Recently, robotic technologies for natural communication with a human have been discussed for the future of human society. A partner robot requires various capabilities for the social interaction based on both of verbal communication and non-verbal communication. To realize the verbal communication with a human, the robot should acquire the environmental knowledge and behaviors for human interactions. In this paper, we propose an utterance system using perceptual information and interaction with a human. The perceptual information is extracted by image processing based on a steady-state genetic algorithm. Furthermore, we conduct several experiments on natural communication with a human.