Using Example-Based Machine Translation to translate DVD Subtitles

Audiovisual Translation (AVT), and in particular subtitling, has been recognised as an area that could potentially benefit from the introduction of machine translation (followed by post-editing) [1],[2],[10],[15],[27],[28]. In recent years the demands on subtitlers have increased, while the payment to subtitlers and time allotted to produce the subtitles have both decreased. Therefore this market is recognised as a potential real-world application of MT. Recent publications have introduced Corpus-Based MT approaches to translate subtitles [15],[27],[28]. An SMT system has been implemented in a Swedish subtitling company to translate between Swedish and Danish and Swedish and Norwegian subtitles, with the company already reporting a successful return on their investment [27]. The hybrid EBMT/SMT system used in the current research, on the other hand, remains within the confines of academic research, and the real potential of the system is not fully known. In this paper we report on the success to date of using this hybrid system to translate English-German DVD subtitles, based on a human evaluation of the system output using real end-users of the subtitles. The subtitles are evaluated in terms of intelligibility and acceptability. We then comment on the developments that are required to make the EBMT/SMT system more accessible for use within the AVT industry, and to put the system in a position where it could compete commercially with the SMT approach.