AUTONOMOUS SYMBOL ACQUISITION THROUGH AGENT COMMUNICATION

In this chapter, we propose a multi-agent system aiming at autonomous symbol acquisition, in which agents acquire symbols through communication instead of symbols being given a priori by the designer. Based on this idea, we extended Steels's language acquisition model to develop a new model featuring three mechanisms: (a) symbol matching; (b) symbol creation; and (c) concept selection. Intensive simulation revealed the following implications: (1) the degree of trade-off between communication success and required lexicon size can be decreased by matching all possible combinations in symbol matching; (2) symbol creation of hearer agents plays a significant role in symbol acquisition, while speaker agents do not; (3) the speed of symbol creation depends on the method used for this step, but it is not related to the trade-off between communication success and lexicon size; and (4) concept selection can also be applied to resolve the trade-off between communication success and required lexicon size.