The Neural Correlates Of Hypothesis Formation And Testing In The Process Of Category Induction

Category induction is the process of making inference from finite samples of instances to general category. It simplizes or generalizes our knowledge or experience about the world. It is one way in which we learn and adapt to environment. It is also the important way of find new things in science. Althougth the problem of induction had been addressed by phylosphers for a long time, psychologists have begun to study its mechanisms systematically over the past three decades. Past studies manily focused on the judgment of inductive inference, the development of inductive inference, and the neural base of induction.Only recently, a few researchers have utilized image methods, and found that induction activated the following brain areas: the left medial frontal gyrus, the left cingulate gyrus, and the left superior frontal gyrus, as well as the right inferior and dorsolateral prefrontal cortex. However, these studies did not address how category induction unfolds in time, and the characteristic of dynamic activation of brain areas is still unclear. In order to study the dynamic process of induction, it is necessary to use the ERP method that is prominent in time resolution. Most firstly, the process of category is presumed to include fowllowing three stages: attention and perception, extracting the shared attributes, hypothesis formation and testing. The later two stages are the important process in category induction, especially the last stage. Up to now only two studies explored the second stage, and two studies explored the last stage. However, these past studies did not foucus directly the critical process of induction itself.The aim of the present study was to investigate the time course and neural correlates of identifying shared attributes, hypothesis formation and testing in a solvable task-category induction task, during which the process of thinking is similar to daily thinking, In the category induction task of the 4 experiments in the present study, participants were provided with series of batteries. Their task was to learn what kind of batteries could produce electricity. When they draw a conlusion, we provided them a pobe test like the way of previous studies to test the correctness of their conclusions. In this task, the process of finding a rule is the process of induction, and the process of applying the rule (ie. Responding to probe) is the process of deduction. The focus is the former in the present study.In Experiment 1 of the present study, participants presented simultaneously with sets of battery were asked to do a perceptual analysis of the attributes of batteries and indicate what kind of batteries can produce electricity. To analyses the process of induction more accurately, we assumed that the induction is based on the relation judgment, because people are sensitive to relationships among information within categories, and making inferences requires people to know how the properties of category members are related. Under this assumption, we designed two tasks. One is inductive generalization, and other is relation judgment. Subjects in generalization task were provided with many of batteries sets that consist of four batteries varied in shape or stripe. In each batteries set, some batteries can produce electricity (i.e. charged batteries) and others cannot (i.e. dead batteries). Subjects should answer what kinds of batteries can produce electricity. The materials used in relation judgment task were same as those in generalization task, but the task of subject in the relation judgment task is not to make an inductive generalization but to make a perceptual relation judgment. We expected the reaction time maybe longer in the inductive generalization task than in the relation judgment task, and also expected the late components of ERP maybe different between these two tasks. Two conditions, a generalizable condition, and a non-generalizable condition were designed in the generalization task. In the generalizable condition, one common attribute was shared with the two charged batteries. Participants should identify this common attribute and form an appropriate hypothesis. In the non-generalizable condition, participants could not make a general conclusion because the charged batteries had no shared attribute.The behavioral results in Experiment 1 revealed that the reaction time were shorter in the generalizable condition than in the non-generalizable condition. The electrophysiological results indicated that there was no main effect of condition on P1 and N1 amplitude. However, the latencies of frontal N1 were significantly longer in the non-generalizable condition than in the generalizable condition. The shorter N1 latency for the generalizable stimuli might be related to the configural and holistic perception of the shared attributes between charged batteries, or was related to the Top-Down attention. After 420 ms, the late positive complex (LPC) was increased for generalizable condition compared with non-generalizable condition, suggested that the cognitive system was updated by newly formed knowledge. In addition, a main difference between the inductive generalization task and relation judgment task, the positive going at about 310 ms in the inductive generalization task possibly reflects the hypotheses formation.Aimed to further examine the neural correlates of hypothesis tesing, a mordified task was used in Experiment 2-4. In these tasks, the series of pictures of battery varied in shape or color were provided sequentially to participants, and a possible H was expected to be formed based on the perceptual analysis of the batteries. For the purpose of the present study, the first two batteries within each trial were both charged and shared one attribute (e.g. both were in red). After the presentation of the second stimulus, subjects could identify the shared attribute and form a preliminary hypothesis (e.g. the red batteries can produce electricity). However, whether the preliminary hypothesis was true or false depended on what the third battery was. Three conditions (including Nonchange, Strengthen, and Reject) were designed in the present study. In the Nonchange condition, the third battery was irrelevant to the preliminary hypothesis. In the Strengthen condition, the third battery was charged and shared the common attribute between the first two batteries, so the preliminary hypothesis was strengthened. In the Reject condition, the third battery did share the common attribute between the first two batteries, but was not charged. So the preliminary hypothesis was rejected.The behavioral results revealed that the accuracies were high across conditions, suggesting that the HT problems in the present study were successfully solved by subjects. The highest accuracies in Strengthen condition compared with other two conditions suggested that the memory of hypothesis or newly formed knowledge was strengthened in the Strengthen condition.Consistent with the behavioral results, the electrophysiological results also revealed the significant effects of condition, which were reflected at the distinct stages of HT. First, in the stage of attention, significant effects of condition were initially seen at posterior P1, which was related to selective attention. The amplitudes of P1 were highest in Strengthen condition than in other two conditions, suggesting that selective attention was paid to the attribute that correlated to the critical property (i.e. the lighting bulb). However, this effect only appeared in Experiment 2.Guided by attention, the course enters into the process of perceptual encoding. During this period, the P2 amplitudes elicited by all three conditions in all experiments were large (see Fig. 3) and there was no difference between conditions, which suggests that perceptual encoding was intensive in all conditions. However, the P2 differed significantly between nonchange condition and other two conditions in Experiment 3, which suggested that participants payed little efforts on the perceptual coding of the simulus in nonchange condition.The process of conflict detection and similarity analyzing was reflected by the frontal N2 and parietal P3. The gradient similarity among three conditions was reflected by the gradient amplitudes at about 300 ms onset of stimulus. A largest parietal P3 was found for the Strengthen condition, suggesting the highest non-congruency between context and stimulus. The amplitude of parietal P3 was the lowest and the amplitude of frontal N2 was the highest in Nonchange condition during this period of time, suggesting the intensive process of conflict detection. This effect was found only in Experiment 2, possibly because the stimuli were more complex in Experiment 2 than Experiment 3 and 4.The results of Experiment 2-4 suggested that the maintaining or abandoning of an H was reflected by the LPC. LPC in the present study might reflect the process of H evaluation and updating of memory-context. There were three predictive results of the evaluation of an H, nonchange, strengthen, and reject. However, the difference of LPC was found only between the Reject condition and other two conditions. The LPC were almost overlapped between Nonchange condition and Strengthen condition. It seems that the brain activities in Strengthen condition did not differ significantly from that of Nonchange condition at the stage of H evaluation. The most possible interpretation is, the former H in Strengthen condition was expected to be strengthened, but the cognitive system was not completely updated, the state of knowledge was unchanged, and the memory was not erased (e.g. the knowledge such as "the red batteries can produce electricity" are still kept active in memory), hence the scalp potentials were the same to those of Nonchange condition. In contrast, the brain activities increased markedly in Reject condition, which was reflected by the increased amplitude of LPC, suggesting the process of cognitive updating and memory erasing.LPC was not only related to the process of memory updating, but was also related to central executive. The fronto-central distributed LPC in the present study is possibly related to the process of CE. This possibility was partially supported by the dipole localization for the difference wave between Reject condition and other two conditions. The results of dipole localization suggested that a cortico-striatal circuitry was associated with the process of H evaluation and memory erasing.To summarize, by using a category induction task, the present study found different brain activities at the distinct stages of hypothesis formation and testing. In the process of hypothesis formation, the non-generalizable stimuli could be differentiated from the generalizable stimuli early at 70 ms onset of stimuli. At about 310 ms the hypotheses were formed, which reflected by a positive going in the induction task. The hypotheses need to be tested further. Initially, the parietal cortex selectively attended to the critical property that was correlated to a H. At the stage of similarity analyzing, the test stimulus was matched to the representation of the former H at the central sites, where the dipole was localized approximately in caudate. Finally, the former H was evaluated and the memory was erased when the test stimulus reject the former H, reflected by the increased amplitude of LPC in the fronto-central sites.