Remote transcription system is one of the communication support systems for hearing impaired students in a lecture. Because the remote transcription system requires several seconds from the utterance of the instructor to displaying the summarized text, hearing impaired students cannot associate the demonstrative word. In this paper, we analyze how demonstrative words appear in the utterance of instructor. About 75% of demonstrative words follow the pause and are uttered within 500ms after the pause. We propose a method for extracting demonstrative words based on acoustic features and conducted experiments. Experimental results show that the recall rate is more than 80% but the relevance rate is less than 20% in the cross validation test.