Prediction of next alphabets and words of four sentences by adaptive junction

Summary form only given. The authors have studied the adaptive junction, which is a feedback-type neural network for recognizing spatiotemporal patterns. Past research suggests that adaptive junction networks have two major internal representations: chain reaction and piling, and that the adaptive junction learning rule can learn the chain reaction with 1-degree feature patterns for any spatiotemporal pattern. Since the chain reaction with 1-degree feature patterns can decide which neurons should activate for the next spatial pattern, the chain reaction must perform prediction. It is has been demonstrated that adaptive junction networks performing the chain reaction with 1-degree feature patterns can behave as predictors of the next alphabets or words of four sentences: `A man cries´, `A baby cries´, `A dog barks´, and `A cat mews´. These results indicate that the chain reaction must play an important role for such cognitive behavior as prediction