Embracing the threat: machine translation as a solution for subtitling
Egileak: Lindsay Bywood Panayota Georgakopoulou
Data: 03.07.2017
Perspectives: Studies in Translatology
Abstract
Recent decades have brought significant changes in the subtitling industry, both in terms of workflow and in the context of the market for audiovisual translation (AVT). Machine translation (MT), whilst in regular use in the traditional localisation industry, has not seen a significant uptake in the subtitling arena. The SUMAT project, an EU-funded project which ran from 2011 to 2014, had as its aim the building and evaluation of viable MT solutions for the subtitling industry in nine bidirectional language pairs. As part of the project, a year-long large-scale evaluation of the output of the resulting MT engines was carried out by trained subtitlers. This paper reports on the impetus behind the investigation of MT for subtitling, previous work in this field, and discusses some of the results of this evaluation, in particular an attempt to measure the extent of productivity gain or loss for subtitlers using MT as opposed to working in the traditional way. The paper examines opportunities and limitations of MT as a viable option for work of this nature and makes recommendations for the training of subtitle post-editors.
BIB_text
title = {Embracing the threat: machine translation as a solution for subtitling},
journal = {Perspectives: Studies in Translatology},
pages = {492-508},
number = {3},
volume = {25},
keywds = {
Subtitling, Machine Translation, SUMAT
}
abstract = {
Recent decades have brought significant changes in the subtitling industry, both in terms of workflow and in the context of the market for audiovisual translation (AVT). Machine translation (MT), whilst in regular use in the traditional localisation industry, has not seen a significant uptake in the subtitling arena. The SUMAT project, an EU-funded project which ran from 2011 to 2014, had as its aim the building and evaluation of viable MT solutions for the subtitling industry in nine bidirectional language pairs. As part of the project, a year-long large-scale evaluation of the output of the resulting MT engines was carried out by trained subtitlers. This paper reports on the impetus behind the investigation of MT for subtitling, previous work in this field, and discusses some of the results of this evaluation, in particular an attempt to measure the extent of productivity gain or loss for subtitlers using MT as opposed to working in the traditional way. The paper examines opportunities and limitations of MT as a viable option for work of this nature and makes recommendations for the training of subtitle post-editors.
}
isi = {1},
doi = {10.1080/0907676X.2017.1291695},
date = {2017-07-03},
year = {2017},
}
