Natural language processing has traditionally relied solely on information extracted from written materials. Recently, as part of natural language processing and speech processing merge into spoken language processing, a new possibility is opening up to exploit prosodic information, which is lost when utterances are transcribed into letters or characters, for language processing. Such a possibility is worth pursuing from both linguistic and technological points of view. This paper focuses on syntactic information contained in prosodic features extracted from read Japanese sentences, and describes a method of exploiting it in dependency structure analysis.