This paper discusses a technique to structure utterance records of the meetings for requirements elicitation based on speech act theory. To elicit requirements to the system to be developed, the customers and the developers often have a series of face-to-face meetings. Utterances in verbal conversation made during the meetings include not only declarative information but also speech acts such as "commitment", "supporting a solution", "explaining a rationale" and so on. Extracting this information leads to clarifying customers' intent, decision rationale, and what problems have not been solved yet and remain unsolved in requirements elicitation phase. The point is how we extract speech acts from utterances and hold structurally them so as to use them afterwards. Our emphasis is on specific words (called keywords) included in utterances. These keywords can suggest what topics are discussed in the utterances and which speech acts the utterances have. Our technique has been assessed by experimental case studies and we have their good applicability to actual meetings for requirements elicitation.