Bidirectional translation of pattern and symbol information

Development of a knowledge representation for a real world operable intelligent system is one of the major targets of this project. For an implementation of intelligent system, there are two major conventional methods. One is a symbolic processing based on a symbolic knowledge representation and a logic inference by the symbolic logic. Though it is effective on some logical tasks of intelligence simulation, it is not so good for a pattern information processing like real world signal. Another method is a pattern recognition. Although it has much accumulation in pattern information processing, an integration with the symbolic knowledge is not realized yet. However, human is said to have both of the symbolic knowledge and the sensuous image. Human’s effective real world recognition is realized by their close interaction. The visual, auditory and language informations look like a single process in human, while they are thought to have different modality and process in engineering. In this paper, we propose a model of symbol and pattern information integration for an image understanding. We used a symbol-pattern bidirectional translation by selective attention model [1] for the integration. In the model, the symbol and the pattern are driven by different theories, and the bidirectional translation reflects their results to each other. For a prototype system, we have implemented a simple image understanding process with the model [2].