Following exposure to consistent stimulus–stop mappings, response inhibition can become automatised with practice. What is learned is less clear, even though this has important theoretical and practical implications. The main contribution of this thesis is to investigate how stimulus-stop associations are acquired and the conditions under which they influence behaviour. To this end, this thesis addressed several outstanding issues concerning the associative architecture of stop learning, the role of expectancies, and the specificity of learning in inhibition tasks. Experiments 1-4 provide evidence that participants can acquire direct associations between specific stimuli and the stop goal without mediation via a single representation of the stop signal. However, these experiments also suggest that the influence of stimulus-stop associations on behaviour depends on top-down attentional settings: if participants begin to ignore the stop-associated stimuli, the effects of stop learning are diminished or eliminated entirely. Across eight experiments, this thesis provides evidence that participants generate expectancies during stop learning that are consistent with the stimulus-stop contingencies in play. However, Experiments 5-6 indicate that there may be some differences in the relationships between stimulus-stop expectancies and task performance under instructed and uninstructed conditions; stimulus-stop associations that are acquired via task instructions or via task practice have similar effects on behaviour, but seem to differ in how they trigger response slowing for the stop-associated items. Experiments 7-8 investigated the role of signal detection processes during the acquisition of stimulus-stop associations. To distinguish between stimulus-stop learning and stimulus-signal learning, the contingencies between specific stimuli and the stop goal and the contingencies between specific stimuli and the spatial location of the stop signal were independently manipulated. Although these experiments showed evidence of stop/go (goal) learning, there was no evidence that participants acquired the stimulus-signal associations. Across four experiments, this thesis investigated the specificity of stop learning. Experiments 9-10 compared the effects of training on behavioural performance in inhibition (go/no-go) and non-inhibition (two-choice) tasks. The results of these experiments revealed that learning in inhibition and non-inhibition tasks could arise through similar associative mechanisms, but suggest that the effects of training in these tasks could also depend on top-down response settings and general non-associative processes. Experiments 11-12 investigated the neural specificity of stop learning. These experiments also revealed similar effects of training across the go/no-go and two-choice tasks adding weight to the claim that training in inhibition tasks primarily influences task-general processes. Combined, the overall conclusion of this thesis is that bottom-up control can influence response inhibition but what is learned depends on top-down factors. It is therefore important to consider bottom-up factors and top-down factors as dependent, rather than independent, influences on response inhibition.

本研究目的主要在探討海洛因成癮者是否具有決策判斷的問題與抑制控制能力缺損，利用數理認知模型的參數值檢驗受試者在新式Go/No-Go作業的表現，該作業結合IGT與原先Go/No-Go作業的特性，並將作業刺激改為中性與海洛因相關的圖片，探討海洛因成癮者在此作業的決策歷程與抑制反應功能，並與控制組作比較。另者，根據誘發─敏感化理論 ( Robinson & Berridge, 1993 ) 與誘發─習慣化模式 ( Di Chiara, 2000 ) 的預測，前者認為成癮越嚴重者的抑制缺損情形會越嚴重，對藥物相關刺激的注意力偏誤也越明顯；後者則認為藥物依賴階段的初期會發生較嚴重的注意力偏誤情形，但隨著依賴程度增加而對藥物相關刺激已習慣化，尋求藥物行為也已變成自動化反應，故注意力偏誤的情形反而會減少。因此，本研究試圖在加入藥物相關刺激的Go/No-Go作業中驗證這兩種理論的預測。 本研究以48位一般控制組、88位海洛因受戒治人為樣本，利用Go/No-Go作業與線索依賴學習模式 ( CD model ) 分析受試者對酬賞或處罰的注意力 ( w參數 )、學習率 ( 參數 )、反應一致性 ( c參數 )、以及海洛因圖片價值的影響 ( I_Heroin參數 )，結果發現，比起控制組，海洛因成癮者對酬賞的注意力較低、受到藥物相關刺激影響較明顯。另外，將戒治組分為物質高依賴組與低依賴組時，比起控制組，高依賴組對酬賞的注意力較低；而低依賴組受到海洛因圖片價值影響較明顯。結果證實當面對同時具有中性與藥物相關刺激時的情況下，海洛因成癮者與過去古柯鹼成癮者在決策作業上的表現不同，不僅受到酬賞或處罰的影響，藥物相關刺激本身亦帶有某種價值，影響成癮者的決策與抑制能力；另外，結果部分支持誘發─習慣化的預測。 / The purpose of present study was to investigate the decision making and inhibitory problem in heroin addicts, and to use the mathematics cognitive model parameters to examine performance in a new Go/No-Go Task of subjects. The task combined the characteristics of IGT and original Go/No-Go Task. We changed the task’s stimuli to neural and heroin related pictures, to explore the decision making and inhibitory reactive function in the task, and compared with control group. On the other hands, according to Incentive-sensitization theory (Robinson & Berridge, 1993) and ‘Incentive ─ habit’ model (Di Chiara, 2000), the incentive-sensitization thoery thought that people with increasing severity of addiction get more inhibitory problems and attentional bias toward drug related cues, but the incentive-habit model suggests that attentional bias for drug cues will be more severe in the early stage of drug dependence., but with the increase in dependence and habit to drug cues, drug-seeking behavior has become automatic reaction, so attentional bias will be attenuated when abusers get more dependence. Therefore, this study attempted to verify the prediction of two theories in the Go / No-Go Task with drug-related stimuli. This research was based on the data collected from the 48 normals and 88 heroin addicts, using Go/No-Go Task and Cue-dependence learning model ( CD model ) analysis their attention to reward or punishment ( parameter w ), learning rate ( ), reactive consistence ( c ), and valence of heroin pictures ( I_Heroin). The result revealed that compare to controls, heroin addicts had fewer attention to rewards, and more obvious influenced by drug related stimuli. Otherwise, to divided addicts group into high dependence group and low dependence group, compared to control group, the high dependence addicts had lower attention to reward, and the lower dependence addicts was higher influenced by drug related stimuli. It confirm that when heroin abusers faced to neural and drug related stimili, their performance in decision making task was different from cocaine abusers, it’s not only influenced from reward or punishment, but also from the valence of drug related cues. In addition, the result partially support the prediction of the incentive-habit model.

L'inhibition est au cœur du contrôle cognitif. Mais son étude est ardue parce qu'elle s'adapte mal des méthodes psychophysiologiques classiques. En effet si l'inhibition est efficacement implémentée, tout comportement observable est supprimé! En conséquence, les modèles psychologiques existants sont fragiles, et les méthodes d'imagerie rapportent des résultats discordants. Une méta-analyse a d'abord été utilisée pour montrer que les travaux classiques confondent activations liées aux mécanismes d'inhibition réactive et sélective et activations liées aux fonctions cognitives corolaires mises en jeu dans les tâches expérimentales. Cela nous a amené à proposer un modèle alternatif d'inhibition, conçu comme un verrouillage anticipé et non-sélectif (contrôle proactif). Ce modèle a d'abord été testé au moyen de deux expériences psychophysiques. Les résultats suggèrent que l'inhibition proactive est le mode de fonctionnement par défaut du système exécutif. Le modèle proactif a ensuite été éprouvé en même temps que ses deux concurrents (réactif sélectif et réactif non-sélectif) à l'aide d'une étude IRMf. Les résultats confirment que le modèle réactif sélectif est peu plausible, au contraire de ses deux concurrents plus récents et moins populaires. La dernière partie est consacrée aux conséquences cliniques de ce renversement théorique. Alors que le modèle standard assume que l'impulsivité est la seule conséquence de troubles de l'inhibition, le modèle proactif prédit aussi des difficultés à initier un mouvement. C'est ce que nous démontrons, en faisant le lien entre hyperactivation du réseau de contrôle proactif et l'akinésie chez le patient parkinsonien / The inhibition is a key function of the cognitive control. But its assessment is made difficult by its very definition. Indeed, if the inhibition is properly implemented, the observable outcomes are suppressed. In consequences, the existing psychological models are fragile, and the imaging results inconsistent. A meta-analysis was first used to prove that the classical studies confound activations related to the reactive and selective inhibition and the activity elicited by parallel functions involved in experimental tasks. This led us to propose a new model of inhibition, defined as proactive mechanisms implemented in anticipation of stimulation. This model was first tested with two behavioral experiments. The results suggest that proactive inhibition is the default mode of the executive control. The proactive model was then assessed together with its two competing models (reactive selective and reactive non-selective) in an fMRI study. The results confirmed that the reactive selective model is less plausible than the two non-selective models. Finally we focused on the clinical consequence of this theoretical upheaval. When the standard model predicts that the impulsivity is the only consequence of inhibition deficit, the proactive model also predicts a deficit in movement initiation. This is what we showed by associating a hyperactivation of the proactive control network and the akinesia in Parkinson’s disease