|
Nosofsky, R.M., & Zaki, S.R. (1998). Dissociations between categorization and recognition in amnesic and normal individuals: An exemplar-based interpretation. Psychological Science, 4, 247-255. Primary Reviewer: Angie Secondary Reviewer: Amanda Purpose In the article, the authors reinterpret data in order to garner support for a single-system exemplar-memory model. According to exemplar models of classification, people store individual exemplars in memory and classify new objects based on their similarity to stored exemplars. Exemplar theory assumes that categorization and recognition performance can be explained using a unified framework that depends on a single representation system. However, in some cases findings of low correlations between classification and recognition have been viewed as problematic for single-system exemplar-memory models. Nosofsky (1988) answered these criticisms by showing that dissociations between classification and recognition performance is well predicted by a single-system exemplar model if different decision rules operate for the different tasks, through a single representational system. More recently, neurological evidence of dissociations between categorization and recognition appear difficult for the exemplar model to explain. The primary objective of this article is to demonstrate that recent findings of dissociations between categorization and recognition performance do not pose problems for single-system exemplar memory models. This is done by testing classification at different time delays, varying sensitivity parameters in mathematical models, offering alternate explanations of some patient data, and bringing into question the generalizability of other patient data. Experimental Work Knowlton and Squire (1993) asked amnesic patients and matched normal control subjects to complete old/new recognition judgments and categorization tasks for sets of dot patterns generated from a prototype. For the categorization task, participants were trained by viewing 40 high distortions of the prototype. They then viewed four presentations of the prototype, 20 low distortions and 40 high distortions of the prototype, and 40 random patterns. Participants had to judge whether or not the new patterns they were viewing belonged to the same category as the original training patterns. The critical finding was that both groups (amnesics and normals) performed similarly on the categorization task. For the old/new recognition judgment task, participants had viewed five random dot patterns eight times. At test, they were presented with the five patterns they had previously seen and five new random patterns. They had to judge whether each pattern was new or old. The normal subjects performed significantly better than the amnesic ones on the recognition test, an expected finding. Also, there was a significant interaction between the groups on the two tasks. The results were interpreted as support for multiple memory systems, an implicit system (still intact in amnesics) responsible for category-level knowledge acquisition and a declarative system responsible for recognition. Nosofsky & Zaki ran an experiment with the same dot patterns as Knowlton’s with “normal” participants. The purpose was to test whether the single-system exemplar-memory model could account for the classification-recognition dissociation reported above. They sought to “simulate” amnesia by introducing delay before testing. Participants were trained on either classification or recognition tasks and tested either immediately afterwards, one day later, or one week later. Performance on the recognition task was significantly worse in the one-week-delay test condition than in the immediate condition, but performance on the classification task did not differ significantly across conditions. Also, there was a significant interaction between condition and task. These results indicate, to the authors, that the dissociation may be attributable to parameter changes (e.g., memory sensitivity). The authors also fit a version of the exemplar model to Knowlton and Squire’s data in order to provide further support for a parameter change explanation. According to the exemplar model, classification decisions are made by summing the similarity of a stimulus to the exemplars stored in memory and then comparing the summed similarity with category criterion. The key was to have different sensitivity weights assigned to the normal and amnesic participants. Taking memory sensitivity into account, the predicted levels of performance in the categorization and recognition tasks mirror the results observed by Knowlton and Squire. Knowlton et al. (1994, 1996) used another paradigm to document dissociations between categorization and recognition tasks in normal and amnesic patients. Participants had to classify stimuli composed of four binary-valued dimensions that were probabilistically associated with two categories. After the classification task, participants were given a memory questionnaire. During the first 50 learning trials, the groups did not differ significantly in their performance. The normal subjects, though, performed significantly better than the amnesics after extended learning trials. Also, the normal subjects performed significantly better on the memory questionnaire. The authors’ explanation is that early classification learning is governed by an implicit memory system, which is intact in amnesics. During later trials, though, normals may have performed better because they used a declarative memory system. By again assuming that amnesics have lower memory sensitivity than normals, Nosofsky and Zaki ran simulations of 1,000 random stimulus sequences that satisfied the constraints of Knowlton et al’s experiment using a probability equation from an exemplar model. They set different memory parameters for the amnesics and normals. When the memory sensitivity parameters were adjusted, the predictions from the exemplar model fit with Knowlton’s observed findings. During the first 50 trials, predicted performance was the same for both groups because background noise dominates and responding is close to chance for all. The high-sensitivity group, however, is predicted to gain advantage because similarity grows as trials continue and background noise plays less of a role. This mirrors Knowlton et al’s finding. Nosofsky and Zaki then address a double dissociation provided by data from patients with PD. Groups of PD patients displayed reverse trends compared to amnesics in recognition and categorization tasks. The PD patients performed as well as the normals on the memory-questionnaire, but performed significantly worse than both normals and amnesics on the early classification task. Nosofsky and Zaki claim many possible explanations exist to account for this apparent double dissociation. They suggest that these results can be produced in a single-memory-system model with multiple component processes interacting. Patient E.P., according to the authors, may provide the greatest challenge to the single-system exemplar model. E.P. recognized a consecutively presented pattern at chance levels (normals perform with 95% accuracy), but performed with the same accuracy as normals in a categorization task. Such a pattern of data does not fit with the version of the exemplar model presented in the article, but explanations may exist. Questions also arise regarding generalizability because the results only pertain to one subject. Conclusions: In response to claims that double dissociation is evidence of separate memory systems underlying recognition and categorization tasks, Nosofsky & Raki demonstrated that the data were not irreconcilable with the idea of a single exemplar-based memory system underlying categorization and recognition, as long as plausible differences in parameter settings are allowed across groups. Regarding the PD patient data, the authors suggested the existence of possible alternative explanations, yet said more research is necessary before any conclusion is drawn. The authors also brought into question the generalizability of patient data to normals. Points for Discussion: Regarding Patient E.P., the authors question the reliability of the data. After all, E.P. is a single subject and how much weight should such data be given in consideration of generalizability? One of the central assumptions of cognitive psychology is that of the existence of a common functional cognitive architecture. Is it fair to allow the authors to question one of cognitive psychology’s basic assumptions? Possibly, E.P. is the “Martian Among Us”, but can his data be so nonchalantly discarded because it does not fit well with a single-memory system explanation? The authors claim “the onus begins to shift back to the multiple-memory-system” regarding the rebuttals to possibly damaging dissociation data. The predictions of data by the exemplar model are rather convincing arguments. The study with time delay before testing was also a relatively strong argument. On the other hand, the explanations regarding the PD patients and Patient E.P. seemed weakly constructed. Which school of thought really has the more empirical support, the single-system view or multiple-systems view? Where does the onus really lie regarding explanations of the data? |
|
This page was last updated: |