The role of chaos in processing language
Jun Tani · 2003
The role of chaotic dynamics in recognition and generation of patterns containing grammatical structure is discussed. Two mechanisms were combined for segmenting patterns and retrieving grammar rules into a process of minimizing the energy of a bidirectionally connected neural network. Valid tree structures for input patterns were generated as the activation pattern of neurons following chaotic searches for the global energy minimum. The ambiguity problem occurring when a unique structure cannot be determined for the given input pattern is considered. It is shown that chaotic dynamics generates the sequence of possible interpretations of such ambiguity in the course of spontaneous state transitions between energy minima. It was observed that the search for valid patterns proceeds preserving the topology of the tree-like pattern space in the dynamics of intermittent chaos.>