Dealing with variation in audio description scripts

Audio description scripts represent a text type structured into several parts helping the speaker produce their recording. The heterogeneous composition and formatting of these parts makes it difficult to describe the linguistic features of audio description (AD) scripts in one go. Hence, it seems useful to implement them into a corpus tool enabling the analysis of the specificities of each AD section. In this paper, the AD scripts of 69 episodes from a German television show serve as a sample to explore a method for dealing with variation when preparing AD scripts for corpus processing. In our article, we offer a short overview of existing research on AD script corpora and on variation, and we present our dataset and the tools we used to prepare and explore the data. We then outline the features of the analysed AD scripts and the treatments applied. In the last section of the article, we discuss our results. Our analysis leads us to conclude that modifying original data for the sake of corpus implementation (e.g. changing formatting features) is a weighty step which may have unforeseen consequences: formal variation in AD scripts conveys more meaning than expected.