Towards 'cinematic' hypertext