How Do Human Translators Utilize AI?

In recent years, human translators have found themselves in what may feel like a tenuous position, as the quality and capabilities of machine translation (MT) engines and generative artificial intelligence (AI) systems have become more advanced and more accessible to both companies and individuals. As we’ve seen across many industries, project directors often view computer automation as a way to optimize time and costs, thereby cutting out the human element of translation. Human Science provides consultations and support for clients who want to explore such options, though our internal research has consistently shown that human oversight is still necessary to ensure translation accuracy, and that machine translation is not yet viable for some document types, such as marketing and other creative writing. In other words, automatic translation or generative AI may not provide a fix-all solution on its own, but they are powerful tools that can greatly aid the translation process when wielded by an experienced professional.

Surveying Translators

With that in mind, we decided to reach out to our list of Japanese to English freelance translators and ask them about their usage of MT and AI tools via a survey. This group included both Japanese and English-speaking natives with many years of technical translation experience, all of whom have passed Human Science’s rigorous translation trial. We wanted to hear about how these translation professionals choose to utilize the latest translation technologies and with what frequency. We were curious what tools the translators found most useful when handling translation work themselves and what their attitudes were towards post editing, the process of editing automatic translation results for accuracy and fluency, as opposed to translating from scratch.

The survey consisted of seven multiple-choice questions and one comment section. The first three questions asked how often translators use MT and AI tools during translation, the purposes for which they are used, and which specific tools the translators use. The latter questions focused on the translators’ current experience and future interest in post-editing. The survey was open for 16 days at the end of 2025 and received a total of 11 responses. While this number of responses is far below what is needed to reliably represent larger trends among all Japanese to English technical translators, nevertheless, some interesting patterns emerged from the results.

MT and AI Usage Among Translators

First, all of the freelance translators who responded to our survey confirmed that they use MT or AI tools for at least some parts of their translation work, but only one individual said they use them on every job they handle. The largest number of respondents (five total) use them “Sometimes,” while two use them “Often” and three “Seldom.” This shows that most modern translators utilize MT and AI tools to some extent, but the vast majority do not feel compelled to rely on these tools for every job.

Regarding which specific services our translators use, there was a clear preference for DeepL (eight responses) followed by Google Translate (six responses). ChatGPT by OpenAI ranked third with three responses, and fourth was Microsoft Translate, with two. Microsoft’s AI assistant, Copilot, was listed but received no responses, nor did any other AI tool. The relative lack of generative AI usage, aside from a minority of ChatGPT users, was somewhat surprising, but when analyzed together with the translators’ reasons for utilizing these tools, more patterns emerged.

In our survey, the listed options for how MT or AI might be used to assist human translation were the following: to generate a baseline translation, to confirm the meaning of the source material, to brainstorm alternate phrases and ideas, and to proofread translations. Of these, a clear majority of respondents (eight total) affirmed that they use MT or AI to brainstorm. Almost half mentioned using an MT engine or AI system to generate an initial translation or to proofread the final translation (five respondents each), and only three people said that they use MT to confirm the meaning of the source text. A space was provided for responder input, but no additional uses were submitted.

Patterns of Use

What makes this information especially interesting is how the tools selected and the frequency of their use intersect with the reasons that translators choose to utilize MT and AI. For example, the three translators who seldom use these services do so almost exclusively for the purpose of brainstorming, with one also using MT to confirm the meaning of some source text. Meanwhile, the one individual who always uses MT uses it solely to generate a baseline from which to work, essentially handling all translation jobs in a manner similar to post editing. Now, while these results alone cannot attest to the skill of these translators or their motivations for using MT and AI the ways that they do, one can extrapolate that the translators who are most confident in their own ability to translate will not rely on generative tools in ways that might impact the final tone of the text, such as generating a translation draft or proofreading. These highly independent translators maintain total control over the phrasing and syntax, rather than trusting the judgement of these systems.

Regarding which services are utilized for each purpose, the neural machine translation engines (Google Translate and DeepL in particular) that predate generative LLMs like ChatGPT are spread across all potential use cases. ChatGPT, however, is only used for brainstorming and proofreading. Our professional translators do not utilize AI for provisional translation and meaning confirmation, tasks that always require accurate understanding and reflection of the meaning of the source text. There are a few reasons why this might be the case. Because AI built on LLMs is still a relatively recent technology, translators may be unaware of the potential that AI has shown as a translation tool or are unfamiliar with how to use it compared to existing MT tools. Another reason could be a sense of distrust for AI results, which have the propensity to hallucinate, compared to neural machine translation engines that were developed specifically for the purpose of translation. Nevertheless, some translators have found value in the LLM’s ability to output variations on a theme and to identify issues in grammar or fluency.

What Does This Mean About Automatic Translation?

If systems that automatically produce translated text were able to reliably do so with the level of fluency and accuracy demanded by users and clients, one would expect that a significant percentage of professional human translators would utilize this function for the majority of their work, to increase their own speed and output. After all, machine translation is near instant compared to the hours or days that traditional translation can take. However, this does not seem to be the case for most translators. Not even half of the translators that we work with utilize this function at all. Instead, they prefer to use MT and AI in more indirect ways, such as brainstorming, to support their own skill and judgement. This further demonstrates what our in-house research has shown: that automatic translation is a valuable support tool with a variety of useful applications but is not a one-stop solution to creating production-ready documents. Human judgement and expertise remain necessary to the process. And while the manner in which human translators choose to incorporate translation technology may vary, most find at least some value in supplementing their translation work with automatic translation and writing tools.

So, if the quality and reliability of automatic translations are not high enough that human translators would choose to incorporate them into the majority of their own work, does this undermine the viability of machine translation and post editing (MTPE) as a translation solution in general? Of course not. Just because the standard of quality that an experienced professional translator holds for their own translations may not be achievable by current automatic translation tools does not invalidate the other benefits of using MT for business. It all depends on your circumstances and priorities. Human translators must uphold a high standard of quality in regards to not only accuracy, but also fluency and localization, in order to maintain their reputations and secure future jobs, but the process takes time. So, when time is a limited resource or budgets are tight, producing a translation through any means is generally better for an expanding business than having none at all, and MTPE can provide a path to near-human quality on a much shorter timeframe.

In the results of our survey, more than 90% of the respondents affirmed their experience in editing documents that were translated by MT, most with some degree of regularity. Only one translator had never performed post editing, and they had no interest in taking such jobs or developing their post editing skills in the future. Most translators seem to acknowledge that MTPE will only become more prevalent as systems improve and build client trust, so they are open to the idea of editing translations that were not human-made. It will be interesting to see if further exposure to MT makes them more or less likely to utilize it in their own translation work.

Conclusion

Our MT/AI survey results revealed that most translators contracted by Human Science for Japanese to English translation work do not view automatic translation output as a suitable substitute for the work they can produce as human translators. However, they do use MT and AI in smaller ways to complement their own skills and expertise, particularly for brainstorming purposes. In practice, DeepL and other neural machine translation engines are clearly favored by our translators, especially for tasks requiring accuracy in meaning, while generative AI has been utilized for more flexible tasks such as brainstorming and proofreading. Ultimately, despite some reluctance to replace their own efforts with these systems, they are willing to use MT and AI as tools to support their work and are increasingly open to accepting automatic translation as an initial step in MTPE tasks.

If you’re looking for a translation service provider or you’re curious whether MT is right for your task, please contact us here for a free consultation.