater effect on the probability that participants will claim that they remember an old item than on the probability that they will merely know it had been presented (Dewhurst & Anderson, 1999; Hirshman, Fisher, Henthorn, Arndt, & Passannante, 2002; Hirshman & Henzler, 1998). Method Participants Twenty-four students from introductory-psychology classes participated to fulfill a course requirement.
Experiment 4 Method Participants Sixteen students from introductory-psychology classes participated to fulfill a course requirement. Procedure All participants saw a 60-item study list (30 words and 30 pseudowords intermixed) in anticipation of an unspecified memory test. Immediately after presentation of the list, participants were given a similarity-rating test. They were shown 30 new words and 30 new pseudowords intermixed. They were told that none of these items had appeared on the study list. They were asked to rate on a seven-point scale (with seven being maximum) the similarity of each test stimulus to the study list. Participants did not request, and were not offered, information as to how they should conceive of similarity for these ratings. Results Fifteen of the 16 participants gave higher mean ratings to pseudowords than to words. There was a significant difference between the mean rating given to pseudowords (4.56) and the mean rating given to words (3.31), F ð1; 15Þ ¼ 125:00, MSe ¼ 0:10.
Procedure Presentation rate of the study list (1 vs. 3 s) was manipulated between participants. Immediately after presentation of the 60-item study list (30 words and 30 pseudowords intermixed), a yes–no recognition test on 120 items was given. Results Overall proportions of positive responses are shown in Table 1. For the 3-s presentation rate group, hit rates were .77 and .73 for pseudowords and words respectively; false-alarm rates were .39 and .15. For the 1-s group, hit rates were .81 and .61, and false-alarm rates were .41 and .12 for pseudowords and words. An analysis of variance on number of positive responses was conducted using the between-subject factor of presentation rate and the within-subject factors of stimulus type (word/pseudoword) and study status (old/ new). Stimulus type had a significant main effect, F ð1; 22Þ ¼ 68:10, MSe ¼ 11:57, as did study status, F ð1; 22Þ ¼ 298:56, MSe ¼ 15:34. Critically, there was a significant interaction between stimulus type and presentation rate, F ð1; 22Þ ¼ 4:80, MSe ¼ 11:57, indicating that the pseudoword effect was reduced with the slower presentation rate. There was also a significant interaction between stimulus type and study status, F ð1; 22Þ ¼ 16:94, MSe ¼ 7:31, indicating that the
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
pseudoword effect was greater on false alarms than on hits. No other interactions were significant.
Experiment 6 If the pseudoword effect reflects familiarity differences between stimuli, then one should expect to see differences between words and pseudowords in the subjective experiences of remembering. A common way of studying subjective experiences in recognition has been to ask participants to discriminate between two different states, a Remember state where the prior occurrence of an item could be specifically remembered and a Know state where the participant knows an item was on the list without recollecting the experience (Tulving, 1985). Gardiner and Java (1990, Experiment 2) reported that pseudowords received more Know responses than did words (see also Rajaram, Hamilton, & Bolton, 2002); that study is replicated here using the stimuli of Whittlesea and Williams (2000). Following the recommended procedure of Gardiner and RichardsonKlavehn (2000), all participants were asked to classify their positive recognition responses as Remember, Know, or Guess judgments. Method Participants Twenty-four students from introductory-psychology classes participated to fulfill a course requirement. Procedure The study list consisted of 30 words and 30 pseudowords intermixed shown at a 1 s rate. Participants were then given a 120-item yes–no recognition test. For each positive recognition response, participants were asked to classify their subjective recognition experience as a Remember experience, a Know experience, or a Guess (with instructions based on the sample instructions provided in Gardiner & Richardson-Klavehn, 2000). That is, if the participant felt that the recognition decision was accompanied by a recollection of the experience, then a Remember response would be made. If the item appeared familiar but no recollective experience occurred, then a Know response would be made. The participant was also allowed to classify responses as Guesses. Results Overall proportions of positive responses are shown in Table 1. The pseudoword effect was evident in overall recognition and in Know responses and Guess responses but not in Remember responses. In overall recognition, the hit rates were .76 for pseudowords and .58 for words;
263
the corresponding false-alarm rates were .43 and .18. In overall recognition, there were significant effects of study status (old/new), F ð1; 23Þ ¼ 242:32, MSe ¼ 12:17, and stimulus type, F ð1; 23Þ ¼ 86:54, MSe ¼ 11:87, as well as a significant interaction between these factors, F ð1; 23Þ ¼ 4:38, MSe ¼ 7:45. Participants were asked to classify recognition decisions based upon recollection as Remember responses. There was no pseudoword effect on these items. The proportion of old items that received Remember responses were .30 and .37 for pseudowords and words respectively; the corresponding false-alarm rates were .08 and .05. An analysis of variance was carried out on the number of Remember responses with the factors of study status and stimulus type. The main effect of study status was significant, F ð1; 23Þ ¼ 110:69, MSe ¼ 14:17. The interaction between study status and stimulus type was also significant, F ð1; 23Þ ¼ 16:31, MSe ¼ 3:13, reflecting the advantage for words over pseudowords in accuracy. However, the main effect of stimulus type was not significant, F ð1; 23Þ ¼ 1:19, MSe ¼ 3:51. Items that were familiar but were not recollected received Know responses. The proportion of old items receiving Know responses were .38 and .17 for pseudowords and words respectively; the corresponding falsealarm rates were .27 and .08. An analysis of variance on the number of items receiving Know responses found significant main effects of study status, F ð1; 23Þ ¼ 36:49, MSe ¼ 6:08, and stimulus type, F ð1; 23Þ ¼ 85:66, MSe ¼ 10:09. The interaction was not significant, F ð1; 23Þ ¼ 1:01, MSe ¼ 4:12.2 Participants were asked to make a Guess response only when they made a positive recognition response in the absence of both recollection and familiarity. As a result, this option was not used often. However, a pseudoword effect was found on Guess responses. The proportion of old items that received Guess responses were. 08 and. 05 for pseudowords and words respectively; the corresponding false-alarm rates were. 08 and. 04. There was a significant effect of stimulus type, F ð1; 23Þ ¼ 11:94, MSe ¼ 1:85. Neither the main effect of study status nor the interaction were significant (in both cases, F < 1:0). The presence of a pseudoword effect on Guess responses is consistent with the claim by Yonelinas (2002) that these responses are familiarity-based and involve a subjective experience similar to that found with Know responses.
2
Yonelinas (2002) advocated measuring familiarity as the probability of a Know response given that a Remember response was not given (i.e., K=½1 R). When this measure is used, the probability of a positive response based on familiarity was .18 for words and .42 for pseudowords. The pseudoword effect was still significant using this measure, F ð1; 23Þ ¼ 100:34, MSe ¼ 0:02.
264
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
Experiment 7 If the pseudoword effect reflects familiarity differences, then it should be found on tests other than recognition that also involve assessments of familiarity. One such task is frequency judgment (Hintzman, 1988, 2001; Hintzman & Hartry, 1990). In this experiment, participants received a study list of items presented once or twice. The frequency-judgment test required participants to remember how often a test stimulus had occurred on the list. Method Participants Sixteen students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants were asked to study a list for an unspecified memory test. The study list consisted of 180 presentations (60 items presented twice and 60 items presented once). A different random assignment of stimuli to frequencies was obtained for each participant, with the constraint that equal numbers of words and pseudowords must be shown at each frequency. Occurrences of a repeated item were separated by 10–15 presentations. Immediately after the study list, participants were given a test in which they were shown the 120 unique stimuli shown on the study list. They were asked to give each stimulus a response of 1 or 2 to indicate the number of times it had been shown on the study list. Results Proportions of positive responses (i.e., frequency judgments of 2) are shown in Table 1. A pseudoword effect was found, with participants giving responses of 2 to .60 of twice-presented pseudowords, .51 of twicepresented words, .38 of once-presented pseudowords, and .16 of once-presented words. The main effects of stimulus type, F ð1; 15Þ ¼ 18:90, MSe ¼ 16:44, frequency, F ð1; 15Þ ¼ 137:83, MSe ¼ 8:45, and the interaction, F ð1; 15Þ ¼ 62:02, MSe ¼ 5:55, were all significant.
Experiment 8 In associative recognition, participants study a list of pairs. On a subsequent test, they are asked to give positive responses to intact pairs (i.e., pairs containing two items that had been presented together on the study list) and negative responses to rearranged pairs. Associative recognition is often viewed as a test relying primarily on recollection, rather than familiarity (e.g., Bain & Humphreys, 1988; Cameron & Hockley, 2000; Clark &
Gronlund, 1996; Westerman, 2000). If this interpretation is true and if the pseudoword effect is due to the greater familiarity of pseudowords than words, then no pseudoword effect should be found in associative recognition. Method Participants Twenty students from introductory-psychology classes participated to fulfill a class requirement. Procedure Participants studied a list of 60 pairs, shown at a 2 s rate. Each pair consisted of either two words or two pseudowords. Assignment of stimuli to pairs was done randomly for each participant. Immediately after the study list, participants were given an associative recognition test. This test consisted of 60 pairs, half of which were intact from the study phase and half of which were rearranged. Participants were asked to give positive responses to intact pairs and negative responses to rearranged pairs. Results As is evident from the proportions shown in Table 1, no pseudoword effect was found. Stimulus type did affect accuracy. Hit rates were .57 for pseudoword pairs and .78 for word pairs. False-alarm rates were .35 for pseudoword pairs and .14 for word pairs. An analysis of variance on the number of positive responses found a significant effect of study status, F ð1; 19Þ ¼ 213:17, MSe ¼ 3:93, and a significant interaction between study status and stimulus type, F ð1; 19Þ ¼ 33:54, MSe ¼ 5:64. The main effect of stimulus type was not significant, F < 1:0. In short, a mirror-effect pattern emerged, with word pairs having higher hit rates and lower false-alarm rates than pseudoword pairs. This replicates previous claims that mirror effects may sometimes be found in associative recognition (Greene, 1996; Hockley, 1994). In this case, the mirror pattern found may well be a result of recollective processes. People could use each item of a test pair as a cue to retrieve the mate it had been paired with on the study list. Such a recollection strategy, when successful, would allow participants to give positive responses to intact pairs and negative responses to rearranged pairs. If it is assumed that this recollection strategy would be more effective with words than with pseudowords (as the overall difference in accuracy suggests), a mirror-effect pattern would result.
General discussion The pseudoword effect, the tendency to give more positive responses to pseudowords than to words, is a
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
reliable effect. The results presented here appear to be inconsistent with an account that attributes the effect to participantsÕ use of a lower criterion (or recoding process) for pseudowords as a result of their perceived difficulty. The pseudoword effect was found in forcedchoice recognition (Experiment 1). In addition, it was found even when the word advantage in accuracy was eliminated through repetition of pseudowords (Experiment 2) or reversed through directed forgetting of words (Experiment 3). The conclusion that the pseudoword effect does not reflect the use of different response criteria for words and pseudowords is consistent with recent evidence demonstrating that participants are ‘‘remarkably reluctant’’ to move response criteria back and forth on a single list (Morrell, Gaitan, & Wixted, 2002, p. 1107). An account that attributes the pseudoword effect to familiarity differences between the two stimulus classes received more support. New pseudowords are rated as more similar to a study list than are new words (Experiment 4). The pseudoword effect was reduced when a slower presentation rate was used (Experiment 5). The pseudoword effect was found only when recognition was accompanied by a Know experience, not a Remember experience (Experiment 6). The pseudoword effect was found on frequency judgment (Experiment 7), another task relying on familiarity, but not on associative recognition (Experiment 8), a task demanding recollection. The discrepancy-misattribution account The results reported here suggest that recognition differences between words and pseudowords reflect the greater familiarity of the latter than the former. The results of Experiment 4 suggest one possible explanation for this familiarity difference, namely, that pseudowords are more similar to the study list as a whole than are words. However, there is an important alternative explanation for these familiarity differences. Whittlesea and Williams (2000) see familiarity as being determined not only by the similarity of test stimuli to a list but also by the fluency with which test stimuli are processed (see Lindsay & Kelley, 1996; Whittlesea, Jacoby, & Girard, 1990). Fluent processing of a stimulus leads to a heightened sense of familiarity. Critical to the application of this account to the pseudoword effect is the assumption ‘‘that when people are aware of a reason why processing of a test stimulus might be especially fluent, they do not experience a feeling of familiarity’’ (Whittlesea & Williams, 2000, p. 548). Words are processed fluently. However, because participants expect to be able to perceive common words easily, they do not attribute this fluent processing to prior presentation on the study list. In contrast, participants are assumed to believe that nonwords should not be perceived fluently. Although these pseudowords are not processed as fluently as
265
words (when measured by pronunciation time), Whittlesea and Williams assume that they are processed more fluently than participants would expect them to be. Whittlesea and Williams argue that participants attribute this discrepancy between expected fluency and actual fluency for pseudowords to their presentation on the study list. Because this account attributes the pseudoword effect to differences in familiarity, it would naturally be consistent with the results of Experiment 1 and Experiments 5–8. Moreover, to the extent that attributions for discrepancies between expected and actual fluencies occur unconsciously, this account would also be consistent with the results of Experiments 2 and 3. Therefore, this approach is a possible way for explaining familiarity differences between words and pseudowords. Implications for Theories of the Mirror Effect The pseudoword effect represents an empirical challenge to the generalization that the mirror effect is a regularity of stimulus comparisons in recognition memory. The mirror effect was notable because it was difficult for standard global-matching models of recognition (e.g., Gillund & Shiffrin, 1984; Hintzman, 1988; Murdock, 1982b) to offer a principled explanation of this pattern. Indeed, models that have been termed ‘‘new generation global memory models’’ (Ratcliff & McKoon, 2000, p. 575) have been proposed specifically to account for mirror effects (McClelland & Chappell, 1999; Shiffrin & Steyvers, 1997) by the use of likelihood ratios in the decision process (see also Glanzer et al., 1993). Such an approach seems necessary if the mirror effect truly is an underlying regularity of recognition memory. However, the need for such an approach is less obvious if the mirror pattern is not an inevitable result of recognition judgments. The mirror-effect pattern, whereby a condition with higher hit rates will have lower false-alarm rates, may be found in between-subject or between-list manipulation of experimental conditions. This does not challenge any plausible theory of memory insofar as it is easy to account for such a pattern by suggesting that participants establish different criteria in different conditions; that is, if two lists differ markedly in strength and participants utilize different response criteria for the lists, it is possible for one list to have higher hit rates and lower falsealarm rates than the other (Stretch & Wixted, 1998). However, the mirror effect is more difficult to explain when different stimulus classes are compared on a single list. Given that participants seem disinclined to shift response criteria back and forth on a single list (Morrell et al., 2002; Stretch & Wixted, 1998) and that the mirror effect can be found even when word frequency is varied continuously (Glanzer, Kim, & Adams, 1998), a simple criterion-shift explanation would not be sufficient for
266
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267
such experiments. However, if it turns out that relatively few stimulus comparisons regularly display the mirroreffect pattern in such designs, then it may be more profitable to explain the mirror effect not in terms of general principles of recognition memory but rather as a result of specific properties of the stimuli used in such experiments. An example of such an approach would be theories that attribute the mirror effect in word-frequency experiments to greater recollection of low-frequency targets than high-frequency targets and greater familiarity of high-frequency lures than low-frequency lures (Joordens & Hockley, 2000; Reder et al., 2000). Because such theories attribute mirror effects to properties of the specific stimuli typically used in experiments comparing high-frequency and low-frequency words, rather than to decision processes always used in recognition, they could explain why mirror effects are not found in experiments such as the ones reported here. In any event, theories of the mirror effect cannot merely offer accounts of its occurrence but must also offer principled explanations of times when it fails to be found, as with the pseudoword effect.
References Bain, J. D., & Humphreys, M. S. (1988). Relational context: Independent cues, meanings, and configurations. In G. Davis & D. M. Thomson (Eds.), Memory in context: Context in memory (pp. 97–137). London: Wiley. Cameron, T. E., & Hockley, W. E. (2000). The revelation effect for item and associative recognition: Familiarity versus recollection. Memory & Cognition, 28, 176–183. Clark, S. E., & Gronlund, S. D. (1996). Global matching models of recognition memory: How the models match the data. Psychonomic Bulletin & Review, 3, 37–60. Dewhurst, S. A., & Anderson, S. J. (1999). Effects of exact and category repetition in true and false recognition memory. Memory & Cognition, 27, 664–673. Gardiner, J. M., & Java, R. I. (1990). Recollective experience in word and nonword recognition. Memory & Cognition, 18, 23–30. Gardiner, J. M., & Richardson-Klavehn, A. (2000). Remembering and knowing. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 229–244). New York: Oxford University Press. Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall. Psychological Review, 91, 1– 67. Glanzer, M., & Adams, J. K. (1985). The mirror effect in recognition memory. Memory & Cognition, 13, 8–20. Glanzer, M., Adams, J. K., Iverson, G. J., & Kim, K. (1993). The regularities of recognition memory. Psychological Review, 100, 546–567. Glanzer, M., Kim, K., & Adams, J. K. (1998). Response distribution as an explanation of the mirror effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 633–644.
Greene, R. L. (1996). Mirror effect in order and associative information: Role of response strategies. Journal of Experimental Psychology: Learning, Memory and Cognition, 22, 687–695. Groninger, L. D. (1976). Predicting recognition during storage: The capacity of the memory system to evaluate itself. Bulletin of the Psychonomic Society, 7, 425–428. Hicks, J. L., & Marsh, R. L. (1998). A decrement-to-familiarity interpretation of the revelation effect from forced-choice tests of recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 1105– 1120. Hintzman, D. L. (1988). Judgments of frequency and recognition memory in a multiple-trace memory model. Psychological Review, 95, 528–551. Hintzman, D. L. (2001). Similarity, global matching, and judgments of frequency. Memory & Cognition, 29, 547–556. Hintzman, D. L., & Curran, T. (1997). Comparing retrieval dynamics in recognition memory and lexical decision. Journal of Experimental Psychology: General, 126, 228–247. Hintzman, D. L., & Hartry, A. L. (1990). Commensurability in memory for frequency. Journal of Memory and Language, 29, 501–523. Hirshman, E., Fisher, J., Henthorn, T., Arndt, J., & Passannante, A. (2002). Midazolam amnesia and dual-process models of the word-frequency mirror effect. Journal of Memory and Language, 47, 499–516. Hirshman, E., & Henzler, A. (1998). The role of decision processes in conscious recollection. Psychological Science, 9, 61–65. Hockley, W. E. (1994). Reflections of the mirror effect for item and associative recognition. Memory & Cognition, 22, 713– 722. Hockley, W. E., & Niewiadomski, M. W. (2001). Interrupting recognition memory: Tests of a criterion-change account of the revelation effect. Memory & Cognition, 29, 1176–1184. Joordens, S., & Hockley, W. E. (2000). Recollection and familiarity through the looking glass: When old does not mirror new. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 1534–1555. Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Lindsay, D. S., & Kelley, C. M. (1996). Creating illusions of familiarity in a cued recall remember/know paradigm. Journal of Memory and Language, 35, 197–211. Lockhart, R. S. (2000). Methods of memory research. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 45–57). New York: Oxford University Press. MacLeod, C. M. (1998). Directed forgetting. In J. M. Golding & C. M. MacLeod (Eds.), Intentional forgetting: Interdisciplinary approaches (pp. 1–57). Mahwah, NJ: Erlbaum. Macmillan, N. A. (1993). Signal detection theory as data analysis method and psychological decision model. In G. Keren & C. Lewis (Eds.), A handbook for data analysis in the behavioral sciences: Methodological issues (pp. 21–57). Hillsdale, NJ: Erlbaum. McClelland, J. L., & Chappell, M. (1999). Familiarity breeds differentiation: A Bayesian approach to the effects of experience in recognition memory. Psychological Review, 105, 724–760.
R.L. Greene / Journal of Memory and Language 50 (2004) 259–267 Morrell, H. E. R., Gaitan, S., & Wixted, J. T. (2002). On the nature of the decision axis in signal-detection-based models of recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28, 1095–1110. Murdock, B. B. (1982a). Recognition memory. In C. R. Puff (Ed.), Handbook of research methods in human memory and cognition (pp. 1–26). New York: Academic Press. Murdock, B. B. (1982b). A theory for the storage and retrieval of item and associative information. Psychological Review, 104, 839–862. Rajaram, S., Hamilton, M., & Bolton, A. (2002). Distinguishing states of awareness from confidence during retrieval: Evidence from amnesia. Cognitive, Affective & Behavioral Neuroscience, 2, 227–235. Ratcliff, R., & McKoon, G. (2000). Memory models. In E. Tulving & F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 571–581). New York: Oxford University Press. Reder, L. M., Nhouyvanisvong, A., Schunn, C. D., Ayers, M. S., Angstadt, P., & Hiraki, K. (2000). A mechanistic account of the mirror effect for word frequency: A computational model of remember-know judgments in a continuous recognition paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 294–320. Shiffrin, R. M., & Steyvers, M. (1997). A model for recognition memory: REM: Remembering effectively from memory. Psychonomic Bulletin & Review, 4, 145–160.
267
Stretch, V., & Wixted, J. T. (1998). On the difference between strength-based and frequency-based mirror effects in recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 1379–1396. Tulving, E. (1985). Memory and consciousness. Canadian Psychologist, 26, 1–12. Westerman, D. L. (2000). Recollection-based recognition eliminates the revelation effect in memory. Memory & Cognition, 28, 167–175. Whittlesea, B. W. A., Jacoby, L. L., & Girard, K. (1990). Illusions of immediate memory: Evidence for an attributional basis of familiarity and perceptual quality. Journal of Memory and Language, 29, 716–732. Whittlesea, B. W. A., & Williams, L. D. (2000). The source of feelings of familiarity: The discrepancy-attribution hypothesis. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 547–565. Whittlesea, B. W. A., & Williams, L. D. (2001). The discrepancy-attribution hypothesis: I. The heuristic basis for feelings of familiarity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 3–13. Wixted, J. T. (1992). Subjective memorability and the mirror effect. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 681–690. Yonelinas, A. P. (2002). The nature of recollection and familiarity: A review of 30 years of research. Journal of Memory and Language, 46, 441–517.