INTRODUCTION
The regulation of action and behavior involves the complementary mechanisms of selection and inhibition. Schizophrenia patients show reduced ability to select the appropriate responses from among irrelevant responses, as in the n-back task (i.e., Carter et al., 1998; Perlstein et al., 2003; Rubia et al., 2001; Weisbrod et al., 2000), and to inhibit prepotent responses in tasks of random generation (RNG) of numbers (Artiges et al., 2000; Horne et al., 1982; Rosenberg et al., 1990; Shinba et al., 2000), rhythms (Jimeno et al., 1999), and of letters (Salamé et al., 1998). RNG consists of producing at random a long sequence of verbal (numbers/letters) and nonverbal (finger taps) responses. The randomization process involves the selection of one response from the set of possible items and the inhibition of strategies and prepotent responses such as repeating, counting, and producing stereotyped digrams (Baddeley, 1966). RNG tasks are resistant to the effects of practice (Evans, 1978; Horne et al., 1982) and are considered to be on-line executive tasks subserving the supervisory function of working memory (Baddeley, 1986, 1996; Baddeley et al., 1998). In RNG, clear explanations of the concept of randomization and examples of deviations from randomness are provided so that participants know what types of responses should be overridden. Recently, M.C. Anderson and C. Green developed the Think-No-Think (TNT) paradigm as a means of addressing the mechanisms of intentional inhibition and suppression of unwanted memories (Anderson and Green, 2001). TNT is a long-term memory paradigm that includes explicit instructions and a lengthy training process to “push out of mind” (suppress) certain specific target words and respond to other cue-words. It also manipulates experimentally the number of Respond and Suppress trials (1, 8, and 16 times in each case), capitalizing on recurrent retentions/inhibitions. The Respond trials were included to act as a control for cumulative learning and improved memory. The selective intentional inhibitory effects could only be demonstrated in the Suppress trials. Thus according to the logic of TNT, the most efficient inhibition shows up as a poorer memory, following multiple attempts to forget relative to baseline (items learned but not repeated), taking the form of a negative slope from baseline (0 repetition) to 1, 8, and 16 repetitions to Suppress. In healthy subjects, they showed that, while 16 Respond repetitions resulted in improved accuracy, 16 Suppress repetitions effectively resulted in reduced accuracy, which was subsequently interpreted as reflecting the workings of a controlled inhibitory mechanism that prevents unwanted memories from accessing consciousness thanks to intentional repetitions to forget.
To our knowledge, the ability of schizophrenia patients to inhibit intentionally using the TNT paradigm has not been studied. A few explorations of intentional forgetting in schizophrenia were carried out using the directed forgetting (DF) paradigm (Muller et al., 2005; Sonntag et al., 2003). The DF and TNT paradigms differ in at least three major dimensions: (a) the number of times each item is presented in the learning phase (once in DF vs. at least twice in TNT); (b) the type of memory test (recognition vs. recall); and (c) the frequency—and insistence—with which the words to be remembered or forgotten are repeated, only once in DF and many times (up to 16) in TNT. Thus these paradigms are unlikely to tap the same cognitive processes.
We sought to examine the ability of patients with schizophrenia to exert intentional inhibition in TNT and attempted to reproduce the patterns of results of Anderson & Green. Moreover, given previous RNG studies that all point to reduced performance in schizophrenia, and because it is assumed that both TNT and RNG address the ability to inhibit prepotent responses, we sought to compare the outcome of these tasks within the same study. Reduced patient performance was predicted in RNG, in line with existing evidence. In TNT, we reasoned that, if the patients' intentional inhibition were defective, their retention of Suppress items should not be influenced by the number of repetitions (i.e., little or no forgetting after 16 Suppress attempts). Relative to RNG, such an outcome would point to some generalized inhibition deficit in schizophrenia. Alternatively, if the patients' retention rates were lower after 16 Suppress repetitions relative to baseline, as depicted by Anderson & Green (2001), this finding would be evidence of their preserved ability to inhibit prepotent responses in TNT. It would also indicate that the patients' inhibition deficit is selective, not general. The present study aimed to examine these alternatives.
MATERIALS AND METHODS
Participants
Twenty-three outpatients with schizophrenia were recruited. All met the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV; American Psychiatric Association, 1994) criteria for the paranoid (N = 12), residual (N = 5), and undifferentiated (N = 6) forms of schizophrenia as determined by consensus of the current treating psychiatrist and two senior psychiatrists in the research team. Patients who had a history of traumatic brain injury, epilepsy, substance abuse, other neurological conditions, or organic mental disorders, or were being treated with antidepressants, benzodiazepines, or lithium were excluded. All patients were clinically stabilized at the time of testing. Twenty-four healthy controls were also recruited and were matched with the patients in terms of age and years of education. None had a history of alcoholism, drug abuse, or neurological or psychiatric illness, and none were taking any drugs. The demographic, behavioral, and clinical characteristics of the two groups are presented in Table 1.
Demographic, behavioral, and clinical characteristics of the schizophrenia patients and healthy controls
Table 1 shows that the two groups were comparable in terms of their age, educational level, intelligence quotient, as measured using the Wechsler Adult Intelligence Scale-Revised (WAIS-R; Wechsler, 1981), and forward and backward digit spans. The research was conducted in accordance with the Helsinki Declaration. The Strasbourg ethical committee approved the study, and each participant signed an informed consent form before the experiment and received financial compensation for taking part.
The TNT Paradigm
It was borrowed from Anderson & Green (2001) and adapted into French. The task consists of a paired-associate learning coupled with a Respond/Suppress learning procedure and comprises three main phases: a study phase, a TNT-practice phase, and a phase of probe tests. The study phase consists of learning 50 pairs of unrelated words (i.e., CODE ARMÉE; see Appendix). With each pair, the participant is instructed to study the words and attempt to link them in any way that would best allow them to recall the target (right-hand) word if they are later given the cue (left-hand) word that was studied with it. Then their ability to recall the target word of each pair is examined. A trial starts with the presentation of a fixation cross in the center of the screen, followed by a cue word to which the participant is required to respond as quickly and accurately as possible by saying aloud its associated target word. A 5200-ms response delay is allowed. As soon as the computer's microphone detects the response, the computer stops and the experimenter types it in on the keyboard. Then the correct answer is displayed in blue for 2000 ms, regardless of the nature of the response provided (i.e., correct, error, omission), before the computer moves on to the next trial. The purpose of this feedback is to allow re-learning the pairs in the event of erroneous or missing responses, and/or to strengthen the cue–target associations and enable responding more quickly. The purpose of the next practice phase is to give participants explicit and detailed instructions regarding the Respond/Suppress procedure. Participants are told they will see cue words on the screen and that for most of these cues they have to respond and provide the associated target words learned previously. For some of the cues, however, they must not only refrain from saying the associated target word of the cue, but also try to forget the item and dispel it from their minds. If they mistakenly respond to these cues, the computer beeps loudly, reminding them that it was a Suppress item. Following a short practice, participants have to demonstrate that they have learned all 15 critical cue words corresponding to the Suppress items perfectly (100% correct) before they can move on to the TNT-practice phase in which the experimenter re-emphasizes the importance of trying to forget the responses for the Suppress trials rather than just avoid saying them aloud. Once the TNT-practice phase is over, retrieval is tested in two modalities. In one modality, it is cued with the same initial cue-words studied earlier (i.e., CODE ARMÉE), or “same probe” (SP). In the other modality, retrieval is cued with an unstudied category and letter-stem cue of the target words (i.e., CODE ARMÉE would be cued as “Militaire–A”), or “independent probe” (IP). The IP modality was used to isolate the contribution of inhibition from alternative explanatory mechanisms of reduced recall following repeated suppressions, such as the generation of diversionary thoughts using alternative associations, or unlearning (see Levy & Anderson, 2002, for a lengthier argumentation).
Procedure
The study was carried out in two separate sessions of approximately 100 min each, devoted to TNT, and RNG, and span measures, respectively. This task order was set for all subjects. TNT was controlled by a Dell laptop equipped with a built-in microphone and connected to an additional TFT screen. In the study phase, each pair was displayed for 5000 ms and a 600-ms blank interval separated successive pairs. During this phase, the participant went through as many as three successive test-learning cycles with feedback, where a single cycle consisted of presenting the 50 cue words. The only difference between the three cycles (other than the words order, which changed with each cycle) was that the display time for each cue word decreased from an initial 2000 ms to 1000 ms in the second cycle and 500 ms in the final cycle. A threshold of 94% (47 pairs) of correct responses was set to ensure effective learning of the utmost word-pairs before proceeding to the subsequent phases of the experiment. If the threshold was met at the end of a cycle, the computer moved on to the next phase. If not, a new cycle started. If the requirement of 94% accuracy was not met by the end of the third cycle, the participant was allowed to start again from the early stage of the experiment, if he/she so wished, or was withdrawn from the study. (Note: in the study by Anderson & Green (2001), a level of 50% correct was required, and subjects who failed to reach this criterion after the third cycle were eliminated.) As regards practice with the Respond/Suppress procedure, a sequence of 32 (24 Respond and 8 Suppress) trials was set up. Eight “filler” pairs (Respond items) were used, and one critical (ALBUM JUNIOR) Suppress item was excluded from the subsequent phases of the experiment. As regards learning the 15 critical words that required later suppression, the list was displayed in one central column on the screen and participants were allowed full time to learn it. A recognition procedure, involving the cue words mixed up with 14 distracters selected from filler pairs learned earlier, was used to check that the words had been learned to perfection. All 29 words were randomly displayed in two columns, and the participants' task was to select the critical words by mouse-ticking the appropriate boxes. They were allowed up to six attempts, and with each new attempt, the whole list was randomly re-presented and followed by a recognition test in which the critical and distracter words were also scrambled. Once the participants achieved perfect recall, the computer moved on to the next phase.
The TNT-practice phase comprised Suppress, Respond, and filler trials. Of the initial 50 word-pairs learned, 15 critical cue words were used in the Suppress trials in such a way that 5 were repeated only once, 5 eight times, and the remaining 5 sixteen times. A similar distribution was applied to other 15 word-pairs used as Respond trials. The filler trials consisted of nine pairs subjected to approximately 14 repetitions each. The remaining 11 pairs were discarded from this particular phase of the study, because 10 of them acted as 5 Respond and 5 Suppress baseline items—learned but nonrepeated—and the last pair (ALBUM JUNIOR) was used for illustration. Suppress, Respond, and filler trials were randomly mixed. For the Respond and filler items, the computer displayed the correct response in blue either once a response had been typed in or after a delay of 3000 ms (omission), before moving on to the next trial. For the Suppress items, each cue word was displayed for 3000 ms. The specifications for the TNT-practice phase were identical to those of the practice phase in all respects except the feedback (i.e., correct response), which was only displayed in filler and Respond trials when the participant failed to respond. In the SP and IP modalities, the display of the stimuli was identical. In each trial, the retrieval cue (either the studied cue word in the SP modality or the unstudied category plus letter-stem retrieval cue in the IP modality) was displayed in the center of the screen. As soon as the participant's response was given and typed in, or after a 4000-ms delay, the computer moved on to the next trial. Each of the SP/IP modalities comprised 40 trials that randomly cued Respond and Suppress items in equal number, plus 2 filler trials.
The word-pairs were displayed in uppercase, and the microphone sensitivity level was adjusted to the participant's normal voice before the start of the task. Each phase started with written instructions displayed on the screen, read out and paraphrased by the experimenter to ensure perfect comprehension. Eight counterbalancing orders were used. Of the 50 pairs of words, 40 were broken up into eight units of 5 pairs each, and only these units, not the remaining 9 filler pairs, were rotated through each of the different orders. There were eight orders, as a result of crossing four steps of repetition during the TNT-practice phase (0, 1, 8, or 16 repetitions) with the condition required (Suppress vs. Respond). The SP/IP modality order was counterbalanced between participants. In all phases of the experiment and for each trial, the participant's reaction time (RT, in milliseconds) was recorded. A record was also kept of the number of cycles needed to reach the learning threshold and of the TNT testing duration.
The RNG task consisted of orally producing sequences of numbers (1 to 10 inclusive) at random. Before the start of the task, the concept of randomness was explained carefully using the “hat” analogy (Horne et al., 1982) as an attempt to produce responses with approximately equal frequency while avoiding repetitions, counting in ones in ascending or descending order and stereotypes (i.e., 3–8 … 3–8). Responses were guided by computer-generated tones at the rates of one tone every second, 2 s, or 4 s, and for each rate 100 responses were required.
Data Analysis
Statistica 7.1 software was used to analyze accuracy and speed (RT) of correct recall in TNT, and the frequencies of ascending and descending counting and stereotypes (Evans, 1978) in RNG. A linear model of analysis of variance (ANOVA) with repeated measures was used, it included group (schizophrenia vs. controls) as a between-subject factor, and condition (Suppress vs. Respond) and repetition (0, 1, 8, 16) in TNT, or rate (1 s, 2 s, and 4 s) in RNG, as within-subject factors.
Score transformations
The arcsine transform (Cohen, 1988) is a means to evaluate differences between proportions (or percentages) with better accuracy. The square root transform taking the form:
(Freeman & Tukey, 1950, cited in Edwards, 1960), is appropriate to normalize data and stabilize variances and is helpful to correct zero values. Higher percentages were likely to be present in the TNT, whereas lower percentages and zeros could be encountered in the RNG tasks. In addition to the initial analyses based on percentages, arcsine and square root transformations were also applied on our data and the subsequent re-analyses were carried out. It came out that the outcomes of the three scoring methods proved similar, therefore, subtending uniformly the subsequent interpretations. For coherence, results based on the square root transform method of TNT and RNG data are reported. Power (1 − β) with alpha = .05 was computed. Student's t tests were applied for simple comparisons, and Newman–Keul's (NK) tests for multiple comparisons of means.
RESULTS
The TNT task
SP
The patterns of responses in accuracy and speed for each group as a function of condition and repetition are illustrated in Figure 1.
The Think-No-Think Task, same probe (SP) modality. A: Accuracy. B: Reaction time (RT). The patterns of responses in accuracy and speed for each group as a function of condition and repetition are illustrated. SCZ = schizophrenia patient; Cont = control; Supp = Suppress; Resp = Respond.
Accuracy (Figure 1A). The ANOVA showed no significant effect of group (F1,45 = 2.86; p < .10) and the condition × repetition × group interaction was not significant (F3,135 = 2.45; p < .07; Power = .60). Data of both groups were then pooled and re-analyzed in an attempt to replicate the findings of Anderson & Green (2001). There was an effect of condition (F1,46 = 33.15; p < .000001; Power = 1.0) and a significant condition × repetition interaction (F3,138 = 8.39; p < .00004; Power = 1.0). For Suppress, it revealed an effect of repetition (F3,138 = 3.46; p < .02; Power = .77), and the NK test showed that the difference between baseline and 16 repetitions was significant (p < .04). For Respond, there was an effect of repetition (F3,138 = 10.70; p < .00002; Power = 1.0), and the NK test revealed that recall improvement was significant after one repetition relative to baseline (p < .002), and leveled off after eight repetitions.
RT (Figure 1B). The ANOVA failed to show a statistically significant effect of group (F1,45 = 3.77; p = .059), but revealed significant main effects of condition (F1,45 = 66.92; p < .00001; Power = 1.0) and repetition (F3,135 = 5.70; p < .002; Power = .94). It also showed a group × condition × repetition interaction (F3,135 = 5.46; p < .002; Power = .93). Subsequent analyses examined the conditions separately to disentangle the terms of the interaction. In Suppress, there was a group × repetition interaction (F3,135 = 8.46; p < .00004; Power = .99). Paired comparisons showed that RTs of both groups were similar in baseline and after one and eight repetitions but differed significantly from each other after 16 repetitions (Patients, 1833 ms; Controls, 1342 ms; t45 = 3.70; p < .0006). In Respond, there was no effect of group (F1,45 = 3.18; p = .09), the repetition effect was significant (F3,135 = 15.20; p < .0003; Power = 1.0), and there was no interaction. The NK test revealed that RTs decreased from baseline (1430 ms) to 16 repetitions (1171 ms) and that all repetition steps differed from each other (p < .003 or better), except between steps 8 and 16 (p > .05).
IP
The patterns of responses in accuracy and speed for each group as a function of condition and repetition are illustrated in Figure 2.
The Think-No-Think Task, independent probe (IP) modality. A: Accuracy. B: Reaction time (RT). The patterns of responses in accuracy and speed for each group as a function of condition and repetition are illustrated. SCZ = schizophrenia patient; Cont = control; Supp = Suppress; Resp = Respond.
Accuracy (Figure 2A). An overall level of 77% was achieved. The ANOVA only showed an effect of condition (F1,45 = 8.39; p < .006; Power = .81), with accuracy for Suppress (74.38%) being lower than for Respond (78.39%).
RT (Figure 2B). The mean RT was approximately 1696 ms. The analysis showed an effect of condition (F1,45 = 8.49; p < .007; Power = .81), with Respond RTs (1643 ms) being shorter than Suppress RTs (1743 ms). There was no effect of repetition and no interaction with the group factor.
The RNG Task
Ascending counting
The ANOVA showed an effect of group (F1,45 = 9.46; p < .004; Power = .85; Patients, 10.64%; Controls, 6.80%) and of rate (F2,90 = 18.34; p < .00001; Power = 1.0). The NK test revealed that ascending counting decreased steadily (p < .01 or better) from 1/s to 1/4s. The group × rate interaction was not significant.
Descending counting (Figure 3A)
The analysis revealed an effect of group (F1,45 = 5.64; p < .03; Power = .64) and rate (F2,90 = 25.95; p < .00001; Power = 1.0) and a significant rate × group interaction (F2,90 = 4.13; p < .02; Power = .72). Paired comparisons showed that at the rate of 1/4s, patients still produced significantly more descending counting than controls (t45 = 3.72; p < .0006).
RNG task. The percentages (Mean ± SE) of descending counting in ones (A) and of stereotypes (B) in schizophrenia patients (SCZ) and normal controls (Cont) as a function of rate of generation.
Stereotypes (Figure 3B)
The ANOVA showed an effect of rate (F2,90 = 24.39; p < .000001; Power = 1.0) and a nearly significant interaction of group × rate (F2,90 = 2.91; p = .059; Power = .56). Paired comparisons showed that patients produced significantly more stereotypes than controls at the rate of 1/s (t45 = 2.03; p < .05).
Additional Analyses
The number of cycles needed to reach the threshold in the initial learning-phase was examined. On average, participants in both groups needed approximately four cycles to reach criterion (Patients, 4.09 cycles ± .31; Controls, 3.87 ± .27; t45 < 1). In addition, participants of both groups needed a comparable time to achieve the TNT test (Patients, 108 min ± 5; Controls, 97 min ± 5; t45 = 1.56; p > .05). Correlation coefficients were also computed to evaluate the degree of potential relatedness between TNT and RNG. The comparisons involved speed and accuracy scores for Suppress after 1, 8, and 16 repetitions in the SP modality of TNT, and stereotypes and ascending and descending counting scores in each rate of RNG. A significant threshold was set at p < .01 to take into account the multiplicity of computations. No significant correlation emerged either when the groups were considered separately or when they were pooled.
DISCUSSION
We examined the ability of schizophrenia patients to exert intentional inhibition in TNT and in RNG, a task known to involve the inhibition of prepotent responses. In TNT, the patients' ability was evidenced in SP, in particular when response speed was considered. In RNG, patients performed poorly and gave more ascending and descending counting and more stereotyped responses than controls, pointing to inhibition failures and impaired on-line executive control. In addition, there was no evidence of a correlation between TNT and RNG.
Before these findings can be discussed, it is important to consider the potential impact of several factors. As regards the demographic variables, patients and controls were comparable in terms of their age and years of education. Consequently, none of these independent variables could explain the experimental results obtained. As for the impact of neuroleptic intake on performance in schizophrenia patients, two recent meta-analyses that examined this issue (Aleman et al., 1999; Heinrichs and Zakzanis, 1998) failed to show any reliable effect on performance in several cognitive tasks. An interpretation of the outcome in terms of neuroleptic intake, therefore, seems unlikely, although it cannot be ruled out. Moreover, patients and controls proved comparable in IQ and in forward and backward spans. This finding suggests that the patient sample was broadly cognitively preserved and thus only partially representative of the general patient population.
In SP, regarding accuracy, no statistically significant interaction with the group factor was obtained, suggesting that the groups behaved similarly, and the pooled data analysis showed that our study clearly reproduced the patterns obtained by Anderson & Green (2001). In Suppress, accuracy decreased significantly (<.05) from baseline to 16 repetitions. In Respond, recall increased from baseline to eight repetitions before leveling off. RT results revealed significant between-group differences and clear interactions with the group factor (Figure 2B). In Respond, RTs of both groups were comparable across repetitions, and overall RTs decreased significantly from baseline to 16 repetitions. In the Suppress condition, however, patients' RTs increased by approximately 490 ms after 16 repetitions relative to baseline, whereas controls' RTs remained virtually flat across repetitions. In IP, the analyses showed an overall condition effect in accuracy and speed, indicating that Suppress responses were less accurate (Figure 2A) and slower (Figure 2B) than Respond responses.
Two major aspects of this outline would merit more detailed explanations: one is the actual relevance of the RT results obtained in SP, the other relates to the relationship between the 94% inclusion threshold and subsequent performance.
RT results in SP are an important new feature of our study as it is the first time this qualitative variable is described in the TNT paradigm. Patients' RTs were not monotonically longer than those of controls; if this were the case, the longer RTs would have been attributed to the patients' known slowness, and would, therefore, lose their specific predictive value. Because this is not the case, patients' RT increases are likely to reflect an active search for memory traces that became weaker following 16 Suppress repetitions. Since correct responses are considered, this finding indicates that the retrieved items were fairly inhibited and about to be forgotten. Patients' RTs in the Suppress condition show, therefore, that the speed variable is a valuable, qualitative, indicator of intentional inhibitory control, which appears preserved in schizophrenia.
As for the impact of the 94% threshold, data indicated that it took patients and controls approximately four cycles to achieve the learning criterion. In other words, most participants benefited from a complete re-start of the experiment, in accordance with the procedure followed. It may be assumed that such a re-start favored a deeper semantic processing of the verbal material (Craik and Lockhart, 1972), rendering the word-pairs and their associative links harder to forget and reducing, therefore, the sensitivity of the whole TNT task.
Interestingly, inspection of graphs by Anderson & Green (2001; Figure 1A and B) reveals that, although the baseline scores were the same, overall performance in IP was lower than in SP. In the former, subjects learned less and forgot less than in the latter after 16 Respond and Suppress repetitions, respectively. It is likely that in IP the participants would have adopted a higher decision criterion for responding (Koriat and Goldsmith, 1996), owing to the unexpectedness of the unstudied category plus letter-stem cueing method. However, a significant suppression effect (condition × repetition interaction) was still found in IP in the study by Anderson & Green, not in our study, where the IP analyses revealed only a condition effect, indicating that the learning and inhibitory mechanisms were active during the test. Thus the unexpected lack of effects in IP might be the consequence of both the lowered sensitivity of TNT induced by the higher threshold, and the novelty of the cueing method. However, other interpretations remain possible.
In RNG, the typical effect of pacing, according to which deviations from randomness are more pronounced with increased task demands, was found in both groups. At faster rates and compared with controls, schizophrenia patients proved poorer randomizers and provided more ascending and descending counts and stereotypes, as previously reported (Artiges et al., 2000; Daniels et al., 2003; Salamé et al., 1998). These results suggest that schizophrenia patients experience difficulties overriding the production of pre-existing schemata (Shallice, 1982).
So far, our findings suggest that schizophrenia patients are able to inhibit intentionally in TNT (at least in the SP test), and at the same time encounter difficulties carrying out on-line inhibition in RNG. Neuropsychologically, our findings argue against a widespread executive impairment in schizophrenia that would include generally deficient inhibitory control. This suggestion raises the question of the reasons of this discrepancy. While TNT and RNG address the ability to inhibit inappropriate responses, they proved not correlated. Whether the discrepant results in TNT and RNG would reflect the operations of two distinct inhibitory mechanisms, or the differential involvement of long-term and working memory, or even some combination of these factors cannot be established. Moreover, it might be argued that finding no effect in one and impairment in another neuropsychological task can be explained by differences in tasks difficulty. The level of difficulty was clearly manipulated in RNG by the use of different rates of number generation, but not in the TNT task, making it impossible to make a conclusion about this issue. This is a limitation to our study.
Recent neuroimaging studies have suggested that the same prefrontal cortex regions, including the anterior cingulate cortex (ACC) and the dorso-lateral prefrontal cortex (DLPFC) are activated during TNT (Anderson et al., 2004; Levy and Anderson, 2002) and RNG (i.e., Artiges et al., 2000; Jahanshahi et al., 1998, 2000; Jahanshahi and Dirnberger, 1999; Joppich et al., 2004). To implement inhibitory control, prefrontal cortex may exert its effects on subcortical or posterior cortical regions, depending on the task. A positron emission tomography (PET) study comparing patients with Parkinson's disease and normal controls during RNG showed in both groups a significant decrease in randomness of responses at faster rates, which was differentially greater for the patients at the faster rate. Normal controls showed hyperactivations in the right internal segment of globus pallidus associated with left DLPFC hypoactivations when task demands increase at faster rates (Dirnberger et al., 2005). The patients showed the opposite pattern of subcortical and frontal activation with faster rates, suggesting disrupted processing in basal ganglia. A PET study of patients with schizophrenia during RNG found different patterns of covariation between randomness of responses and activation in the ACC and superior parietal regions in patients and normal controls (Artiges et al., 2000). This finding was consistent with the hypothesis of a functional disconnection between the prefrontal and parietal cortices during inhibitory processing in schizophrenia (Kim et al., 2003). Finally, using functional magnetic resonance imaging during TNT, Anderson et al. (2004) showed in normal participants that controlling unwanted memories was associated with dorsolateral prefrontal activation and reduced hippocampal activation.
Based on these accounts, it would appear that a behavioral and anatomo-functional study common to TNT and RNG tasks is needed to see whether inhibitory control is associated or not with the same or different patterns of activation in these tasks in schizophrenia patients and normal controls. Further studies should also focus on experimental factors, such as the level of learning required before the TNT-practice phase. Such work would shed light on the actual modulation of intentional forgetting and allow for a better assessment of the propensity of schizophrenia patients to exert intentional inhibition.
ACKNOWLEDGMENTS
This research was supported by Institut National de la Santé Et de la Recherche Médicale (INSERM). Thanks are due to Michael C. Anderson and Benjamin J. Levy for having generously provided full descriptions and materials of their original TNT paradigm.
Appendix
List of the 50 French word pairs used in the TNT task.
