Growing self-organizing map approach for semantic acquisition modeling

Based on the incremental nature of knowledge learning, in this study a growing self-organizing neural network approach for modeling the acquisition process of semantic features is proposed. The Growing Self-Organizing Map (GSOM) algorithm is extended and applied to the problem of language acquisition. Based on that algorithm, experiments are conducted using Standard German children's books corpus. A cyclic reinforcing and reviewing training procedure is introduced to model the teaching and learning process between children and their communication partners. Experimental results indicate that (1) GSOM has good ability to learn the semantic categories presented within the training data, that (2) clear semantic boundaries can be found in the network representation, and that (3) cyclic reinforcing and reviewing training leads to a detailed categorization of lexical items as well as to a detailed clustering, while keeping the already-learned clusters and already-developed network structure stable. Experiments show that our GSOM approach is a good method for modeling semantic learning during language acquisition.