Investigating Repetition and Reusability of Translations in Subtitle Corpora for use with Example-Based Machine Translation

Repetition and reusability are two notions that lie at the heart of a number of current approaches to computer-aided translation (CAT) and machine translation (MT), but are rarely problematized in the literature. In this paper, we discuss these notions in the context of the example-based machine translation (EBMT) of movie subtitles in a project currently underway at Dublin City University. We start by describing, in sections 2 and 3 below, how repetition and reusability have been dealt with by other researchers in CAT and MT, before going on to outline our own approach in section 4. We characterize our approach as combining both ‘prospective’ and ‘retrospective’ phases. The prospective phase relies on measurements of levels of repetition in our texts, on the one hand, and human evaluations of the reusability of existing translations, on the other, while the retrospective phase relies on real-user evaluations of automatically translated subtitles after they have been inserted into relevant movie clips. While the focus of this paper is on the prospective evaluation of repetition and reusability in our corpora, we will also make some comments on why prospective measures and judgements might be expected to correlate with retrospective evaluations. We will provide some quantitative analysis results from the corpora used with the EBMT system, as well as an overview of the responses generated in the human evaluation. The paper concludes with comments on the usefulness of a prospective phase, and the benefits this has for the next phase of research.