In order to develop a system capable of understanding natural conversational speech, along with the current developments in technology for phonetic information processing, a technology must be developed that will utilize prosodic information of natural speech.

Me propose here an algorithm for generating a parsing tree that represents for the semantical relationships between phrases, based on analysis of fundamental frequency contour patterns. The feasibility and validity of our algorithm are confirmed by computer simulation experiments. The capability of prosodical sentence structure inference is confirmed using natural Japanese speech. Also, it is confirmed that our algorithm is speaker independent and task independent. Furthermore, it is confirmed that the algorithm is applicable to another language, by experiments using English speech.