高齢者健康診断システムでは,問診に対する返答を"はい","いいえ"の2値で求める.しかし,医師が行う問診では,結論に至るまでにいくつかの対話がなされ,必ずしも2値で返答されるとは限らない.本論文では,複数の発話文から相手の発話意図を判断するために,真偽疑問文に対する返答発話の肯定/否定意図の抽出を試みた.特に会話文で使用される単語を副詞,感動詞,直接・間接表現述語に分類し,これらの語に対する肯定値を質問紙調査により求めた.さらに,文末表現において変化する肯定値の変化を示す式を定義した.また,真偽疑問文1個について生じる対話中で,結論としての肯定/否定意図を総合的に決めることを可能とした.本手法の有効性を検証するために,高齢者健康診断システムのインタフェース部分に適用した結果を報告する. / A computer system asks elderly people questions about their health and quality of life and usually allows them the alternatives of yes or no as their answers. A human doctor, on the other hand, has some conversations with their patients and often receives more ambiguous answers than a straightforward yes/no. This paper proposes a method to extract affirmative/negative degrees of the answers to yes-no questions in order to judge speakers' intentions from their utterances. We paid special attention to adverbs, interjections, direct/indirect representation predicates and defined their affirmative values based on the result of question-naire. We also defined a formula to calculate affirmative value change caused by different verb aspects. As a result, our proposed method can determine an overall affirmative/negative intention to a given question by analyzing users' utterances. Finally, in order to verify the validity of our method, we present the result of its application to our web-based analytical system interface for elderly people.