This paper proposes an attention expression mechanism for a robot to generate utterances spontaneously. The robot must direct the attention of a person to the same real world information as the robot for the person to interpret an utterance from the robot. The attention expression mechanism exposes the robot's focus of attention by turning its head towards the target of attention. In particular, the mechanism exposes the attention of the robot to the person as well as real world information for the person to develop joint attention. In addition, we introduce an utterance generation mechanism named Linta-III, which was developed with the attention expression mechanism.