The semi automatic generation of audio description from screenplays. Technical report CS-06-05

Audio Description is an important accessibility medium for visually impaired people, allowing them to enjoy visually-based experiences such as television and cinema. Its use is increasingly widespread, due in part to increased Government legislation. However,the production of audio description is time-consuming – it may take 60 person-hours film. This paper investigates the novel idea of generating a first draft audio description script from a film screenplay. This can be viewed as a text summarisation problem in which relevant sentences are identified in the screenplay and are then adapteto suit the style guidelines of audio description. Through a systematcomparision of screenplays anaudio description scripts we discovered that on average a screenplay contains about 60% of the information required for an audiodescription, though not necessarilyexpressed in a suitable form. We present algorithms that can recall 80% of the available sentences from a screenplay at a precision rate of 50%. Of the resulting sentences tare not in an appropriate form for audio description, our set of heuristics can then map 66% of themto a suitable form. These results, along with feedback from evaluation sessions with BBC audio describersuggest that the semi-automatic generation of audio description isossible and applicable in this important real-world scenario.