Conclusions

In the preceding chapters, it has been shown that a textual topic segmentation algorithm designed to work on structured, paragraph-segmented text can be made to work effectively for the domain of transcripts of spoken dialogue.

An exploration of the background research of topic segmentation and of spoken dialogue structure has been undertaken; and a series of controlled tests have been performed to examine the effectiveness of a customised version of the TextTiling [6] algorithm on spoken dialogue transcripts, using various kinds of spoken dialogue data and various algorithm-tuning parameters. The results of these experiments have been plotted and analysed