A Hebbian model of the lexical-semantic memory exploiting differences between distinctive and salient features

An important issue in lexical-semantic memory models is the formation of categories and taxonomies, and the different role played by shared vs. distinctive, and salient vs. marginal features (Cree et al., 2006). Aim of this work is to extend our previous model of the semantic and lexical memory (Ursino et al., 2011) to discuss the mechanisms leading to the formation of categories, and to investigate how feature saliency can be learned from past experience. The model assumes that the object semantics is represented as a collection of features, which belong to different cortical areas and are topologically organized. Lexical items are stored in a distinct network. Excitatory synapses among features in the semantic network, and among object representation and lexical items, are created on the basis of past experience of object and word presentation. In particular, we tested an Hebbian paradigm, including the use of potentiation and depression of synapses, and thresholding for the pre and post-synaptic activity (i.e., the strength of the Hebbian rule depends on the comparison of the pre or post-synaptic activity with a given threshold) to find which rules are compatible with a robust self-organized memorization of categories. The model was trained using words + simple schematic objects as input (i.e., vector of features) having some shared features (so as to realize a simple category) and some distinctive features with a different saliency, simulated assuming a different frequency of features during training. Results show that categories can be formed from past experience, using Hebbian rules and normalization of synapses. The trained network is able to solve simple word and object recognition tasks, by maintaining a distinction between categories and individual members in the category, and providing a different saliency to features occurring more frequently. Results can provide indications on which neural mechanisms can be exploited to form a robust representation of categories, within the grounded cognition framework, and on which mechanisms could be implemented in artificial connectionist systems to extract concepts and categories from a continuous input stream of word and objects.