Consequences of an inhibition deficit for word production and comprehension: Evidence from the semantic blocking paradigm

Abstract

We investigated the semantic blocking effect in picture naming and word–picture matching for two nonfluent aphasic patients who show evidence of a deficit in inhibiting verbal representations (M.L. and B.Q.), one fluent aphasic patient (K.V.), and neurologically intact control participants. In two picture-naming tasks (Experiments 1A and 1B), M.L. and B.Q., relative to controls, showed a greatly exaggerated semantic blocking effect in naming latencies that increased dramatically across repeated presentations. On two corresponding word–picture matching tasks (Experiments 2A and 2B), both also showed an increasing semantic blocking effect, though the effects were not as large nor as consistent as those in naming. The fluent patient, K.V., showed a pattern like controls on both tasks. On an associated word–picture matching task, both M.L. and B.Q. showed results paralleling those of controls. The contrast between the production and comprehension patterns for M.L. and B.Q. supports the conclusion that their exaggerated blocking effect in production arises during lexical rather than semantic selection. We postulate that M.L.'s (and potentially B.Q.'s) production effect is due to difficulties in postselection inhibition, which results in overactivation of lexical representations. This overactivation is likely to be one source of their nonfluency in spontaneous speech.

Keywords:

• Aphasia
• Lexical selection
• Inhibition
• Production
• Semantic blocking

This work was supported by National Institutes of Health (NIH) Grant DC-00218 to Rice University. The authors would like to thank Katherine Forquer for her help in testing the participants.

Notes

¹ Recently, the claim has been put forward that the competition is not involved in selection at the lexical level, but instead occurs at an output buffer stage (e.g., Mahon, Costa, Peterson, Vargas, & Caramazza, 2007; Miozzo & Caramazza, 2003). This view has been put forward mainly to account for findings from picture–word interference tasks in which the picture name and the word are available simultaneously and could compete for output. It is somewhat difficult to see how this hypothesis could be applied to the semantically blocked naming effect, as one would have to hypothesize that output representations persist across trials. One would assume that participants could clear the buffer at the end of the previous response, given prior assumptions about participants' ability to clear the buffer of irrelevant responses that are available early (Miozzo & Caramazza, 2003).

² For instance, suppose that in the undamaged system after some time (t), the target response typically has an activation of 150 units whereas the nearest competitor has an activation of 100 units. Due to noise in the system, which has some variance, say 30 units, the activation of the nearest competitor will occasionally be higher than that for the target, resulting in a semantic error. In the damaged system with overinhibition, at time t, activation of the target might typically be 75 units and activation of the competitor 50 units. Assuming that the variance due to noise is at least as great in the damaged system, there will be a greater proportion of trials on which activation for the competitor exceeds that for the target. Whether or not a semantic error or an omission would be predicted would depend on whether one assumed some (high) threshold for production of a response (which should lead to more omissions for the overly inhibited system) or a ratio threshold for response (which would lead to more semantic substitutions).

³ These latencies are for the naming of pictured nouns. For verb naming, M.L. is also highly accurate; however, his latencies are substantially outside the range for controls (Biegler, Martin, & Potts, 2005). As the studies reported here dealt only with noun production and comprehension, this word class difference is not considered further.

⁴ It was difficult to calculate a span measure on these probe tasks for B.Q. as had been done previously with other patients (e. g., Martin & He, 2004), as his performance fluctuated across list lengths rather than decreasing systematically.

⁵ The discrepancy in performance between the Stroop and picture–word interference paradigms is puzzling, especially since he obtained very few errors in both tasks (3/70 for Stroop and 4/60 for picture–word interference). However, recent unpublished verbal Stroop data obtained in our laboratory have also shown large interference effects in reaction times for patients who are more similar to K.V. than M.L. with regard to speech fluency, short-term memory, and language processing in a semantic context (Crowther, 2006). Conceivably, the verbal Stroop task could be especially difficult for a wide range of aphasic patients and not restricted to patients with large left frontal lesions.

⁶ An independent measures ANOVA was also run without replacement values, and similar results were found in the analysis.

⁷ Given that some production models assume that lateral inhibition is involved in lexical selection (e.g., Stemberger, 1985), one might hypothesize that M.L. has a deficit in lateral inhibition (rather than postselection inhibition) that results in overactivation. However, a deficit in lateral inhibition could not account for M.L.'s poor performance on the recent negatives task. Moreover, the predictions of a deficit in lateral inhibition are not straightforward. Presumably, lateral inhibition for patients like M.L. would work in the same manner as that for controls, but only more slowly, such that more time would be needed until a critical difference between the target and its competitors was achieved. The carryover effect of activation from previously named items from the same category would depend on the state of the target and its competitors once the selection criterion for the target was reached. If patients used the same criterion, then no difference from controls would be expected in terms of the carryover across cycles.

⁸ As discussed earlier, Wilshire and McCarthy (2002) found a higher proportion of errors on the “no” trials than on the “yes” trials of a similar task. We also observed a greater relatedness effect for the “no” than the “yes” trials for two of three patients (M.L. and B.Q.). However, we were specifically interested in whether any increasing effect of semantic blocking across cycles would be observed. For each patient, we analysed the data including yes/no as an additional factor. There was no sign of a different pattern of Relatedness × Cycle interaction for the yes versus no trials, as all Fs were less than 1.0 for the three-way interaction. Hence the data are only reported collapsing across this factor.