Connectives Acquisition in a Humanoid Robot Based on an Inductive Learning Language Acquisition Model
Dai Hasegawa, Rafał Rzepka, Kenji Araki · InTech eBooks · 2009
5.1 Compositionality of Rules In Fig. 9, the system responded to unknown connective commands with an accuracy of about 0.7 after 90 commands. These commands are new combinations of EoAs. The number of possible commands that can be made by such combinations is 2,500, because the number of EoAs is ten and the number of connectives that were input by users is 25. Still the system could respond with high accuracy by learning only a small number of commands. Thus, we can conclude that the acquired connectives that are abstract rules have compositionality and the rules can handle many new meanings by combination of simple sentences. Such compositionality is one important aspect of natural language. The system is able to produce different (and not previously input) results by varying EoAs. This is one advantage of our learning algorithm. 5.2 Learning Algorithm In the learning algorithm which we suggested, the system can only connect two simple