We propose a learning method of the speaker's taste information from natural language dialogs. Especially in this paper, we attempt to calculate favorable values from utterances containing objects whose values are undefined and emotional expressions for the utterances. First, we calculate the whole utterance's emotional degree by detecting relational words and direct emotional expressions, such as happy and sad. Next by applying Emotion Generating Calculations to this utterance, we calculate the favorable value of an undefined word. In order to verify this method, we provide the result of our experiment to extract new favorable values from a naturally-occurring dialog corpus.