Limits...
Why Verbalization of Non-Verbal Memory Reduces Recognition Accuracy: A Computational Approach to Verbal Overshadowing.

Hatano A, Ueno T, Kitagami S, Kawaguchi J - PLoS ONE (2015)

Bottom Line: These results demonstrate the plausibility of the recoding interference hypothesis to account for verbal overshadowing, and suggest there is no need to implement separable mechanisms (e.g., operation-specific representations, different processing principles, etc.).In addition, detailed inspections of the internal processing of the model clarified how verbalization rendered internal representations less accurate and how such representations led to reduced recognition accuracy, thereby offering a computationally grounded explanation.Finally, the model also provided an explanation as to why some studies have failed to report verbal overshadowing.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Graduate School of Environmental Studies, Nagoya University, Nagoya, Japan.

ABSTRACT
Verbal overshadowing refers to a phenomenon whereby verbalization of non-verbal stimuli (e.g., facial features) during the maintenance phase (after the target information is no longer available from the sensory inputs) impairs subsequent non-verbal recognition accuracy. Two primary mechanisms have been proposed for verbal overshadowing, namely the recoding interference hypothesis, and the transfer-inappropriate processing shift. The former assumes that verbalization renders non-verbal representations less accurate. In contrast, the latter assumes that verbalization shifts processing operations to a verbal mode and increases the chance of failing to return to non-verbal, face-specific processing operations (i.e., intact, yet inaccessible non-verbal representations). To date, certain psychological phenomena have been advocated as inconsistent with the recoding-interference hypothesis. These include a decline in non-verbal memory performance following verbalization of non-target faces, and occasional failures to detect a significant correlation between the accuracy of verbal descriptions and the non-verbal memory performance. Contrary to these arguments against the recoding interference hypothesis, however, the present computational model instantiated core processing principles of the recoding interference hypothesis to simulate face recognition, and nonetheless successfully reproduced these behavioral phenomena, as well as the standard verbal overshadowing. These results demonstrate the plausibility of the recoding interference hypothesis to account for verbal overshadowing, and suggest there is no need to implement separable mechanisms (e.g., operation-specific representations, different processing principles, etc.). In addition, detailed inspections of the internal processing of the model clarified how verbalization rendered internal representations less accurate and how such representations led to reduced recognition accuracy, thereby offering a computationally grounded explanation. Finally, the model also provided an explanation as to why some studies have failed to report verbal overshadowing. Thus, the present study suggests it is not constructive to discuss whether verbal overshadowing exists or not in an all-or-none manner, and instead suggests a better experimental paradigm to further explore this phenomenon.

No MeSH data available.


Distribution of polarity values for the “old” and “new” faces.Simulation 1 (without verbalization). The dotted vertical line indicates the decision criterion that produces “old”/“new” judgment performance comparable to humans.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4464652&req=5

pone.0127618.g005: Distribution of polarity values for the “old” and “new” faces.Simulation 1 (without verbalization). The dotted vertical line indicates the decision criterion that produces “old”/“new” judgment performance comparable to humans.

Mentions: Fig 5 shows the outcome of this analysis. Polarity values for each unit were measured on the 10th cycle of a visual recognition trial, and were averaged across the units in the visual image layer. To produce a smoother distribution curve, the outcomes were averaged across five different models (each initiated with a different random status). In this figure, the polarity distribution of the 32 “old” faces (circular markers) is located to the right of the polarity distribution of the 32 “new” faces (triangular markers). The dotted vertical line at polarity value of 0.942 indicates the “old”/“new” decision criterion that produced comparable performance to humans. With this criterion, 100% of the “old” items were categorized as “old” (i.e., hits), whereas 30% of the new items were correctly categorized as “new” (i.e., correct rejections). Thus, the averaged “old”/“new” judgment accuracy resulted in 65% correct performance (SE = 0.02). This performance is compatible with normal human recognition accuracy, as reported in the verbal overshadowing literature (e.g., 64% correct, [1]). Note that human accuracy refers to the proportion of the participants who correctly chose the target face in a single-trial test. Thus, the polarity analysis also suggests that the model was successfully trained for face recognition in terms of “old”/“new” recognition, based on the polarity values.


Why Verbalization of Non-Verbal Memory Reduces Recognition Accuracy: A Computational Approach to Verbal Overshadowing.

Hatano A, Ueno T, Kitagami S, Kawaguchi J - PLoS ONE (2015)

Distribution of polarity values for the “old” and “new” faces.Simulation 1 (without verbalization). The dotted vertical line indicates the decision criterion that produces “old”/“new” judgment performance comparable to humans.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4464652&req=5

pone.0127618.g005: Distribution of polarity values for the “old” and “new” faces.Simulation 1 (without verbalization). The dotted vertical line indicates the decision criterion that produces “old”/“new” judgment performance comparable to humans.
Mentions: Fig 5 shows the outcome of this analysis. Polarity values for each unit were measured on the 10th cycle of a visual recognition trial, and were averaged across the units in the visual image layer. To produce a smoother distribution curve, the outcomes were averaged across five different models (each initiated with a different random status). In this figure, the polarity distribution of the 32 “old” faces (circular markers) is located to the right of the polarity distribution of the 32 “new” faces (triangular markers). The dotted vertical line at polarity value of 0.942 indicates the “old”/“new” decision criterion that produced comparable performance to humans. With this criterion, 100% of the “old” items were categorized as “old” (i.e., hits), whereas 30% of the new items were correctly categorized as “new” (i.e., correct rejections). Thus, the averaged “old”/“new” judgment accuracy resulted in 65% correct performance (SE = 0.02). This performance is compatible with normal human recognition accuracy, as reported in the verbal overshadowing literature (e.g., 64% correct, [1]). Note that human accuracy refers to the proportion of the participants who correctly chose the target face in a single-trial test. Thus, the polarity analysis also suggests that the model was successfully trained for face recognition in terms of “old”/“new” recognition, based on the polarity values.

Bottom Line: These results demonstrate the plausibility of the recoding interference hypothesis to account for verbal overshadowing, and suggest there is no need to implement separable mechanisms (e.g., operation-specific representations, different processing principles, etc.).In addition, detailed inspections of the internal processing of the model clarified how verbalization rendered internal representations less accurate and how such representations led to reduced recognition accuracy, thereby offering a computationally grounded explanation.Finally, the model also provided an explanation as to why some studies have failed to report verbal overshadowing.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Graduate School of Environmental Studies, Nagoya University, Nagoya, Japan.

ABSTRACT
Verbal overshadowing refers to a phenomenon whereby verbalization of non-verbal stimuli (e.g., facial features) during the maintenance phase (after the target information is no longer available from the sensory inputs) impairs subsequent non-verbal recognition accuracy. Two primary mechanisms have been proposed for verbal overshadowing, namely the recoding interference hypothesis, and the transfer-inappropriate processing shift. The former assumes that verbalization renders non-verbal representations less accurate. In contrast, the latter assumes that verbalization shifts processing operations to a verbal mode and increases the chance of failing to return to non-verbal, face-specific processing operations (i.e., intact, yet inaccessible non-verbal representations). To date, certain psychological phenomena have been advocated as inconsistent with the recoding-interference hypothesis. These include a decline in non-verbal memory performance following verbalization of non-target faces, and occasional failures to detect a significant correlation between the accuracy of verbal descriptions and the non-verbal memory performance. Contrary to these arguments against the recoding interference hypothesis, however, the present computational model instantiated core processing principles of the recoding interference hypothesis to simulate face recognition, and nonetheless successfully reproduced these behavioral phenomena, as well as the standard verbal overshadowing. These results demonstrate the plausibility of the recoding interference hypothesis to account for verbal overshadowing, and suggest there is no need to implement separable mechanisms (e.g., operation-specific representations, different processing principles, etc.). In addition, detailed inspections of the internal processing of the model clarified how verbalization rendered internal representations less accurate and how such representations led to reduced recognition accuracy, thereby offering a computationally grounded explanation. Finally, the model also provided an explanation as to why some studies have failed to report verbal overshadowing. Thus, the present study suggests it is not constructive to discuss whether verbal overshadowing exists or not in an all-or-none manner, and instead suggests a better experimental paradigm to further explore this phenomenon.

No MeSH data available.