This paper suggests that video indexing is an interesting and important natural language application for which it is crucial to identify temporal information in collateral text that articulates the semantic content of moving images. Recently a rich source of information about the content of films and television programmes has become available in the form of audio description scripts. The analysis of the expression of temporal information in a corpus of audio description scripts leads to a discussion of some consequences for schemes to annotate such information in a video indexing application.