The popular video-sharing platform, YouTube, has long made it possible for video creators to add subtitles to their visual content. Today, it’s not only possible to watch videos with subtitles but you’re also able to see these subtitles in your preferred language, regardless of whether subtitles have been added by a video’s creator or not.
However imperfect, those automatically translated captions are easily accessible to billions of users around the globe. So if you are wondering if machine translation works for audiovisual content, the answer might be closer to ‘yes’ than you might think.
The notion of whether to use machine translation or not is a tough question. Now, it’s more a case of in what capacity do we use machine translation? Businesses are increasingly turning to machine translation as an option to improve efficiencies in their localisation strategies. In the same light, language service providers that primarily handle audiovisual content now find themselves in the same situation.
Localising audiovisual content such as films, series, TV programmes and TV adverts is a delicate matter. There are very specific requirements that govern audiovisual translation – on-screen space limitations is a prime example.
Translated subtitles have to be a suitable length to ensure they don’t take up too much space on the screen and so the viewer can read them comfortably without rushing within the allocated time for each subtitle line. This is also important for dubbing purposes.
Another aspect to consider when working with audiovisual content is that it is typically rich in puns, humour and colloquial expressions. The localised versions need to retain all those elements that make the content lively and engaging but in a form that is relevant to the target language and culture.
Turnaround times are also an important aspect. With shows airing internationally, audiences around the world expect to be able to watch their favourite shows in their local language as soon as the original is released. This has a direct impact on linguists and companies responsible for localising audiovisual content.
If you’re wondering what machine translation has to do with all of this, turnaround time is the very area where MT, if applied properly, can truly make a difference. While machine translation isn’t always a silver bullet and certainly has its limitations, it can offer linguists a good starting point for producing their final version much faster than they would starting from scratch.
Tackling machine translation quality
The quality of machine translation varies between languages. However, language pairs where results are guaranteed to be good, applying this technology takes away a significant portion of effort that a linguist would otherwise have to invest in when localising content from scratch. If linguists need less time to localise content, it means less stress when deadlines are tight and more time for additional quality checks.
We haven’t yet reached a point where machine translation systems would be culture-aware and as a result, able to aptly handle creative forms of expression such as puns and jokes. Having said that, they can be trained to handle specific types of content very well.
Customised engines trained on millions of bilingual segments of subtitling data will maximise the chances of delivering optimal translation results on audiovisual content.
The latest generation of machine translation, aptly named adaptive neural machine translation, also comes with the ability for MT systems to learn the required text length over time. If it’s systematically fed corrections in the form of post-edits created by linguists, over time, it will start adapting the length of the translated text to the typical length of strings post-edited by linguists.
Machine translation may have been far from penetrating the audiovisual landscape just a few short years ago, but it seems machine translation is well on its way to becoming an integral part of the localisation process to assist linguists with the demands of translating high volumes of video content.