Multilingual Aligned Corpora From Movie Subtitles Mathieu Mangeot

This paper describes a methodology for build-ing aligned multilingual corpora form movie subtitles found on the Web. The subtitles have specific formats and encodings. In a first step, we convert them to our multilingual sub-title format based on XML. In a second step, we align the subtitle sentences with the time used to display them on the screen. We im-plemented the tool Jimaku in order to semi-automatically perform these steps. The last step consists in aligning the sentences at the sub-sentence level and to index the corpus for contextual lookup. For this step, we use the WIMS platform, result of previous research on text collections management.