Source monitoring and associative structure

Journal of Memory and Language 87 (2016) 144–156

Contents lists available at ScienceDirect

Journal of Memory and Language journal homepage: www.elsevier.com/locate/jml

Source monitoring and associative structure Francis S. Bellezza ⇑, Jennifer K. Elek, Ru Zhang Department of Psychology, Ohio University, United States

a r t i c l e

i n f o

Article history: Received 24 September 2014 revision received 18 September 2015

Keywords: Source monitoring Episodic memory Context information Event memory Mediated learning

a b s t r a c t Paired associates were used to study source memory. In three studies each word of the pairs was presented in one of the four locations of a two by two array. An event code explanation of memory representation, based on the hierarchical propositional network of Anderson and Bower (1974), was used to explain two seemingly paradoxical results: (a) Location identification of the cue word depended on successful target recall, and (b) source memory for the cue and target words was the same. Furthermore, the creation of an event code in memory can explain why the source locations of unrecalled target words were identified above chance level. This explanation seems preferable to a word code explanation in which source information for each item is attached to itself and also to the other item in the pair. It is suggested that the event code explanation may also play a role in accounting for source identification without recognition. Ó 2015 Elsevier Inc. All rights reserved.

Introduction It is as important to remember details of an event, such as its time or place, as it is to remember the event itself (Johnson, Hashtroudi, & Lindsay, 1993; Johnson & Raye, 1981, 1998). How source information is remembered has been the focus of intensive research. Moreover, the assessment of source-monitoring performance has typically occurred in conjunction with the testing of recognition memory. This is true in the studies of the neuroscience of source monitoring (Mitchell & Johnson, 2009) and in the development of mathematical models of source monitoring (Banks, 2000; Batchelder & Riefer, 1990; Glanzer, Hilford, & Kim, 2004; Rotello, Macmillan, & Reeder, 2004), as well as in multiple-item paradigms such as the associative recognition task (Achim & Lepage, 2003; Hockley, 2008), where pairs of items are presented followed by a recognition test of the complete pairs. Researchers have generally assumed source information ⇑ Corresponding author at: Department of Psychology, Ohio University, Athens, OH 45701, United States. E-mail address: [email protected] (F.S. Bellezza). http://dx.doi.org/10.1016/j.jml.2015.09.004 0749-596X/Ó 2015 Elsevier Inc. All rights reserved.

to be stored in memory at the item level. If an item is not recognized, then its contextual information cannot be remembered. Thus, source-identification tests have been accompanied by recognition tests with source identification assumed to be contingent on successful recognition. But source information in memory about real events is not typically accessed by another presentation of that event. By the predominant use of the recognition paradigm we may have unintentionally limited our knowledge of the learning and retention of source information. Because real events are typically assembled from contiguous but often unrelated features of an experience, source information must often be remembered when only some aspect of the event is referred to in a test query. For example, the sparse question ‘‘How was your vacation?” can result in a variety of information being remembered, including source and context information. Consequently, for the successful identification of source information the components of each event must have first been bound together in memory (Chalfonte & Johnson, 1996). Our results presented below suggest this to be the case. Paired associates learning is rarely used in source monitoring experiments. (For exceptions see Bröder, 2009;

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

Cook, Marsh, & Hicks, 2006.) In our preliminary studies using a paired-associates learning task, the two words of each presented pair occupied two locations in a 2  2 matrix, as shown in Fig. 1. The words used represented concrete objects, and participants were told to try to create a visual image combining the words. Later, when one of the items was presented as the cue word, the participants had to recall the other word, the target word. They also had to identify the original location of each word in the pair. Word pairs, of course, are not complex events; but the recall of an ensemble of information before making a source judgment is more difficult than simply recognizing a single word. Because of the prevailing belief that an individual word is the locus and effective cue for its source information in memory, we expected to find more source information remembered for cue words than for target words. We assumed that a cue word could elicit its source information during each recall test, but that a target word had to be recalled before its location could be identified. To our surprise, we found in preliminary studies that successful source identification of a cue word was greatly dependent on recall of its corresponding target word. Secondly, it appeared that levels of source monitoring performance for the cue word and the target word were identical. These two results have implications about how source information is stored in episodic memory. The phenomena of interest we describe here represent the less than expected source identification of the cue words of studied pairs compared to the source identification of the target words, which are much less available. In our studies every cue word was available at test, but only about half of the target words were, namely, only those that could be recalled. Thus, with regard to source identification there were approximately twice as many presented cue words present as test items as there were target words. When source information is identified better than expected There is another result reported in the research literature that may be related to those focused upon here. This phenomenon involves the greater than expected source identification of studied items that have not been recognized or recalled (e. g., Starns & Hicks, 2008; Starns, Hicks, Brown, & Martin, 2008). For example, using a paired-associates paradigm, Cook et al. (2006) found that

145

under certain conditions the sex of the voice presenting an item could later be identified, even when the test item itself was not recalled. Ball, DeWitt, Knight, and Hicks (2014), using extra-list cues semantically related to unrecognized test items, found that the rating task previously used to evaluate an item could be identified at a level greater than expected by guessing. Other authors have also demonstrated this phenomenon of source identification without recognition (e. g., Kurilla & Westerman, 2010). A common explanation of the nature of source memory is that the representation of the word in memory becomes associated with information indicating that the word was presented during the experiment. Various types of source information, such as time of occurrence, location, type font, or sex of speaker’s voice may become associated as information tags (Quillian, 1968) to the representation of the word. We refer to this as the word-code explanation. According to this view, a participant may be able to retrieve temporal information for an item from memory, but not be able to retrieve location information. That is, recognition of the word may occur but not the retrieval of other source information. Similarly, location information may be retrieved but not temporal information. In this latter situation the item may not be recognized but other source information may be retrieved. Some types of source information may be remembered and not others because in the word-code explanation of source memory each word is made up of features which may be selectively activated and used in associations (Anisfeld & Knapp, 1968). Hence, semantic components of the word provide context to be associated to cooccurring information. A similar mechanism generating associations rather than features has been labeled by Underwood (1965) as implicit associative responses. These types of generated context can act as a mediator for the source information. That is, different types of source information may become associated to different aspects of the generated context. Context information may be inadequate to support the recognition of an item but may be able to retrieve enough information from memory to identify source information associated with the item. For a further discussion of this explanation of source information without recognition or recall see Cook et al. (2006) and Ball et al. (Figs. 1–5 and 2014). Word code explanation

Fig. 1. Example of words in a pair and their source locations.

The word code explanation can be applied to the paired associates paradigm used here. One possible version of the word code explanation is diagrammed in Fig. 2. The Link a in memory associates the two words in the pair. Because participants do not know which word will be the test cue, during presentation participants may bind the source information of each word to both items of the pair. Link x1 binds source information about the location of the cue to the cue word, and Link y1 binds source information about the target word to the cue word. Similarly Links x2 and y2 bind source information about the two words to what will be the target word. Thus, regardless of which item is the cue, recall of the other item plus source information for both items can be retrieved. In this word code

146

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

Fig. 2. Diagram of a possible word code explanation for associating location information to words in pair. The associations originating at the cue and target words may originate from different features or meanings of the words. Parameter a is the probability of retrieving the target using the cue of the pair; parameter x1 is the probability of the cue retrieving the proposition connecting the cue with its location; parameter y1 is the probability of the cue retrieving the proposition connecting the target with its location; parameter x2 is the probability of the target retrieving the proposition connecting the cue with its location; parameter y2 is the probability of the target retrieving the proposition connecting the target with its location.

explanation all the source information is associated to memory representations of both items presented in the event. Event-code explanation We propose an explanation somewhat different from the word-code explanation given above. We label it the event code explanation to emphasize our assumption that source information becomes associated to a mental construct we call an event rather than to the representations of individual items making up that event. The event must be created in memory before source information can be attached to it. Anderson and Bower (1974) suggested that the associations relating list information with various types of context information form a hierarchical network. The nodes in this network represent interrelated propositions utilizing labeled associations. This formulation allows flexibility in the variety of events represented. To deal with our results, we use a simplified version of Anderson and Bower’s proposal (see also Anderson, 1976, 1983a, 1983b; Anderson, Bothell, Lebiere, & Matessa, 1998; Anderson & Bower, 1972, 1973).

As shown in Fig. 3, we assume that if a visual image can be constructed for the pair, then an event code is formed in memory that includes that image. The image is the event node in an associative hierarchy to which the cue and target words become associated, as represented by Links a and b, respectively. This image can be thought of a mediator (Montague, 1972; Paivio, 1969, 1971; Richardson, 1998) for a set of organized context in which the pair is embedded. Once the event code is created, the source information for each item in the pair can become associated to it by Links x and y. These associations depicted in Fig. 3 used for successful retrieval in different combinations can generate various kinds of performance. The first result in our preliminary data showed that successful source identification of the cue word was highly dependent on recall of the target word. With an event code explanation recall of the target indicates that the event code has been retrieved. There is then the possibility that the Links x to the source information of the cue can also be accessed. Our second result can also be explained: Because the participant does not know which item will be used as a cue, the availability of Links x and y should be comparable. Hence, the probability of identifying the source of each word will be approximately the same. Because of the event code’s associative structure, there is the possibility that the target word can be recalled but not its source information. There is also the possibility that source information can be identified without recalling the target word. That is, target source information and target recall information may co-vary but one is not necessary for the other. We suggest that the event code explanation is preferable to the word code explanation. There are a number of reasons for this. First, it seems that a representation of the event should be formed in memory before source information is associated to the event. That is, source information is associated to an event not to each of the multiple components making up the event. Second, if the source information is associated directly to the representations of the items making up the event, as depicted in Fig. 2, then the number of associations needed increases in a combinatorial manner as the number of component items increases. Third, one of our preliminary results was that source information for cue words is remembered no better than source information for the target words. This means that the cue word is as strongly associated to targetword source information as it is associated to its own cue-word source information. This does not correspond to the notion that information should be most strongly associated with stimuli in closest proximity to it. The differences between the two explanations are discussed in more detail below.

The three studies Fig. 3. Diagram of the event code model. Parameter a is the probability of retrieving the event code from the cue in the pair; parameter b is the probability of retrieving the target word in the pair from the event code; parameter x is the probability of retrieving the proposition giving the location of the cue word; parameter y is the probabilities of retrieving the proposition given the location of the target word.

The three studies described address various issues considered above. In all three studies we expected the two basic outcomes of (a) location identification of the cue word highly dependent on recall of the target word and (b) location identification of the cue and target word

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

almost identical. In addition, we also looked for identification of source location of the target words in the absence of recall of those target words. Study 1 was a straight-forward demonstration of these expected results. In Study 2, pairs consisted of both related and unrelated words were presented in the study list. The pairs with related words should be easier to learn because related words are more similar, and therefore associated, than unrelated words. Source identification, however, should be poorer for the related words. According to the event code explanation, there should be more associative interference when identifying the locations of similar cues. That is, in the event code explanation shown in Fig. 3 the instantiations Cue2 and Target2 will be more difficult to discriminate for related words than unrelated words (Osgood, 1949), even though the representations of Cue2 and Target2 result from being embedded in the composite image. A variety of predictions can be made from the word code hypothesis depending on precisely how Fig. 2 is interpreted. However, Ball et al. (2014) suggest that when recall does not occur source information should be better identified for the related items because the word used as a cue more easily accesses both sets of source information compared to when the items were unrelated. In Study 3, half the pairs were made up of unrelated words and the other half of single letter cues with word targets. It is more difficult to associate information to single letters than to concrete nouns, so recall should be poorer for letter cues than for word cues (Day & Bellezza, 1981). According to the event code hypothesis, information is associated to instantiations Cue2 of the cue and Target2 of the target word. Therefore, letter cues should exhibit poorer source discrimination than word cues. These expectations also follow from the word code explanation because the source information for both items will be more difficult to associate to letter than to word representations in memory. Method Materials for Study 1 Eight hundred and forty-five words were sampled from Toglia and Battig (1978) such that all words had mean ratings above 4.50 on seven-point scales of concreteness, imagery, familiarity, and meaningfulness. Ninety-six words were then randomly selected to construct 48 word pairs with the two words in each pair not obviously related. Examples are father – beach, bird – nurse, and heart – window. The words from each pair were randomly placed in different cells of a two-by-two grid as shown in Fig. 1. Each of the six possible spatial configurations of two words was utilized 8 times. Participants and procedure for Study 1 Forty-one undergraduate students participated for extra course credit and were tested with each participant seated before a computer station. Past research with our paired-associates procedure using 48 items indicated that

147

the mean proportion of correct recalls was about .50. This suggests that about twice as many cues words would be available as target words for source identification. Consequently, twice as many locations should be correctly identified for cues than for targets. Assuming that this is a medium effect size involving correlated means and setting power at .90, a sample size of 36 was appropriate (Faul, Erdfelder, Lang, & Buchner, 2007). Participants were informed that each pair would be later tested for recall and word location. The participants were instructed to try to form a mental image containing representations of both words when each pair was presented. The use of mental imagery in paired-associates learning has been shown to enhance performance (Paivio, 1971; Richardson, 1998). However, no instructions were provided regarding how the source locations should be remembered. Each word pair was presented for 3 s followed by 3 s of blank screen, with the entire list presented twice in identical order. Following presentation, participants solved anagrams for 3 min to eliminate any possible effects of shortterm memory. They then received a test booklet of 48 blank two-by-two grids with the cue word from each of the 48 pairs printed to the left of each grid. Each cue word was selected randomly from the two words in the presented pair. Participants recalled the appropriate target word by printing the cue word and the target word in their studied locations. Every test item had to be completed fully, even if the target word and the word locations were guessed. Materials for Study 2 Forty-eight related cue and target word pairs were sampled from Nelson, McEvoy, and Schreiber’s (1998) norms such that the words in each word pair scored a 5.85 or greater on the concreteness dimension. An additional 48 concrete words were sampled to serve as alternate cue words, with each cue word not obviously related to its target word. Hence, 48 word triads were made up of a target word, a related cue, and an unrelated cue. Two presentation lists of materials containing 48 word pairs each were created. In one list, half of these word pairs consisted of the related cue and target words, such as cheese – goat, tower – castle, and shoe – box, with the remaining half of the pairs containing the unrelated cue and target words. The second list contained the same target words but the relatedness of the cues from the 48 three-word sets were opposite of those in the first list. Examples of unrelated pairs were towel – goat, ticket – castle, and lamp – box. The spatial configuration corresponding to a target word to be recalled was identical for both lists. Participants and procedure in Study 2 Seventy-one undergraduate students completed Study 2 for course credit. A larger sample size was used compared to Study 1 to because two different types of pairs were tested using a within-subjects design. The procedure was that of Study 1 with participants randomly assigned to one of two forms of the experimental list. The words were

148

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

projected on a large screen in the testing room. After testing of the pairs, participants rated the relatedness of the two words in each presented pair on a 7-point rating scale, with 1 indicating that the words were completely unrelated and 7 indicating completely related. Materials for Study 3 Using of words from Study 1, forty-eight noun pairs were created with one word randomly chosen as the cue word. The target item of each pair was always a word, but half the cue items were replaced by a single letter of the alphabet. This was counterbalanced so that one list replaced the cue word with a letter in 24 of the 48 word pairs, and a second list replaced the cue word with a letter in the other 24 word pairs. In both lists half the pairs were word pairs and half the pairs were made up of a letter and a word. Examples of unrelated word pairs are pocket – bottle, perch – test, and tank – ocean. Examples of presented letter–word pairs are pocket – M, W – test, and tank – A. Participants and procedure in Study 3 Eighty-nine undergraduate students completed Study 3 for course credit. Approximately half were administered each list. After presentation of the pairs but before testing the participants were informed that all the targets to be recalled were words and that letters would appear only as cues. No relatedness ratings were collected, but the instructions and procedure were the same as in Study 2. Results In Study 1 all 41 participants recalled at least one target, a necessary condition for analyzing source identification conditional on recall, so all participants were retained. In Study 2 eight participants were eliminated because they did not recall at least one target word for both related and unrelated pairs. Similarly, eight participants were eliminated in Study 3 because they did not recall at least one target word for both word cues and letter cues. The final number of participants in Study 2 was 63 and in Study 3 was 81.

Table 1 Proportion of target words recalled and the proportion of spatial locations identified in the three studies. Condition

Target recall

Source identification Source of cue

Source of target

Study 1 Unrelated

.44

.54

.51

Study 2 Unrelated Related

.25 .50

.48 .53

.45 .53

Study 3 Letter cue Word cue

.39 .66

.54 .54

.53 .55

interest in the three studies, however, was sourcemonitoring performance. Source-monitoring performance Identifying locations of cues and targets As can be seen in Table 1, in all three studies there was almost no difference in the identification of the location of cue and target items. The largest difference was .03. An ANOVA was performed on the source identification performance of Studies 2 and of Study 3 with the two withinsubjects factors of pair condition and cue versus target. In Study 2 there was an effect of pair-relatedness, F (1, 62) = 19.60, p < .001, partial eta-squared = .24, with related pairs resulting in better source identification but with no effect of cue versus target. In Study 3 there was no effect of word cue versus letter cue on source identification nor of cue item versus target item. Source identification conditional on recall Figs. 4–6 show mean performance across participants in identifying the source of the items in each pair conditional on recall of the target word in each of the three studies. In Fig. 4 the CIs indicate that the only factor in Study 1 that was related to source identification was whether the target

0.8

Cue

Target

0.7

A recalled word was scored as correct even if it was the plural of the target word or a misspelling. As shown in Table 1, participants correctly recalled .44 of the target words in Study 1. In Study 2 target words for the pairs with semantically related words were better recalled (.50) than pairs with unrelated words (.25), where the 95% CI for the difference was [.21, .28]. A manipulation check of the materials showed that for the related pairs a mean relatedness rating of 5.85 was obtained compared to a mean relatedness rating of 2.06 for the unrelated pairs, with a 95% CI for the difference of [3.56, 4.02]. In Study 3, target words for word cues were better recalled (.66) than for letter cues (.39), with a 95% CI for the difference of [.23, .30], replicating the results of Day and Bellezza (1981). Of primary

Proportion Identified

Recall performance 0.6 0.5 0.4 0.3 0.2 0.1 0

No Recall

Recall

Recall Condition Fig. 4. Proportion of cues and targets in the recalled and unrecalled pairs whose source location was identified in Study 1.

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

Cue

following no recall and that source identification was superior for letter cues than for word cues but only following recall. A 3-way within-subjects ANOVA confirmed these two results. In all three studies source identification was equal for cues and targets and was positively correlated with target recall.

Target

1.0 0.9

Proporon Idenfied

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

No Recall Unrelated

Recall

No Recall

Recall

Relatedness Condion Related

Fig. 5. Proportion of cues and targets in the recalled and unrecalled pairs whose source location was identified for pairs consisted of unrelated words and for pairs consisted of related words in Study 2.

1.00 0.90 Cue

0.80

Proportion Identified

149

Target

0.70 0.60 0.50

Source identification performance for unrecalled targets For those pairs for which correct recall did not occur, participants had only the cue word to use for identifying both locations. Already noted was the result that source identification was the same for both cues and targets. But another interesting finding was that with no recall of the target word source identification for both items from the pair was significantly above the chance level of .25, as indicated by the CIs in Figs. 4–6. The conclusion that chance performance is equal .25 is based on the argument that there are four alternatives for guessing the location of the first word placed and three alternatives for the second word. Hence, there are twelve different combinations available for randomly placing the two words. For only one combination of the twelve possible are both locations correct. For two combinations of the twelve only the cue word is correctly placed, and for two combinations only the target word is correctly placed. For the other seven combinations both words are incorrectly placed. To correctly place the cue word by guessing, either the combination with both words correctly placed must be chosen or one of the two combinations where only the cue word is correctly placed. Thus, getting the cue location correct by chance is 3/12 = .25. The same argument can be made for correctly guessing the target location.

0.40

Correlation of cue and target source identification performance

0.30 0.20 0.10 0.00 No Recall

Recall

word cues

No Recall

Recall

letter cues

Cue Type Condition Fig. 6. Proportion of cues and targets in the recalled and unrecalled pairs whose source location was identified for pairs with words as cues and pairs with letters as cues in Study 3.

word was recalled. The factor of cue versus target word had no effect. An ANOVA with two within-subjects factors confirmed these results. As shown in Fig. 5, source identification in Study 2 was superior following recall, and the pairs of CI’s on the graph suggest that neither type of pair nor type of item in the pair had any effect on source identification once performance was made conditional on target recall. However, a 3-way within-subjects ANOVA indicated that when combining over word type and recall, source identification for unrelated items was .58 and superior to the .51 for related items, F(1, 62) = 16.25, p < .001, partial eta squared = .21. Based on inspection of the CIs in Fig. 6, source identification following recall in Study 3 was superior to that

Source location performance for the cue and target items was correlated for each participant using a phi correlation. This was done separately for the recalled and unrecalled items, as shown in Figs. 4–6. For those pairs for which recall occurred the mean phi correlations were .76 in Study 1, .85 and .85 for the related and unrelated conditions, respectively, of Study 2, and .77 and .76 for the word cue and letter cue conditions, respectively, of Study 3. The corresponding five phi correlations for the pairs in which no recall occurred were .24, .30, .44, .37 and .37. Because the largest 95% CI for these 10 mean correlations involved adding and subtracting a value of .12 from the value, there was no overlap between the phi values of the recalled and not recalled items. In our learning paradigm both words of the pair could not appear in the same location, so placement of one word in a location constrains the choice for the second word. Therefore, if participants simply guess the locations of the two words, a phi correlation of .11 is expected. Nevertheless, all five values of phi for the no-recall response category were significantly above the chance value of .11 using the largest 95% CI of .24. Thus, there is some common factor or factors that elevate the level of source identification of both the cue word and the target word of a pair

150

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

about chance level, even when recall of the target does not occur. An alternative way to investigate the independence of source identification for cue and targets is to compute conditional probabilities. If the target word was recalled, then each word is the pair was available as a prompt for identifying its location. The large values of phi reported above suggest that when the location of the cue word was identified, there was a high probability that the location of the target was identified. That is, the value of P(T|C) was large, where C represents the event of cue location identification and T represents the event of target location identification. If the target word was not recalled, however, only the cue word was available so it is important to determine to what degree target location was identified without target recall. The values of P(T|C) and P(C|T) when recall did not occur are shown in Table 2. If cue location was identified, then the value of P(T|C) based guessing is .33. Similarly, if target location was identified, the value P(C| T) by guessing is also .33. As can be seen in Table 2, the values of both P(C|T) and P(T|C) are similar in value and greater than .33 in all conditions of the studies. Conditional probabilities are related to the phi correlation coefficient in a specific manner. The quantity P(T| C) P(T|C) represents the DP measure (e. g., Bröder, 2009). However, when P(T) = P(C), then this difference also represents phi; that is, P(T|C) P(T|C) = DP = phi. See Falk and Well (1997) for a general discussion of correlation as a probability. In each of the ten conditions analyzed, the probability of correctly identifying the source of the cue, denoted P(C), was observed to be almost identical to the probability of correctly identifying the source of the target, denoted P(T). From this result a number of relationships can be deduced, as shown in Appendix A. The phi correlation was equal to both the differences P(C|T) P(C|T) and P(T|C) P(T|C) in each of the ten conditions of the three studies. By computing these conditional probabilities we found that source memory in the five recall and five norecall conditions were nearly identical for the cues and targets, which shows that assuming P(T|C) = P(C|T) and assuming P(C|T) = P(T|C) is justified. If the cue location and target location were influencing each other differently, then we would expect that P(T|C) – P(C|T). Furthermore, such an inequality also implies P(T) –

Table 2 Proportion of target locations identified given cue location identified, P(T| C), and proportion of cue locations identified given target location identified, P(C|T), when recall of the target word did not occur. Condition

P(T|C)

P(C|T)

Study 1 Unrelated

.48 (.04)

.52 (.04)

Study 2 Unrelated Related

.53 (.03) .48 (.04)

.58 (.03) .53 (.04)

Study 3 Letter cue Word cue

.61 (.03) .59 (.04)

.65 (.03) .61 (.04)

Note. Values in parentheses represent standard errors. All proportions are above the chance values of .33.

P(C), which is contrary to our results. Instead, we get conditional probabilities in these two cases that are the same. This result suggests that the locations are not cuing each other, but are both being affected by some other factor or factors (Starns & Hicks, 2005, 2008). Source identification reversals In Appendix B are shown the frequencies of locations chosen for the two items in each pair with regard to whether the item occupied its own correct location, denoted C or T, the location of the other word in the pair, denoted (C) or (T), or one of the two unused locations, denoted C and T. If participants are guessing both locations, there are 7 possible ways to guess the locations of the cue and target given that both are incorrect. One of these is the reversal error, denoted (C)(T), with an expected relative frequency of 1/7 or .14. However, in the five conditions of Studies 1–3 shown in Appendix B in which the target word was recalled, these relative frequencies of the (C)(T) responses were .50, .50, .54, .51, and .48, respectively. For the five conditions for which the target word was not recalled, the relative frequencies were .19, .18, .20, .17, and .22. All of these values are above the chance value of .14. In Study 2 location reversal errors were slightly smaller for unrelated words, .50, than for related words, .54, when target recall occurred, and were .18 and .20, respectively, with no target recall. Related words can be considered more similar, and thus more confusable, than unrelated words. These proportions are based on aggregated data with each participant usually having a small number of location errors from unrelated and related pairs in which both cue and target words are placed in incorrect locations. Using the frequency data displayed for Study 2 in Appendix B, the proportion of targets recalled for unrelated words was .25 and the proportion of targets recalled for related words was .50. Related and unrelated words, however, also differed in level of recall. Because of the poor target recall for unrelated pairs, only 35 out of 71 participants had at least one unrelated item where two location errors were made when recall of the target occurred. When an analysis was performed based on the mean proportions of these individual participants, the relative frequencies of location reversals given that the target was recalled was .49 for unrelated pairs and .51 for unrelated pairs. The difference was not statistically significant. When the analysis was performed on the data where the target was not recalled, the relative frequencies of location reversals were .17 for unrelated pairs and .25 for related pairs, compared to the respective values of .18 and .20 found in the aggregated data. This difference was statistically significant, t(69) = 1.99, p < .025, indicating that the reversal errors occurred more frequently for related than unrelated pairs when recall did not occur. Discussion The three studies described here reliably demonstrated the two related phenomena found in our preliminary

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

investigations: (a) the probability of identifying the source location of the cue word from a pair depended on the recall of the target word, and (b) source monitoring performance for the cue and target word was virtually identical. These results were surprising because cue words were always available at test, whereas target words were recalled for fewer than half the pairs. Two other effects from previous research literature were replicated. Source identification of unrecalled target words was above chance level, as reported in previous research literature (Ball et al., 2014; Cook et al., 2006). Also, source identification performance for cues and targets was positively correlated as indicated by the values of the phi correlations. This was true even when the target word was unrecalled. Effects of type of pair on source identification The experimental manipulations used in Studies 2 and 3 produced some unexpected results. The analysis conditional on recall supported our expectations in Study 2 but not in Study 3. In Study 2 related and unrelated pairs were used. Not surprisingly, more target items were recalled for related pairs than for unrelated pairs. Without being made conditional on recall performance, source identification was better for related words (.53) than unrelated words (.48), as shown in Table 1. However, when made conditional on recall, source identification was better for unrelated words (.58) than for related words (.51). This result is opposite of the result shown in Table 1. The reason for this discrepancy is that source identification was larger for recalled targets than for unrecalled targets. But, more targets were recalled for the related pairs. So, unless made conditional on recall, performance for related pairs appears to be better, as shown in Table 1, where related pairs resulted in better recall which in turn produced better source identification. The superiority of unrelated pairs compared to related pairs on the source identification measure was a main effect and thus was true for recalled and unrecalled items and for both cue and target items. The result that source identification of unrecalled targets was no different for related versus unrelated pairs is contrary to Ball et al. (2014, p. 1282) who suggested that in the related pairs the cue and target would share features and the cue word would thus be better able to access context generated by the target word which, in turn, would be associated with location information regarding the target word. This can be seen in Fig. 2 by considering that the features shared by the related words enabled access to location information that may have be associated to those features from either word. The event code explanation, as diagrammed in Fig. 3, makes a greater distinction between item encodings that are the basis of recall versus encodings that are created for source identification. The results of Study 3, in which word cues were compared to letter cues, are difficult to explain. For recall of the target words the word cues were more effective than letter cues, replicating Day and Bellezza (1981). As shown in Table 1, however, there was no overall difference in source identification for pairs with letter cues versus pairs with word cues. Because, in general, single letters are less

151

easily associated to new information than words, we expected that source information would be better for word cues. Contrary to expectations, when the target word was recalled, identification of the location of both cue and target was better when a letter was the cue than when a word was the cue. When target recall did not occur, there was no difference in identifying the locations of the two types of cues. When source identification performance was made conditional on recall, a significant interaction was found in Study 3. Contrary to expectations, not word-cue, but letter-cue superiority occurred but only for recalled pairs. One possible explanation of the result is that the pairs in Study 3 consisted of a total of 24 letters but 72 words. Not only because of their smaller total number but also because of their smaller number among the recalled pairs, the letters may have been more distinctive than the words (Nairne, 2002; Schmidt, 1991), thereby resulting in better source discrimination for the letter–word pairs. Distinctiveness as an organizational factor may have been operating to enhance source identification for the letter cues (Bower, 1970). Furthermore, global matching models suggest that items forming small categories based on their number of shared features are more easily retrieved (Clark & Gronlund, 1996; Hicks & Starns, 2006). The correlation of source memory for targets with source memory for cues When the target word could not be recalled, the level of source identification performance for both cues and targets was about .36 across the five conditions compared to about .75 when the target word could be recalled. The values of phi correlations indicated that identification of one component of the pair was highly predictive of identification of the other component, but only if the target was recalled. If the target was not recalled, predictability was less but still above chance level. This result of dependence of source memory on recall is similar to results of previous researchers (Ball et al., 2014; Bröder, 2009; Cook et al., 2006; Starns et al., 2008). Similarly, identification of source without recognition has been found in previous studies (Kurilla & Westerman, 2010; Starns & Hicks, 2008; Starns et al., 2008; Vogt & Bröder, 2007), but we performed no recognition tests, so the degree to which the unrecalled targets in our three studies would be recognized is unknown. In a word code explanation, as shown in Fig. 2, each word in the pair is associated with the source information of both words. Even if the link connecting the cue to the target, Link a, is not available the link of the cue to the location information of the target, Link y1, may be. If the target is recalled, then location information can be accessed from both cue and the target words, using Links y1 and y2, respectively. An event code explanation is somewhat different. Even without target recall the event code can sometimes be available. As shown in Fig. 3, Link b may fail, but Links x and y may function. Dependency between target recall and source identification is created by the presence or absence of Link a, which cannot be observed but sometimes can be inferred. For both explanations, the correlation between source identification for the cue and the

152

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

target does not imply that location information for one word is acting as a cue for the location information of the other (Starns & Hicks, 2005, 2008; Vogt & Bröder, 2007). Because the 48 pairs consist of 96 items, each location is associated with 24 items in the list. The degree of associative interference, or fanning (Anderson, 1983b), makes difficult the retrieval of one location of a pair item from the location of the other pair item. Evidence was found that participants were sometimes reversing the locations of the cues and targets after identifying the locations of the two items. This was especially true when the target word was recalled. This outcome can be described using either a word code explanation, as shown in Fig. 2, or an event code explanation, as shown in Fig. 3. In either model information identifying two of the four locations can be identified without the participant knowing whether each contained the target word or cue word. If both locations are identified without the participant knowing the item either location contained, then the reversals of the kind shown in Appendix B can result. Whether the data values can be accounted for precisely will depend on a formal model being developed and applied to these data. There was also evidence that location reversals were more frequent for pairs made up of related words compared to pairs containing unrelated words, as would be expected. The effects reached conventional levels of statistical significance only for pairs for which recall of the target word did not occur. Multidimensional source attributes Positive correlations in source monitoring have been previously reported in the literature using pairs of words with source information for each item. Bröder (2009) presented two words from the same taxonomic category either adjacent in a list or separated by seven other words. Also, each word was presented either at the top or bottom of a screen. In a test of free recall of the words the screen location of each word recalled had to be identified. When two category words were presented adjacently, the correct identification of the location of one word in the pair was highly predictive of the location identification of the other word. This was not true for words from pairs not presented adjacently. Because each word in the pair belonged to a common category, the participants, perhaps, became aware of the common category for adjacent words and used it as the basis of for relating the two words, but this awareness decreased as the two words appeared further apart in the list. In addition, there have been a number of studies in which a recognition procedure has been used where each individual item had values on two different source dimensions. For example, Meiser and Bröder (2002) varied the font size of visually presented words as well as their location on a computer screen. They found a positive relation between identifying the font of the presented word and identifying its location. But this occurred only with remember recognition responses but not with know responses for which source identification measures for font and location were independent. Vogt and Bröder (2007) and Meiser

(2014) also found a positive covariance when identifying different types of source information attached to the same presented items. From these results and those of our Studies 1–3 the conclusion can be drawn that performance on multiple source monitoring dimensions will covary in a positive way. This is true whether one item is used, as in recognition, or two items in a pair, as in paired-associates learning. When single words are presented along with two dimensions of source information for later recognition, a word code rather than an event code, may possibly act as a mediator for both sources of location information. Identification of the source values on both dimensions may be contingent on successful recognition, with each attribute associated with the representation of the word in memory, but not directly associated with one another (Starns & Hicks, 2008). However, we discuss below the possibility and, perhaps, the benefit of event codes acting as mediators for source information in the recognition paradigm.

General discussion We proposed an event code explanation, as opposed to a word code explanation, to account for the following source-memory results using a paired-associates learning task: that location identification of the cue word was strongly dependent on recall of the target word, that identification of the location of the cue word and target word was nearly identical, that source identification of the cue and target words occurred above chance level, even when recall of the target word did not occur, and that performance identifying the location of the cue and of target word was positively correlated across items regardless of whether recall of the target word occurred. As described above, a word code explanation can also explain these results, but we believe that the event code explanation is preferable for the following reasons: First, a representation of an event should be formed in memory before source information is associated to that event. Events are integral representations in memory and are remembered in a unitized manner. That is, source information is associated to events not to the components making up the event. This occurs in the event code explanation but not in the word explanation. This last assumption implies that participants who are unable to assemble an event in memory out of its diverse components will be less able retain source information about that event. In confirmation of this, source information was less available when a target item could not be recalled. Second, if an event is made up of more than two components, then source information should not be duplicated among the various components making up the event. However, if word codes were used, this would have to happen. Third, if the source information is associated directly to the representations of the items making up the event, as depicted in Fig. 2, then the number of associations needed increases in a combinatorial manner as the number of these items increases. Fourth, according to a word cue explanation, it seems reasonable that a single item in a pair should be the focus of attention when source information

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

regarding that item is processed. That is, one word is being cognitively processed to a greater degree than the other word because an individual word rather than the entire event is being attended to. Consequently, source information should be strongly associated to that item but more weakly associated to its paired item. Source information for the cue word should later be identified better in the presence of the cue word than source information for the target word. This is a consequence of the principle of encoding specificity (Thomson & Tulving, 1970; Tulving & Thomson, 1973). The role of encoding specificity in source-monitoring paradigms The event code is a central hub in a hierarchy of four associations represented by a link to each of the words in the pair and to two propositions encoding the location of each word. The order in which the hierarchy is formed is important. First, the event code is formed associating the two words, and then the location information is attached to it. Thus, the cognitive processes and context occurring during the two learning stages are different. The location information is not associated to representations of single words. Thus, if a participant was unable to assemble an event in memory out of two seemingly unrelated words, then he or she cannot store source information about them, unless a different learning strategy is used and the items are studied individually. Therefore, when explaining word recall and source memory in the paired associates paradigm, there are two instantiations, that is, encodings, of each word in the pair. When a pair of words is first presented, the participant reads them separately but because of instructions quickly tries to relate them to one another to form a composite image. The first encoding process involves considering the meaning of each word thereby creating memory representations Cue1 and Target1, as indicated in Fig. 3. A key assumption of the event code explanation is that only after a representation of the event is created, by interassociating the two words of the pair, is source information associated with the representation of the event. The event code is then completed by producing representations Cue2 and Target2 and making them part of the image. The representations Cue2 and Target2 are then associated to their respective location information. The encodings Cue1 and Cue2 are not the same. The encoding specificity principle suggests that representations Cue1 and Cue2 are not identical in meaning. The encodings are not trans-situational and, also, there is no automatic access from encodings to the more general representation of the word in semantic memory (Tulving & Thomson, 1973, p. 365). It is also true that Target1 and Target2 are not identical in meaning. When the pair is first presented, encoding processes create Cue1, the mediating event code, and Target1. The context of the encoding processes includes the experimental setting before and during the formation of the image. After the image has been formed it becomes a prominent part of the context of the processes linking the location information to Cue2 and Target2. Because of the principle of encoding specificity,

153

the encoding processes engaged in at encoding must be reproduced at test for the item to be accessed in memory. Cue1 is used to retrieve the event code mediator which retrieves Target1. The event code mediator also contains Cue2 and Target2 which are then used to retrieve the location information. Thus, the encoding processes are reproduced to enable retrieval (Thomson & Tulving, 1970; Tulving & Thomson, 1973; cf. Anderson et al., 1998). The factors which influence the formation and retrieval of the event code may not be the same factors that influence the encoding and retrieval of the source information. This was shown in Study 2 where similarity of the words in the pair facilitated the target word recall, representing the first stage of learning, but interfered with source identification, representing the second stage. The event code explanation for source memory in recognition We have cited studies in which source information is identified though a word has not been recognized (Kurilla & Westerman, 2010; Starns & Hicks, 2008) or a studied word not recalled (Cook et al., 2006). However, these two situations involving failure of recognition and failure of recall are not necessarily equivalent. Tulving and Thomson (1973) give the example of presenting the pair glue – CHAIR and later finding participants would sometimes fail to recognize CHAIR if presented alone but recall CHAIR when provided with the cue glue. The principle of encoding specificity (Thomson & Tulving, 1970; Tulving & Thomson, 1973) operates in either paradigm. We have suggested that the imagery of an event code supports recall from a cue, whereas in recognition a word code mechanism may suffice, and learning context may reside in word-generated associations when using the word code. However, the question may be asked whether an event code is also necessary when learning source information in a recognition paradigm. In recognition tasks the nature of any event code formed is not apparent. Participants are not typically instructed in a learning strategy. Yet, the phenomenon of source identification in the absence of recognition may be dependent on event codes being previously formed, even if participants were not explicitly instructed to do so. This possibility is supported by the fact that source identification of unrecognized items depends on strong encoding procedures involving semantic processing (Ball et al., 2014; Cook et al., 2006; Meiser, 2014; Starns et al., 2008), and seems to occur only if learning context is recollected (Kurilla & Westerman, 2010). What kind of instructions could be provided in recognition learning to create those event codes necessary for the phenomenon of source identification to occur without recognition taking place? We suggest that an event code for a single presented item can result from an elaboration based on semantic information (Fisher & Craik, 1980; Montague, 1972), such as using the presented word in a sentence. To explain the formation of an event code in recognition, Fig. 3 could be modified to include the one presented word with a Link a to the event code produced by elaboration but with no Link b. Furthermore, rather than two links to location information, there would be one link to location information and one link to temporal information

154

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

(Anderson & Bower, 1974), where access to the latter would provide the basis for a recognition decision. As in the paired-associate paradigm, in recognition there would not be perfect concordance between retrieval of the temporal information and retrieval of the location information. If the event code was not formed or was not accessible at test, then neither the location nor the temporal information would be available. If the event code was accessible, then one, both, or neither of the two pieces of source information, temporal and location, could be retrieved. Each type of information must be separately cued by the event code. Starns and Hicks (2008) have shown that when different contextual dimensions (color and location) were bound to item information, they were not directly linked to each other. This conclusion is consistent with both the word code and event code explanations as diagrammed in Figs. 2 and 3. Goals, strategies, and awareness Whether an event code or a word code mediates learning depends, in part, on learning strategies, in turn, influenced by learning instructions. Participants in our studies were given a learning goal and a mediation strategy for achieving that goal; namely to create an event, a visual image, from the two words with location information being of less importance. We assume that they were aware of their success in this reaching goal and could verbalize some of the contents of their visual image, although they were not asked to do so. Performance would change if instructions emphasized only remembering the location of each word with no emphasis on associating the two words. Without specific instructions each participant could use one of many different learning strategies, or, perhaps, no strategy at all, other than passive attention. How an individual creates an event in memory depends upon his or her expectations and goals. People typically try to find meaningful relations among the multiple components of an experienced event; that is, people try to comprehend events. We assume that mediated learning reflects what typically occurs when creating events in memory outside the laboratory. Events are usually created using pre-existing information in memory about naturally occurring objects and people. In general, expectations and goals can activate structures in memory, sometimes referred to as chunks, schemas, scripts, frameworks, and so on, whose purpose is to organize information to resemble familiar past events (Anderson, 1983a, 1983b). These mental structures formed from previously experienced events provide a framework for the event code. Even events formed from streams of behavior can be organized by a process Newtson (1973) has labeled unitization. In our three studies two consequential experimental manipulations were used. First, participants were instructed in a visual-mediation learning strategy. Second, when the pairs were presented for study, participants did not know which of the two words would be later used as the cue. The mediation instructions were used so that participants would form an event code. Without mediation instructions, which increase the likelihood of associating the two words, participants would be more likely to form

to separate word codes for each pair; that is, be less able to interrelate the two words in the pair. Under these conditions source identification should be greater for the cue word than for the target word because no event code would exist to enable location identification of the target word if it was not recalled. If participants know which word in each pair will be the cue word but nevertheless received mediated learning instructions, then performance may not be very different from that reported here. This is because when using concrete words visual imagery produces bidirectional associations (Paivio, 1971, pp. 276–285). If high-imagery materials are not used, however, creating an event code may become difficult. In this case location of the target item in the pair may only be identified if the target item is recalled. Without instructions for mediated learning knowing which item will be the cue may also have an effect on which learning strategy a participant chooses. For example, some participants may decide to focus on learning the location of the target item, based on the assumption that when presented the cue at test they will automatically remember its location. Or, some participants may focus on learning the location of the cue because they know it is certain to be tested. In addition, strategies may also be used at the time of memory retrieval. If a participant remembers a location associated with the pair but is unsure of whether it is for the cue or target word, the strategy used could be always to assign the location to the cue word. The cue word is salient because it is always present at test, but the target word is present only if it is recalled. This strategy would introduce a bias in location assignment. At this point we do not know what strategies, if any, participants use in source learning and retrieval. Limitations of the recognition paradigm Our use of a paired-associates paradigm for studying memory for source information resulted from dissatisfaction with the predominant use of the recognition paradigm for this purpose. There are three problems when using the recognition paradigm for studying source memory of an event. One is that only a single item typically comprises the experimental event, which does not reflect the more natural situation where a number of diverse, but contiguous, perceptual components must be organized in memory to represent an event. It is not often that a real event occurs that is already comprised of a completely integrated unit in memory. But this is what typically occurs in the recognition paradigm. Second, the memory test in the recognition paradigm requires a complete re-presentation of the event. These are not the circumstances under which we usually access source information in memory. Third, the required response is typically an old versus new judgment, resulting in the possibility of guessing and response bias. Many correct judgments could be the result of guessing. As mentioned above, researchers have built neuroscience accounts and mathematical models of source memory from data based almost exclusively on recognition data. We have suggested two possible explanations for the recall and source monitoring results we have obtained

155

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

here. Each of these explanations can be realized with a variety of formal models. Quantitative analysis may be necessary to make predictions with enough precision to determine if a particular model fits the data and to determine if parameter values exhibit values consistent with the memory processes proposed. Using a paired-associates paradigm with components that cannot be easily guessed seems like only a small step away from using the recognition paradigm, but, as we try to demonstrate here, the use of the paradigm raises new and interesting questions about the storage and retrieval of source information. These include, but are not limited to, Why is source memory for the event cue relatively poor? and Why is source memory for the unrecalled target relatively good?

Appendix A Proof that phi equals DP when the marginal distributions are identical in a 2  2 contingency table. Let the event C be the identification success of the location of the cue of the pair and let event C be the identification failure of the cue. Similarly, event T is the location identification of the target and T is its identification failure. The 2  2 frequency table is denoted as below: Event T

Event T

Marginal total

Event C Event C

n11 n21

n12 n22

n1 n2

Marginal total

n1

n2

n

If P(C) = P(T), then n1/n = n1/n and n1 = n1. Also, n2 = n2. In addition, 1/P(C) = 1/P(T), so P(C&T)/P(C) = P(C&T)/P(T). P(T|C)P(C)/P(C) = P(C|T)P(T)/P(T), so P(T|C) = P(C|T). Similarly, 1 P(C) = 1 P(T), so P(C) = P(T). Also, P(C&T)/P(C) = P(C&T)/P(T). Therefore, P(C|T) = P(T|C). phi = (n11 n22 n12 n21)/sqrt(n1 n2 n1 n2) = (n11 n22 n12 n21)/n1 n2. phi = (n11 n22)/n1 n2. (n12 n21)/n1 n2 = P(C|T)P(C|T) P(C|T)P(C|T). phi = P(C|T)[1 P(C|T)] P(C|T)[1 P(C|T)] = P(C|T) P(C|T)P(C|T) P(C|T) + P(C|T)P(C|T). phi = P(C|T) P(C|T) = DP, where also phi = P(T|C) –[1 P(C|T)] = P(T|C) [1 P[T|C)] = P(T|C) P(T|C). Because the assumption P(T) = P(C) holds for pairs when targets are recalled and for pairs when targets are not recalled, then phi = P(C|T) P(C|T) = P(T|C) P(T|C) is true for both types of pairs.

Appendix B Condition

Frequencies of item response locations for related and unrelated pairs in Studies 1–3. Response category CT

CT

CT

(C)T

C(T)

Study 1, unrelated words in pairs Correct recall 603 Incorrect recall 195

38 221

44 165

21 166

20 121

Study 2, unrelated words in pairs Correct recall 288 Incorrect recall 236

23 173

11 147

6 152

Study 2, related words in pairs Correct recall 543 Incorrect recall 119

19 124

27 114

Study 3, letter cue, word target Correct recall 597 Incorrect recall 245

36 173

Study 3, word cue, word target Correct recall 771 Incorrect recall 133

63 100

CT

(C)(T)

Total

45 145

86 98

– 1968

9 140

14 181

29 103

– 1512

26 110

17 94

31 119

86 83

– 1512

32 147

13 180

13 172

24 156

51 105

– 1944

80 90

57 99

68 89

62 84

172 76

– 1944

Note: C = cue in correct location, T = target in correct location, (C) = cue in target location, (T) = target in cue location, C = cue in location other than cue or target location, T = target in location other than cue or target location.

156

F.S. Bellezza et al. / Journal of Memory and Language 87 (2016) 144–156

References Achim, A. M., & Lepage, M. (2003). Is associative-recognition more impaired than item recognition memory in Schizophrenia? A metaanalysis. Brain and Cognition, 53, 121–124. http://dx.doi.org/10.1016/ S0278-2626(03)00092-7. Anderson, J. R. (1976). Language, memory, and thought. Hillsdale, NJ: Erlbaum. Anderson, J. R. (1983a). A spreading activation theory of memory. Journal of Verbal Learning and Verbal Behavior, 22, 261–295. http://dx.doi.org/ 10.1016/S0022-5371(83)90201-3. Anderson, J. R. (1983b). Retrieval of information from long-term memory. Science, 220, 25–30. http://dx.doi.org/10.1126/science.6828877. Anderson, J. R., Bothell, D., Lebiere, C., & Matessa, M. (1998). An integrated theory of list memory. Journal of Memory and Language, 38, 341–380. Anderson, J. R., & Bower, G. H. (1972). Recognition and retrieval processes in free recall. Psychological Review, 79, 97–123. http://dx.doi.org/ 10.1037/h0033773. Anderson, J. R., & Bower, G. H. (1973). Human associative memory. Washington, DC: Winston. Anderson, J. R., & Bower, G. H. (1974). A propositional theory of recognition memory. Memory & Cognition, 2, 406–412. http://dx.doi. org/10.3758/BF03196896. Anisfeld, M., & Knapp, M. (1968). Association, synonymity, and directionality in false recognition. Journal of Experimental Psychology, 77, 171–179. http://dx.doi.org/10.1037/h0025782. Ball, B. H., DeWitt, M. R., Knight, J. B., & Hicks, J. L. (2014). Encoding and retrieval processes involved in the access of source information in the absence of item memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40, 1271–1286. http://dx.doi.org/10.1037/ a0037204. Banks, W. P. (2000). Recognition and source memory as multivariate decision processes. Psychological Science, 11, 267–273. http://dx.doi. org/10.1111/1467-9280.00254. Batchelder, W. H., & Riefer, D. M. (1990). Multinomial processing models of source monitoring. Psychological Review, 97, 548–564. http://dx.doi. org/10.1037/0033-295X.97.4.548. Bower, G. H. (1970). Organizational factors in memory. Cognitive Psychology, 1, 18–46. http://dx.doi.org/10.1016/0010-0285(70)90003-4. Bröder, A. (2009). Semantically clustered words are stored with integrated context: Validating a measurement model for source memory, storage, and retrieval in free recall. Zeitschrift für Psychologie, 217, 136–148. http://dx.doi.org/10.1027/0044-3409.217.3.136. Chalfonte, B. L., & Johnson, M. K. (1996). Feature memory and binding in young and older adults. Memory & Cognition, 24, 403–416. http://dx. doi.org/10.3758/BF03200930. 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. http://dx.doi.org/10.3758/BF03210740. Cook, G. I., Marsh, R. L., & Hicks, J. L. (2006). Source memory in the absence of successful cued recall. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 828–835. http://dx.doi.org/10.3758/ BF03195921. Day, J. C., & Bellezza, F. S. (1981). The alphabet mnemonic and visualimagery mediation. Paper presented at the meeting of the Psychonomic Society, Philadelphia, PA, November. Falk, R., & Well, A. D. (1997). Many faces of the correlation coefficient. Journal of Statistics Education, 5. Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G⁄Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175–191. http://dx.doi.org/10.3758/BF03193146. Fisher, R. P., & Craik, F. I. M. (1980). The effects of elaboration on recognition memory. Memory & Cognition, 8, 400–404. http://dx.doi. org/10.3758/BF03211136. Glanzer, M., Hilford, A., & Kim, K. (2004). Six regularities of source recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 1176–1195. http://dx.doi.org/10.1037/ 0278-7393.30.6.1176. Hicks, J. L., & Starns, J. J. (2006). Remembering source evidence from associatively related items: Explanations from a global matching model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 1164–1173. http://dx.doi.org/10.1037/0278-7393.32.5. 1164. Hockley, W. E. (2008). The picture superiority effect in associative recognition. Memory & Cognition, 36, 1351–1359. http://dx.doi.org/ 10.3758/MC.36.7.1351. Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin, 114, 3–28. http://dx.doi.org/10.1037/h0028139.

Johnson, M. K., & Raye, C. L. (1981). Reality monitoring. Psychological Review, 88, 67–85. http://dx.doi.org/10.1037/0033-295X.88.1.67. Johnson, M. K., & Raye, C. L. (1998). False memories and confabulation. Trends in Cognitive Sciences, 4, 137–145. http://dx.doi.org/10.1016/ S1364-6613(98)01152-8. Kurilla, B. P., & Westerman, D. L. (2010). Source memory for unidentified stimuli. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36, 398–410. http://dx.doi.org/10.1037/a0018279. Meiser, T. (2014). Analyzing stochastic dependence of cognitive processes in multidimensional source recognition. Experimental Psychology, 61, 402–415. http://dx.doi.org/10.1027/1618-3169/a000261. Meiser, T., & Bröder, A. (2002). Memory for multidimensional source information. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28, 116–137. http://dx.doi.org/10.1037/0278-7393.28.1.116. Mitchell, K. J., & Johnson, M. K. (2009). Source monitoring 15 years later: What have we learned from fMRI about the neural mechanisms of source memory? Psychological Bulletin, 135, 638–677. http://dx.doi. org/10.1037/a0015849. Montague, W. E. (1972). Elaborative strategies in verbal learning and memory. In G. H. Bower (Ed.). The psychology of learning and motivation: Advances in research and theory (Vol. 6). New York: Academic Press. Nairne, J. S. (2002). The myth of the encoding-retrieval match. Memory, 10, 389–395. http://dx.doi.org/10.1080/09658210244000216. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). The University of South Florida word association, rhyme, and word fragment norms. . Newtson, D. (1973). Attribution and the unit of perception of ongoing behavior. Journal of Personality and Social Psychology, 28, 28–38. http://dx.doi.org/10.1037/h0035584. Osgood, C. E. (1949). The similarity paradox in human learning: A resolution. Psychological Review, 56, 132–143. http://dx.doi.org/ 10.1037/h0057488. Paivio, A. (1969). Mental imagery in associative learning and memory. Psychological Review, 76, 241–263. http://dx.doi.org/10.1037/ h0027272. Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart, and Winston. Quillian, M. R. (1968). Semantic memory. In M. Minsky (Ed.), Semantic information processing (pp. 227–270). Cambridge, MA: The MIT Press. Richardson, J. T. E. (1998). The availability and effectiveness of reported mediators in associative learning: A historical review and an experimental investigation. Psychonomic Bulletin & Review, 5, 597–614. http://dx.doi.org/10.3758/BF03208837. Rotello, C. M., Macmillan, N. A., & Reeder, J. A. (2004). Sum-difference theory of remembering and knowing: A two-dimensional signaldetection model. Psychological Review, 111, 588–616. http://dx.doi. org/10.1037/0033-295X.111.3.588. Schmidt, S. R. (1991). Can we have a distinctive theory of memory? Memory & Cognition, 19, 523–542. http://dx.doi.org/10.3758/ BF03197149. Starns, J. J., & Hicks, J. L. (2005). Source dimensions are retrieved independently in multidimensional monitoring tasks. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 1213–1220. http://dx.doi.org/10.1037/0278-7393.31.6.1213. Starns, J. J., & Hicks, J. L. (2008). Context attributes in memory are bound to item information, but not to one another. Psychonomic Bulletin & Review, 15, 309–314. http://dx.doi.org/10.3758/PBR.15.2.309. Starns, J. J., Hicks, J. L., Brown, N. L., & Martin, B. A. (2008). Source memory for unrecognized items: Predictions from multivariate signal detection theory. Memory & Cognition, 16, 1–8. http://dx.doi.org/ 10.3758/MC.36.1.1. Thomson, D. M., & Tulving, E. (1970). Associative encoding and retrieval: Weak and strong cues. Journal of Experimental Psychology, 86, 255–262. http://dx.doi.org/10.1037/h0029997. Toglia, M. P., & Battig, W. F. (1978). Handbook of semantic word norms. Hillsdale, NJ: Lawrence Erlbaum Associates. http://dx.doi.org/10. 1037/0033-295X.111.3.588. Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80, 352–373. http://dx.doi.org/10.1037/h0020071. Underwood, B. J. (1965). False recognition produced by implicit verbal responses. Journal of Experimental Psychology, 70, 122–129. http://dx. doi.org/10.1037/h0022014. Vogt, V., & Bröder, A. (2007). Independent retrieval of source dimensions: An extension of results by Starns and Hicks (2005) and a comment on the ACSIM measure. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 443–450. http://dx.doi.org/10.1037/02787393.33.2.443.