Item-specific effects in recognition failure: Reasons for rejection of the Tulving-Wiseman function

The present paper addresses the problems of whether recognition failure of recallable words is a function of both recognition and recall, and whether recognition failure is restricted to a small and specifiable subset of study items. A meta-analysis of the Nilsson—Gardiner database (Nilsson & Gardiner, 1993) showed that recognition given recall was positively correlated with recognition and negatively correlated with recall. Two new experiments are reported, the first one using 48 word pairs for which recognition failure was found in previous studies. An item analysis of the data demonstrated that recognition failure occurred primarily with noun—adjective pairs. The second experiment compared Norwegian—American and American—Norwegian name pairs. Wide deviation from the Tulving—Wiseman function (Tulving & Wiseman, 1975) was observed for the latter condition. In both conditions, recognition failure occurred with only the items for which the beginnings of names shared three or more letters. It is concluded that recognition failure occurs when there exists a relationship between the members of an A—B pair that is independent of their pairing in the study context. The Tulving—Wiseman function is the result of collapsing across items in the analysis of previous studies.     

The effects of amount of processing on recall and recognition

Johnson-Laird, Gibbs and de Mowbray (1978) and Ross (1981) have argued that amount of processing, as indexed by the overall number of decisions a subject makes, provides a good predictor of incidental memory performance. Conversely, McClelland, Rawles and Sinclair (1981) have provided evidence that it is normally the number of positive decisions rather than the overall number of decisions per se that determines level of recall. The present study replicated and extended the findings of McClelland and his colleagues. In free recall (Experiment 1 and 3), cued recall (Experiment 3) and recognition in the presence or absence of context cues (Experiments 2 and 3), an account based on number of positive decisions provided by far the better explanation. Experiment 3 also revealed that the experimental manipulations had a somewhat greater effect on recall than recognition. This is explained in terms of Tulving and Pearlstone's (1966) distinction between availability and accessibility. It is suggested that words associated with negative decisions are not only less accessible in memory, there are also fewer of them available for recall and recognition.     

Duaal processes in recognition and in recognition failure

It is argued that adherence of the results of experiments in the recognition failure paradigm to the function observed by Tulving and Wiseman (1975) reflects the operation of two distinct processes in recognition similar to those described by Mandler (1979, 1980). Familiarity decisions based on the target word alone can occur when target and list cue have not been integrated and result in some measure of dependence between recall and recognition. Contextual retrieval can operate in recognition when integration has occurred, and is independent of success in recall. It is further contended that deviations from the Tulving-Wiseman function arise because special instructions given to subjects produce more or less cue-target integration than is normal, and differentially affect the relative proportion of items recognized via contextual retrieval and via familiarity decisions. The first part of the paper argues that the contextual retrieval involved in recognition failure experiments takes the form of an attempted backward recall of the list cue from the target, consistent with the suggestions of Rabinowitz, Mandler, and Barsalou (1977) and claims that the results of a paper by Fisher (1979) strongly support this view.     

Backward recognition failure: A case of retrieval success

Tulving and Wiseman (1975) have demonstrated that a systematic relationship exists between the recognition probability of recallable words and the overall level of recognition. However, this relationship has only been shown in the paradigms examining “forward” recognition failure, as defined by Rabinowitz, Mandler and Barsalou (1977). A reanalysis of their data shows that just such a systematic relationship exists in “backward” recognition failure experiments, and a model developed from the tenets of Fragmentation Theory is shown to predict this. A critical experiment which falsifies the dual-coding, dual-access model used by Rabinowitz et al. (1977) is also reported, thereby extending the generality of the Fragmentation Model.     

Exceptions to recognition failure as a function of the encoded association between cue and target

The relation of the recognition failure of recallable words to overall recognition rates is largely invariant across conditions that influence both recall and recognition separately. In two experiments, the influence of the integration of the members of A-B word pairs on this relation was investigated. In Experiment 1, it was found that deviations of observed recognition failure from predictions of the Tulving-Wiseman function (Tulving & Wiseman, 1975) were produced by shallow, nonsemantic encoding. In Experiment 2, the association of category-to-instance pairs was varied. It was found that weak associates caused larger deviations of observed recognition failure from predicted recognition failure than did strong associates. Such results suggest that a strongly encoded association between cue and target elements of A-B pairs is a necessary condition for the adherence of data to the Tulving-Wiseman function. The implications of these findings for general models of memory are discussed.     

Integration and cue overlap in recognition failure of recallable words

A highly systematic relation between the conditional probability of recognition given recall and the overall recognition hit rate has been demonstrated in a wide variety of experiments. A function describing this relationship was developed by Tulving & Wiseman (1975). Exceptions to this function have, in retrospect, been interpreted in terms of (a) a low integration between cue and target items, or (b) a high cue overlap between the two tests involved: recognition and recall. The experiment reported was designed to evaluate the joint and the separate contributions from integration and cue overlap for obtaining conformity with and exception from the Tulving-Wiseman function. In line with the predictions made, the results showed that these two factors in combination can account for data that fit the function and for exceptions above the function. In relative terms, the contribution from integration was somewhat more pronounced than that from cue overlap.     

Test-expectancy and semantic-organization effects in recall and recognition

Two experiments examined whether people expecting recall are, compared with people expecting recognition, more likely to form associations between semantically related words in a list of to-be-remembered words. People were induced to expect either a recall or a recognition test on a critical list that included three conditions of semantic organization. Words in the unrelated (U) condition were semantically unrelated to all other words on the list, whereas words in the two related conditions were semantically related to one other list word. In the related-spaced (R-S) condition, the two related words appeared in input positions separated by 5-11 other items, whereas in the related-massed (R-M) condition, they appeared in adjacent input positions. Different groups received either an expected or unexpected recall (Experiment 1) or recognition (Experiment 2) test on the critical list. In both recall and recognition, (l) people expecting recall did better than those expecting recognition, (2) memory was worst for U words, next best for R-S words, and best for R-M words, and (3) the test-expectancy and semantic-organization effects were additive. A standardizedz-score measure of category dependency in memory indicated that (1) people expecting recall were not more likely than those expecting recognition to form interitem associations between the related words and (2) recognition was category dependent, but less so than recall. Within the framework of Anderson and Bower’s (1972, 1974) theory, these data indicate that, compared with people expecting recognition, those expecting recall are not more likely to form interitem associations by tagging more pathways connecting semantically related nodes but, rather, are more likely to tag the nodes themselves. The implications that semantic-organization effects in recognition have for the Anderson-Bower theory were also discussed.     

Recognition failure of recallable words in semantic memory

In an experiment in which there was no study phase, 54 subjects were tested for recognition of famous surnames and then were tested for cued recall of the same surnames. Subjects failed to recognize 53.4% of names that they subsequently recalled. Recall was significantly higher than recognition. The relationship between overall recognition rate and recognition rate of recallable words closely resembled that reported by Tulving and Wiseman (1975) for episodic memory experiments. The present data therefore extend the generality of this relationship, and of the principle that the probability of retrieval from memory depends critically on the cues provided. It is argued that the similarity between results for episodic memory experiments and the present semantic memory experiment can be more parsimoniously accommodated by tagging theory than by episodic theory.     

Recognition failure of recallable words: free and cued recall compared

The experiment reported was conducted with the purpose of studying whether the phenomenon of recognition failure of recallable words would hold for a paradigm which involves a free recall test rather than a cued recall test used in previous research. The 3×2 design used comprised two between-subjects factors: the subjects were instructed that a recognition test, a cued recall test, or a free recall test would follow the study trial, but the actual test sequence given was recognition followed by cued recall, or recognition followed by free recall. The results demonstrate that cases of recognition failure of recallable words do occur in all six conditions, but the amount of recognition failure for the recognition-free recall test sequence was less than that predicted from the Tulving & Wiseman (1975) function. In line with previous research the data for the recognition-cued recall test sequence showed the amount of recognition failure that was predicted by this function.     

Word frequency and list composition effects in associative recognition and recall

The effects of list composition and word frequency on cued recall, associative recognition, and item recognition were examined in three experiments For pure-frequency lists, cued recall and associative recognition show better performance on common high-frequency (HF) words than on rare low-frequency (LF) words. Item recognition, however, shows an advantage for LF words. In mixed lists, consisting of half HF and half LF words, the HF advantage in cued recall disappeared; however, the word frequency effects in item and associative recognition were unchanged. These results are inconsistent with explanations based on differential attention or co-rehearsal of KF and LF words. However, the results are consistent with list strength results which show that recognition is insensitive to strength-based list composition, but that recall is sensitive to list composition.     

Semantic organization of study materials has opposite effects on recognition and recall

It has been well established for several decades that semantic organization of study materials greatly enhances recall by facilitating access to information during retrieval. However, the effect of organization on recognition, and its relationship to the effect on recall, is in doubt. We report the first direct comparison of the effects of categorically organizing study lists on recognition, cued recall, and free recall. We found that whereas organization improved recall, it impaired recognition. Organization had a larger effect on free recall than on cued recall. Within the categorized lists, recall was superior for items highly associated with the category; the opposite was true of recognition. In recall, organization improved the proportion of categories recalled, but it lowered the proportion of items per category recalled. A simple framework for interpreting the dissociation is offered. Possible mechanisms underlying the detrimental effect of organization on memory and prospects for future research are briefly discussed.     

Recognition failure of categorized words

Previous research has demonstrated that the phenomenon of recognition failure of recallable words does not hold for categorized word pairs (e.g.,flower-ROSE). We tested the hypothesis that such an exceptional finding is attributable to the use of homogeneous lists that create a situation in which the functional recall cues do not contain any information not already contained in the recognition cues. In three experiments in which categorized word pairs were used, but whose design ruled out the invariant informational overlap between recognition and recall cues, recognition failure was found to the extent expected by Tulving and Wiseman’s (1975) function. These results add to the evidence that the relation between recognition and cued recall in recognition-failure experiments is largely invariant over many otherwise relevant variables that affect recognition and recall.     

The role of integration in recognition failure and action memory

In three experiments, we studied memory for action events with respect to exceptions from the Tulving-Wiseman function demonstrated in experiments on recognition failure of recallable words. In Experiment 1, we examined exceptions of poor integration in a regular recognition failure condition (i.e., recognition of targets without contextual cues, followed by recall of targets in the presence of contextual cues). In Experiment 2, we examined exceptions of cue overlap in which subjects also had access to the information of contextual cues at recognition test. In Experiment 3, we attempted to equate the levels of recognition across the action and verbal encoding. In addition, the cue overlap and no-cue overlap conditions were studied in a within-subjects design. Results from the three experiments indicated that encoding enactment (episodic integration) and conceptual integration (semantic integration) are related to each other. As a consequence of this relationship, there is a larger independence between recognition and recall of well-integrated items with encoding enactment. On the other hand, for the poorly integrated items without encoding enactment, there is a larger dependence between recognition and recall. Even in the cue overlap condition, where there is a case of large dependence between recognition and recall, the same pattern of data was observed. The results are discussed in terms of an episodic integration view of encoding enactment.     

Word familiarity and retention

These experiments examine the relationship between subjects' familiarity judgements of words of similar (low) frequency and their recall or recognition of these words. The expected relationship between familiarity and recall was well confirmed, as was the less expected relationship between familiarity and recognition. An analysis of the vocabulary acquisition process led to more specific predictions about performance on delayed, as compared with immediate, retention tests. The most crucial of these predictions was that words which are familiar, but whose meanings are not known, are remembered by tagging sets of phonological (as opposed to semantic) features, leading to good immediate recall but poor delayed recall, and a greater likelihood of acoustic confusions following a delay. Some support was obtained for these predictions. However, subjects showed unexpectedly good retention of unknown words and it was felt that tagging alone does not account for all the findings.     

Word frequency effects in associative and item recognition

The word frequency effect (WFE) has been taken as evidence that recall and recognition are in some way fundamentally different. Consequently, most models assume that recall and recognition operate via very different retrieval mechanisms. Experiment 1 showed that the WFE reverses for associative recognition, which requires discrimination between intact test pairs and recombinations of study list words from different study pairs. Experiment 2, in which word triples were used, revealed an interaction between word frequency and test type: for item recognition, performance was better for low-frequency words; however, for associative recognition and free recall, performance was better for high-frequency words. In Experiment 3, item recognition was tested: although overall performance was better for low-frequency words, the recognition advantage for items in intact pairs was larger for high-frequency words, suggesting two components in recognition memory. These results imply common mechanisms in recall and recognition. Theoretical implications are discussed within the framework of the SAM model.     

Recall failure following successful generation and recognition of responses by alcoholic Korsakoff patients

Amnesic Korsakoff patients and controls were presented with twelve words, each shown individually in the presence of a weak associate. They were then asked to freely associate to 12 other words that were strong associates of the to-be-remembered (TBR) words. The Korsakoff patients generated, and were able to recognize, as many of the TBR words as the controls. Yet, in spite of this recognition ability, these same patients were subsequently unable to recall the critical TBR words when the weak cues were again presented. A follow-up study found that the same recognition results could be obtained with Korsakoffs months after initial presentation suggesting that the patients might initially have “recognized” the most highly associated words simply because they represented the most probable choices. Intermediate association prompts failed to generate correct responses. It was hypothesized that the Korsakoff patients cannot restructure their semantic associative hierarchy during input in such a way as to become sensitized to other than the strongest associates as prompts during recall.     

Retrieval operations in cued recall and recognition

The present study compared the effects of two classes of experimental manipulations on the recognition and cued recall of target words learned in the presence of list cues. One class of manipulations, learning instructions (repetition vs. meaningful rehearsal), had similar effects on recall and recognition, whereas the other, preexperimental association between target and cue words, had separable effects: Cue-to-target association affected only recall, while target-to-cue association affected only recognition performance. Recall and recognition were thus viewed as fundamentally similar processes, both of which require retrieval operations. Differences between the two were attributed to the differential abilities of the recall and recognition retrieval cues to access the original episodic event.     

A single-system account of the relationship between priming, recognition, and fluency

A single-system computational model of priming and recognition was applied to studies that have looked at the relationship between priming, recognition, and fluency in continuous identification paradigms. The model was applied to 3 findings that have been interpreted as evidence for a multiple-systems account: (a) priming can occur for items not recognized; (b) the pattern of identification reaction times (RTs) to hits, misses, correct rejections, and false alarms can change as a function of recognition performance; and (c) fluency effects (shorter RTs to words judged old vs. judged new) and priming effects (shorter RTs to old vs. new words) can be observed in amnesic patients at levels comparable with healthy adults despite impaired or near-chance recognition. The authors' simulations suggest, contrary to previous interpretations, that these results are consistent with a single-system account.     

Recollective experience in word and nonword recognition

The functional relationship between memory and consciousness was investigated in two experiments in which subjects indicated when recognizing an item whether they could consciously recollect its prior occurrence in the study list or recognized it on some other basis, in the absence of conscious recollection. Low-frequency words, relative to high-frequency words, enhanced recognition accompanied by conscious recollection but did not influence recognition in the absence of conscious recollection. By contrast, nonwords compared with words enhanced recognition in the absence of conscious recollection and reduced recognition accompanied by conscious recollection. A third experiment showed that confidence judgments in recognizing nonword targets corresponded with recognition performance, not with recollective experience. These measures of conscious awareness therefore tap qualitatively different components of memory, not some unitary dimension such as "trace strength." The findings are interpreted as providing further support for the distinction between episodic memory and other memory systems, and also as providing more qualified support for theories that assume that recognition memory entails two components, one of which may also give rise to priming effects in implicit memory.