Human or machine: How do translators use AI today?
What do translators really think about AI? This is a question that the whole industry is asking – and the answer is more nuanced than you might think.
What do translators really think about AI? This is a question that the whole industry is asking – and the answer is more nuanced than you might think. We asked around 90 language experts – representing more than 80 different language pairs – and their responses were both interesting and thought-provoking.
In March this year we conducted a survey called “Your perspective on AI”. The purpose was to get a better idea of what language professionals think about AI, how they use it in their work, and which tools they prefer and why.
Although most professional translators use AI in some form or other in their daily work, they never just accept the results it gives them. Human judgement and cultural awareness were highlighted as crucial factors in this context that no algorithm has been able to replace.
The responses show that a clear majority use AI in some way in their translation and proofreading work. But the opinions were by no means only positive: many are cautiously optimistic or have mixed feelings; a smaller group are decidedly sceptical; while another group consisting of enthusiastic fans of AI are curious about the technology and want to learn more.
Despite the potential of AI, the main misgiving is the risk of poor-quality translations. The second main issue is what is happening in terms of acceptance and the understanding that human quality assurance of AI-produced materials is needed. Will this diminish or even disappear over time? This fear is reinforced by concerns over unreliable results, the AI tools’ inability to understand cultural context, and various privacy risks. There are also longer-term fears regarding the industry itself, such as uncertainty about compensation levels and the risk of decreasing demand for professional translators.
Use
The survey shows that AI is mainly used as support in the actual translation process, but also to generate ideas, for language support, terminology searches and fact-checking, and to a lesser extent to review translations. Many see AI as a flexible tool that can help in different stages of the process, but emphasise the importance of a human always having the last word.
The most common tools – a combination is key
Among the most common tools are ChatGPT and DeepL, but Claude and Google Gemini are also gaining ground. Another group of tools consists of professional translation software, with Smartling, Phrase, Trados Studio and memoQ being the ones translators uses the most.
Translators prefer to combine several tools rather than relying on any individual solution. Established computer-assisted translation (CAT) and translation management system (TMS) environments – now including AI support – are still the most important platforms for professional translation. These are supplemented to a high degree with generative AI tools to edit, refine or get help with wording. Most translators simply prefer a toolbox where they can select a solution based on the task at hand.
Strengths and weaknesses of AI tools
When language professionals describe AI’s strengths, they mainly highlight terminology management and speed. AI is perceived as a powerful tool to find the right term or produce a first draft, and in certain cases to improve flow and style. Other positive aspects, such as functionality, user-friendliness and adaptability, are also mentioned but are less important.
The single biggest drawback is considered to be the risk of poor quality. Many therefore stress the need for thorough proofreading and essential editing of AI-generated texts. They also mention the risk of subtle, hard-to-detect errors and differences in nuance that can have a major impact. Privacy issues are also highlighted, as well as limited adaptability and other functional limitations – although these are generally described as secondary compared with the quality aspect.
Evaluating the tools
When translators were asked to evaluate the various tools at three levels – very usable, usable and not usable – most assessments indicated the middle result, i.e. usable. The common denominator is that none of the tools is considered able to deliver consistent high quality for all languages, types of texts, subject matter and contexts.
In the end, in order to continue to deliver high quality translations it is a question of finding a balance between the fantastic opportunities offered by this technology and the irreplaceable value of human expertise. This would seem to be the key to the industry’s continued development and future.