Abstract: Category induction is a process of forming classification criterion. It involves abstraction of common features from limited samples of a specific category and turning the common features into a criterion characterizing this category. The common features provide a basis to judge whether a new stimulus can be classified into the category. Therefore, category induction is a prerequisite of classification processing, and a process of obtaining concepts and rules as well, occupying a primary position in human learning and knowledge acquirement. The event-related potential (ERP) was used to investigate the neural correlation of category induction, with geometric stimuli’s comparison as experimental task. The present study developed Bigman and Pratt’s (2004) study in the following ways: first, the successive presentation of S1 and S2 was changed to simultaneous presentation in order to eliminate the impact of the working memory load so that the meanings of related ERP components be clearer; second, with the limitation of its variation being considered, the size dimension was replaced with stripe orientation dimension, and the number of levels was changed from 3 to 4 to increase the range of feature variation; and third, a “non-induction” condition was designed, where S1 and S2 were identical, and the participants were also required to search common features.
The EEG was recorded from 64 scalp sites using tin electrodes mounted in an elastic cap (BrainProduct Inc.), with the averaged reference on the left and right mastoids, and a ground electrode (GND) was placed on the medial aspect of the frontal (on the middle line, between Fz and Fcz). The ERP waves under three conditions were overlapped and averaged respectively. Brain Electrical Source Analysis program (BESA, Version, 5.0, Software) was used to perform dipole source analysis. For dipole source analysis, the four-shell ellipsoidal head model was used. In order to focus on the scalp electrical activity related to the processing of pure induction processes effect, the averaged ERPs evoked by the “non-induction” condition were subtracted from the ERPs evoked by the two “induction” conditions respectively, and two difference waves were obtained accordingly (S1F-S3F and S2F-S3F). Principal component analysis (PCA) was employed in the interval from 400 to 650 ms in order to estimate the minimum number of dipoles. When the dipole points are determined, software will automatically determine the dipoles location. The relevant residual variance criterion was used.
Two difference waves were obtained through subtracting S3F from S1F and subtracting S3F from S2F. In the time window of 400 and 650 ms, three conditions elicited obvious LPC component, and the topographies indicated that the components activated mainly over central-parietal scalp region. There was no significant difference between two “induction” conditions, but there were significant differences both between two “induction” conditions and “non-induction” condition. Dipole source analysis (BESA 5) of two difference waves (S1F-S3F and S2F-S3F) indicated that there was only one dipole source under each difference wave, and the two dipole sources were located similarly (both were located approximately in hippocampus). It is known that hippocampus is related to the formation of novel connection, therefore the LPC may reflect the key process of category induction: formation of novel connection

Key words: category induction, event-related potential, LPC, hippocampus, dipole source analysis