Machine translation quality in an audiovisual context
Aljoscha Burchardt, Arle Lommel, Lindsay Bywood, Kim Harris and Maja Popović
DFKI, Berlin | University College London | text&form/DFKI, Berlin | Humboldt-Universität zu Berlin
Abstract
The volume of Audiovisual Translation (AVT) is increasing to meet the rising demand for data that needs to be accessible around the world. Machine Translation (MT) is one of the most innovative technologies to be deployed in the field of translation, but it is still too early to predict how it can support the creativity and productivity of professional translators in the future. Currently, MT is more widely used in (non-AV) text translation than in AVT. In this article, we discuss MT technology and demonstrate why its use in AVT scenarios is particularly challenging. We also present some potentially useful methods and tools for measuring MT quality that have been developed primarily for text translation. The ultimate objective is to bridge the gap between the tech-savvy AVT community, on the one hand, and researchers and developers in the field of high-quality MT, on the other.
Table of contents
Audiovisual translation (AVT) has become a fundamental necessity in the 21st century. Media trends such as VHS and LaserDiscs have come and gone, and translation tools have progressed from typewriters to fully integrated real-time web-based translation environments. Our world is becoming ever smaller, while the demand for information in every corner of the globe is growing. Consequently, the sheer volume of data that needs to be accessible in most regions and languages of the world is rising dramatically: every minute, 300 hours of video material is being uploaded to YouTube. Even assuming that only a small fraction of this content is of interest to a broader global audience, the effort required to publish it in multiple languages is a tremendous challenge. This has been recognized and acknowledged by research bodies and governments that have supported early-adopter projects involving automatic AV translation. Such projects include MUSA and eTITLE, which have used rule-based MT combined with translation memory to investigate the potential of these tools for AVT; SUMAT, which has trained statistical machine translation engines on subtitles in seven bi-directional language pairs and performed an extensive evaluation of the resulting MT quality; EU-Bridge, which has focused on further advancing the state-of-the-art in automatic speech recognition combined with MT with a view to applying this technology in several domains, including AVT; HBB4ALL, which, although mainly focused on accessibility, has carried out research into the reception of automatic interlingual subtitles; and ALST, a project whose aim was to implement existing automatic speech recognition, speech synthesis and MT technologies in audio description and voice-over, part of which included quality assessment of voice-over scripts produced using MT and post-editing.