Intralist distractors and recall direction: Constraints on models of memory for serial order.

The serial-order version of the theory of distributed associative memory (TODAM; S. Lewandowsky and B. B. Murdock [see PA, Vol 76:14457]) predicts that disuption of forward serial recall should leave backward recall largely unaffected. This article reports 4 experiments in which the effects of an intralist distractor task were compared for forward and backward serial recall. Regardless of whether Ss could anticipate recall direction at study, the distractor task was found to disrupt forward but not backward recall. Although the existence of that dissociation had been predicted by TODAM, the theory was unable to provide a quantitative account of the data. Instead the authors provide a retrieval-based account within the framework of temporal distinctiveness theory. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Memory for serial order revisited.

The critiques by D. J. Mewhort et al (see record 1994-36081-001) and J. S. Nairne and I. Neath (see record 1994-36082-001) identified at least 6 potentially serious problems with S. Lewandowsky and B. B. Murdock's (see record 1989-14457-001) Theory of Distributed Associative Memory (TODAM) model of memory for serial order. It is shown that the flaws attributed to the memory component of TODAM are less serious than claimed, whereas the problems attributed to the response selection stage necessitated a process implementation of the previously unspecified deblurring mechanism. The deblurring process, implemented by a dynamic autoassociative network, is shown to handle most of the problems identified by the critics without imperiling TODAM's ability to handle basic serial position data. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

TODAM and the list-strength and list-length effects: Comment on Murdock and Kahana (1993a).

B. B. Murdock and M. J. Kahana (see record 1993-32216-001) presented a continuous memory version of the theory of distributed associative memory (TODAM) model. They claimed that this model predicts list-strength and list-length findings, including those reported by R. Ratcliff (1990) and K. Murnane and R. M. Shiffrin (1991). This model is quite similar to one discussed by Shiffrin et al (1990), who rejected the model on the basis of its inability to predict both an absent or negative list-strength effect (when strength is varied by repetitions) and a present list-length effect. This comment elaborates the earlier discussion and demonstrates that the version of TODAM proposed by Murdock and Kahana indeed fails for this reason. This is shown first for a somewhat simplified version of the model for which derivations are obvious and then in a simulation of the complete version using the parameter values suggested by Murdock and Kahana. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

On serial recall: A critique of chaining in the theory of distributed associative memory.

Criticizes 3 aspects of the S. Lewandowsky and B. B. Murdock (see record 1989-14457-001) model for serial recall based on the theory of distributed associative memory. First, recall is based, in the model, on a chain of associations. Once an error has been made, the chain is broken. To lengthen the chain, a tacit response (a facsimile vector) is used as a retrieval probe. It is shown that a facsimile vector is not an effective retrieval cue. Second, the model is biased against items at the beginning of a list, and, because of the bias, response alternatives must be limited in an artificial way. Third, the model implies that recognition failure of recallable items is a routine characteristic of memory-span tasks. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Item and associative recognition with precuing and postcuing.

M. Duncan & B. B. Murdock (2000) compared precued and postcued item recognition and serial recall showing precued-postcued differences for item recognition but not for serial recall. Precuing and postcuing refer to 2 conditions in which the instructions as to the type of recall test following the presentation of short lists of items is given before or after the list presentation. This methodology was extended here to a paired-associate task. In 2 experiments, short lists of paired associates were presented followed by single-item, old-new, or intact-rearranged pair recognition tests; test type was precued or postcued. A fast or slow presentation rate was used to discourage or encourage mediators. TODAM2 (a theory of distributed associative memory) predicts that there should be little or no cuing differences regardless of whether subjects use mediators to remember the pairs. As predicted the recognition data were essentially identical for the precued and postcued conditions. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Concept and measurement of primary memory.

The recency effect in free recall remains relatively constant over experimental conditions that have large and systematic effects on overall level of recall. The effect is commonly attributed to a distinct memory system, known as primary memory. 3 views of primary memory are discussed: the 1st conceptualizes primary memory as a limited capacity store from which information is transferred to a more permanent store, the 2nd equates primary memory with consciousness, are the 3rd sees primary memory as a limited capacity retrieval system. Methods for measuring primary memory have been suggested by N. C. Waugh and D. A. Norman (1965), B. Raymond (1969), B. B. Murdock (1967), and E. Tulving (1968, 1970). These methods are critically evaluated, and their relative merits are assessed with a combination of rational and empirical arguments. (63 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

List-strength and list-length effects: Reply to Shiffrin, Ratcliff, Murnane, and Nobel (1993).

R. M. Shiffrin et al (see record 1994-24299-001) claimed that the theory of distributed associative memory (TODAM) is unable simultaneously to predict an absent (or negative) list-strength effect (LSE) and a positive list-length effect (LLE). However, Shiffrin et al failed to distinguish between situations in which lag (i.e., number of items intervening between study and test) is controlled and situations in which it is not. The authors stand by their previous conclusion: TODAM can explain why there is little or no LSE when at the same time there is an LLE when the LLE is studied under the standard conditions. It is argued that there are no published studies where lag has been controlled. However, this simplified version of TODAM cannot explain an LLE when a scoring window is used. Whether such a result would be inconsistent with a more complete version of TODAM remains to be seen. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

TODAM2: A model for the storage and retrieval of item, associative, and serial-order information.

Presents an extended version of the convolution–correlation memory model TODAM (theory of distributed associative memory) that not only eliminates some of the inadequacies of previous versions but also provides a unified treatment of item, associative, and serial-order information. The chunking model extended the basic convolution–correlation formalism by using multiple convolutions, n-grams (multiple autoassociations of sums of item vectors), and chunks (sums of n-grams) to account for chunking and serial organization. TODAM2 extends the chunking model by including rn-grams (reduced n-grams), labels, and "lebals" (the involution or mirror image of a label) to provide a general model for episodic memory. For paired associates, it is assumed that Ss store only labeled n-grams and lebaled rn-grams. It is shown that the model is broadly consistent with a number of major empirical paired-associate and serial-order effects. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Forced-choice associative recognition: Implications for global-memory models.

Associative recognition requires Ss to discriminate intact from rearranged test pairs. In a 3-alternative forced-choice procedure, an intact test pair is tested against 2 rearranged distractors which may overlap, by sharing a common word in each test pair alternative (OLAP), or may not share words (NOLAP). With the exception of B. B. Murdock's (see record 1983-04936-001) theory of distributed associative memory (TODAM), current global matching models predict that forced-choice performance will be better for OLAP than for NOLAP test trials. TODAM can predict either an OLAP advantage or no difference between OLAP and NOLAP test conditions. The performance of the models is produced by fundamental statistical properties, and with the exception of TODAM, the OLAP advantage cannot be eliminated by varying parameters. Results of 3 experiments, however, show a NOLAP advantage. The implications of these results for global matching models and the relationship between recall and recognition are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

"Levels of processing, encoding specificity, elaboration, and CHARM": Correction to Eich.

Reports an error in "Levels of processing, encoding specificity, elaboration, and CHARM" by Janet M. Eich (Psychological Review, 1985[Jan], Vol 92[1], 1-38). Equation 5 on page 11 was incorrect. The correct equation is provided in the erratum. (The following abstract of the original article appeared in record 1985-13697-001.) A model of cued recall called CHARM (composite holographic associative recall model) is applied to several issues that have been investigated within the depth-of-processing framework. It is shown that, given some straightforward, empirically testable assumptions about the representations of the to-be-remembered items themselves, CHARM can account for the main effect of depth of processing, the problem of the negatives, encoding–specificity interactions, and both facilitative and inhibitory effects of elaboration. The CHARM model is extended to encompass some depth-of-processing effects found in recognition memory. The highly interactive associative, storage, and retrieval mechanisms in the CHARM model are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Correction to Pike.

Reports an error in "Comparison of convolution and matrix distributed memory systems for associative recall and recognition" by Ray Pike (Psychological Review, 1984[Jul], Vol 91[3], 281-294). In this article, there were two erroneous sentences, one on page 284 and one on page 285. The sentences are corrected in the erratum. (The following abstract of the original article appeared in record 1984-27853-001.) Compares 2 closely related distributed memory models in terms of plausibility; arithmetic simplicity; economy of storage space; and ability to account for associative, similarity, and order data in recall and recognition. It is argued that the storage-retrieval system brought about by the convolution-correlation concept outlined by M. A. Eich (see record 1983-04922-001) and B. B. Murdock (see record 1983-04936-001) is neurally implausible, necessitates more complex analyses, and is less economical in storage space than is the matrix memory concept described by J. A. Anderson et al (see record 1978-22353-001). It is shown that the matrix model can easily account for associative symmetry-asymmetry data and for item similarity effects. Means and variances of operating strength for various recall and recognition situations, modeled by the matrix system, are presented, and it is shown how signal-to-noise ratios can be derived. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Levels of processing, encoding specificity, elaboration, and CHARM.

[Correction Notice: An erratum for this article was reported in Vol 92(4) of Psychological Review (see record 2008-10981-001). Equation 5 on page 11 was incorrect. The correct equation is given in the erratum.] A model of cued recall called CHARM (composite holographic associative recall model) is applied to several issues that have been investigated within the depth-of-processing framework. It is shown that, given some straightforward, empirically testable assumptions about the representations of the to-be-remembered items themselves, CHARM can account for the main effect of depth of processing, the problem of the negatives, encoding–specificity interactions, and both facilitative and inhibitory effects of elaboration. The CHARM model is extended to encompass some depth-of-processing effects found in recognition memory. The highly interactive associative, storage, and retrieval mechanisms in the CHARM model are discussed. (90 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Mathematical and connectionist models of human memory: a comparison

Recent convolution-based models of human memory (e.g. Lewandowsky & Murdock, 1989), have accounted for a wide range of data. However such models require the relevant mathematical operations to be provided to the network. Connectionist models, in contrast, have generally addressed different data, and not all architectures are appropriate for modelling single-trial learning. Furthermore, they tend to exhibit catastrophic interference in multiple list learning. In this paper we compare the ability of convolution-based models and DARNET (Developmental Associative Recall NETwork), to account for human memory data. DARNET is a connectionist approach to human memory in which the system gradually learns to associate vectors, in one trial, into a memory trace vector. Either of the vectors can than be retrieved. It is shown that the new associative mechanism can be used to account for a wide range of relevant experimental data as successfully as can convolution-based models with the same higher-level architectures. Limitations of the models are also addressed.     

Predicting free recalls.

This article reports some calculations on free-recall data from B. Murdock and J. Metcalfe (1978), with vocal rehearsal during the presentation of a list. Given the sequence of vocalizations, with the stimuli inserted in their proper places, it is possible to predict the subsequent sequence of recalls--the predictions taking the form of a probability distribution over all possible such sequences. The predictions are parameter free. This article looks at (a) the accuracy of those predictions and how that accuracy can be validated, (b) the principles on which the predictive algorithm is based, and (c) the limits to predictability. Some implications for the understanding of free recall are spelled out. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Comparison of convolution and matrix distributed memory systems for associative recall and recognition.

[Correction Notice: An erratum for this article was reported in Vol 92(4) of Psychological Review (see record 2008-10982-001). In this article, there were two erroneous sentences, one on page 284 and one on page 285. The sentences are corrected in the erratum.] Compares 2 closely related distributed memory models in terms of plausibility; arithmetic simplicity; economy of storage space; and ability to account for associative, similarity, and order data in recall and recognition. It is argued that the storage-retrieval system brought about by the convolution-correlation concept outlined by M. A. Eich (see record 1983-04922-001) and B. B. Murdock (see record 1983-04936-001) is neurally implausible, necessitates more complex analyses, and is less economical in storage space than is the matrix memory concept described by J. A. Anderson et al (see record 1978-22353-001). It is shown that the matrix model can easily account for associative symmetry–asymmetry data and for item similarity effects. Means and variances of operating strength for various recall and recognition situations, modeled by the matrix system, are presented, and it is shown how signal-to-noise ratios can be derived. (25 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Independence and exclusivity among psychological processes: Implications for the structure of recall.

If a particular outcome can arise in each of two or more different ways, how are the different ways related? In particular, to what extent do these different underlying processes overlap in their effects? When the outcome and underlying processes are psychological in nature, theorists directly addressing the issue commonly assume independence among processes and hence some overlap in their effects. In general, however, an alternative is to assume exclusivity among processes and hence some overlap in their effects. It is suggested that theorists may therefore develop both independence and exclusivity forms of multiple-process models, allowing choice between them to be made on empirical rather than a priori grounds. This theoretical approach is adopted in the specific case of memory retrieval. An existing model of retrieval mechanisms in recall is developed to provide independence and exclusivity submodels for the structure of recall whose predictions can be compared. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Integrated retrieval cues as a mechanism for priming in retrieval from memory.

Spreading-activation theories are contrasted with composite retrieval-cue theories in priming of associative judgments. Under spreading activation, priming causes a change of memory state from quiet to active before the appearance of a test item. Under retrieval-cue theories, the prime is combined with the test item during retrieval. Spreading-activation models must predict that a relation between the prime and one element of a mispaired test will damage rejection—the partial-overlap prediction. Some retrieval-cue theories avoid this prediction with composite representations. We present one speed-accuracy trade-off and two reaction time experiments that demonstrate substantial discriminative priming (increasing hits more than false alarms) without consistent partial-overlap effects. Results suggest a combined retrieval-cue account of priming and independent storage of composites and parts (e.g., B. B. Murdock; see record 1983-04936-001). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Resolving the apparent discrepancy between the incongruency effect and the expectancy-based illusory correlation effect: The TRAP model.

The incongruency effect and the expectancy-based illusory correlation effect seem contradictory because they describe apparently contrasting consequences of previously held expectancies: better recall of incongruent than congruent items but overestimation of congruent items. This article resolves this dilemma by presenting a model that is able to simultaneously predict both of these effects. The Twofold Retrieval by Associative Pathways (TRAP) model adopts the encoding assumptions of person memory models but distinguishes between two different retrieval processes, exhaustive and heuristic, hypothesized to underlie recall and frequency estimation, respectively. Experiment 1 showed that expectancy-based illusory correlation effects and incongruency effects are compatible in that they were produced simultaneously. Experiments 2 and 3 tested and rejected alternative explanations for the obtained pattern of results. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Resampling techniques in the analysis of non-binormal ROC data

The methods most commonly used for analyzing receiver operating characteristic (ROC) data incorporate "binormal" assumptions about the latent frequency distributions of test results. Although these assumptions have proved robust to a wide variety of actual frequency distributions, some data sets do not "fit" the binormal model. In such cases, resampling techniques such as the jackknife and the bootstrap provide versatile, distribution-independent, and more appropriate methods for hypothesis testing. This article describes the application of resampling techniques to ROC data for which the binormal assumptions are not appropriate, and suggests that the bootstrap may be especially helpful in determining confidence intervals from small data samples. The widespread availability of ever-faster computers has made resampling methods increasingly accessible and convenient tools for data analysis.