Impact of AI on audiovisual translators
Olga Krysiak's statement at CISAC’s Dramatic, Literary and Audiovisual Works Technical Committee
Olga Krysiak
10/21/20263 min read
AI’s impact on audiovisual translators depends on who controls its use, how it affects the quality of our work and who actually benefits from any time it saves.
Audiovisual translation is about much more than words. It involves interpreting characters, humour, cultural references and what is left unsaid. We decide what can be shortened, what needs to stay and how to make a character sound convincing in another language. Whether we are working on subtitles, dubbing or voice-over, all those decisions have to fit within specific time constraints.
A sentence can sound perfectly natural and still misrepresent a scene. An AI-generated line may change an implication or introduce something that was never in the original. Identifying and correcting these problems requires professional expertise. Just because something sounds fluent doesn’t mean it is accurate.
At AVTE, one of our main concerns is the assumption that a machine-generated draft leaves the translator with very little to do. That assumption can be used to justify lower fees and tighter deadlines, regardless of how much work is actually involved. In reality, checking and correcting an AI-generated draft can take just as long as translating from scratch. And throughout the process, the translator is still responsible for the final result.
Translators must therefore have a say in whether AI is appropriate for a particular task. They must have enough time to assess and revise its output, and be paid fairly for the work involved. They must also be free to reject an unsuitable draft.
Progress should mean better translations and better working conditions. If translators are paid less, are held responsible for machine-generated errors and are given less time to ensure quality, we have every right to ask who is really benefiting.
“Are we against AI?”
We are open to tools that help translators do their jobs well. But their value has to be demonstrated in real working conditions. The fact that a system can generate dialogue quickly tells us very little about how much work is needed to make that dialogue ready for an audience. Translators need to be involved in evaluating these tools and deciding how they are used.
“Can AI save translators time?”
It can help with certain tasks, such as transcription or repetitive checks. But we need to look at the whole process, including checking the translation against the original, correcting errors and ensuring consistency. A draft generated in seconds may still take considerable time to fix. Any claims about increased productivity need to take that work into account, as well as the quality of the final result.
“If the translator checks everything, what is the problem?”
Proper checking takes time, concentration and expertise. If the fee and deadline are based on the assumption that the machine has already done almost everything, meaningful human review becomes difficult. If translators are expected to take responsibility for the finished product, they need both the time and the freedom to make all the necessary changes. Simply having a human in the process doesn’t guarantee quality.
“Should translators accept lower rates when they use AI?”
Lower rates should never be the automatic consequence of using AI. The amount of work involved depends on the material and the quality of the output, and any time savings need to be demonstrated rather than assumed.
Translators still bring professional judgment to the process and remain responsible for the finished work. Where productivity does improve, they should share in the benefit.
“What about royalties and the use of translations for training?”
These are two separate issues. Royalties compensate translators for the use of their translated works. Using those works to train AI systems is a different matter that raises its own questions about permission and payment.
Our position is clear, translators’ creative contributions must be recognised and fairly remunerated. Using their translations to train AI systems must require permission and compensation. Payment for one purpose should not be treated as blanket permission for every other purpose.
“What would we ask commissioners and language service providers to change?”
Involve translators from the outset, before deciding how the project will be handled. Evaluate tools using complete scenes and programmes, with professional quality checks. Set fees and deadlines based on the work actually required. Give translators the freedom to reject drafts that aren’t fit for purpose. And be transparent about how their translations and data will be used. These are practical steps that help ensure the final work meets the standards audiences deserve.
“Will AI replace audiovisual translators?”
We would be cautious about making confident predictions. Our immediate concern is that jobs can be cut or rates reduced before anyone has established whether the resulting quality is acceptable. Those decisions can risk pushing experienced professionals out of the industry and make it harder for new translators to build a career.
We need to address the decisions being made now and their consequences for both translators and audiences.
“What is our main message?”
Audiences deserve translations that preserve the meaning, humour and emotional impact of the original. That takes professional judgment. If the industry wants that level of quality, it must recognise our creative contribution, respect our rights and provide working conditions that allow us to deliver it.