AI cannot replace the human sense for language
For many staff members at the University of Münster, translating a text “on the fly” from German into English (or another language) – or vice versa – using an AI-based translation tool is part of their daily routine – whether it’s for email correspondence, drafting texts or translating webpages. Automated translation saves a lot of time and is a great help. After all, most people do not have as good a command of a foreign language as they do of their own mother tongue.
Professional translators have mixed feelings about automated translations. The Federal Association of Interpreters and Translators, for example, points out noticeable differences in quality between the various tools and emphasises the need for human oversight when dealing with the translations. Depending on the intended use, errors can have consequences of varying severity. Nevertheless, even for many professionals, these programmes have become indispensable. “There are vast quantities of texts at universities that need to be translated. AI enables us to cope with this deluge,” explains Robert Brambeer, who works for the English Support Office at the University of Münster.
His duties include translating the University’s publications from German into English and proofreading translations. These include, amongst other things, study-related regulations and forms, as well as the pages of the central English-language website. Many of the AI-generated texts that land on his desk are translated accurately. “For texts that mainly convey factual information and data – such as event announcements or short press releases – AI translation works surprisingly well,” says the American. And it saves time. However, he warns of serious errors in content that can sometimes be hidden behind elegant-sounding phrasing – these are the kind of errors that he and the English Support Office regularly spot. It is therefore essential to check all texts that the University makes publicly available. For certain types of text, such as regulations, the use of a translation memory system can be helpful. This is a digital database that searches for identical or similar text fragments stored within it.
Accuracy of content is one thing. Another is style, which Robert Brambeer also keeps an eye on. In his view, the English texts he receives for proofreading have become more monotonous. Plays on words and dual meanings have been lost. Furthermore, English texts translated using AI often sound “stilted”. As examples, he cites compound clauses that the AI translates word for word, but which are not common in English. Even relativising expressions popular in German, such as ‘unter anderem’ (amongst other things) and ‘auch’ (also), do not occur as frequently in the English language. ‘I read these translations and think: That sounds acceptable, but not like it was written by a native speaker. That’s when my work begins.” He prefers to retranslate some texts from scratch because proofreading them would be too time-consuming. When it comes to a feel for the language or where a deeper social and cultural understanding is required, Robert Brambeer recommends human translators and editors. “AI tools can be used effectively for translating certain types of text. But when it comes to literary texts, I would advise against it.”
For many staff members at the University of Münster, translating a text “on the fly” from German into English (or another language) – or vice versa – using an AI-based translation tool is part of their daily routine – whether it’s for email correspondence, drafting texts or translating webpages. Automated translation saves a lot of time and is a great help. After all, most people do not have as good a command of a foreign language as they do of their own mother tongue.
Professional translators have mixed feelings about automated translations. The Federal Association of Interpreters and Translators, for example, points out noticeable differences in quality between the various tools and emphasises the need for human oversight when dealing with the translations. Depending on the intended use, errors can have consequences of varying severity. Nevertheless, even for many professionals, these programmes have become indispensable. “There are vast quantities of texts at universities that need to be translated. AI enables us to cope with this deluge,” explains Robert Brambeer, who works for the English Support Office at the University of Münster.
His duties include translating the University’s publications from German into English and proofreading translations. These include, amongst other things, study-related regulations and forms, as well as the pages of the central English-language website. Many of the AI-generated texts that land on his desk are translated accurately. “For texts that mainly convey factual information and data – such as event announcements or short press releases – AI translation works surprisingly well,” says the American. And it saves time. However, he warns of serious errors in content that can sometimes be hidden behind elegant-sounding phrasing – these are the kind of errors that he and the English Support Office regularly spot. It is therefore essential to check all texts that the University makes publicly available. For certain types of text, such as regulations, the use of a translation memory system can be helpful. This is a digital database that searches for identical or similar text fragments stored within it.
Accuracy of content is one thing. Another is style, which Robert Brambeer also keeps an eye on. In his view, the English texts he receives for proofreading have become more monotonous. Plays on words and dual meanings have been lost. Furthermore, English texts translated using AI often sound “stilted”. As examples, he cites compound clauses that the AI translates word for word, but which are not common in English. Even relativising expressions popular in German, such as ‘unter anderem’ (amongst other things) and ‘auch’ (also), do not occur as frequently in the English language. ‘I read these translations and think: That sounds acceptable, but not like it was written by a native speaker. That’s when my work begins.” He prefers to retranslate some texts from scratch because proofreading them would be too time-consuming. When it comes to a feel for the language or where a deeper social and cultural understanding is required, Robert Brambeer recommends human translators and editors. “AI tools can be used effectively for translating certain types of text. But when it comes to literary texts, I would advise against it.”
Background: The new ‘uniGPT Übersetzung’ (‘uniGPT Translation’) service enables data-secure translations between German and English
With ‘uniGPT Übersetzung’, the University of Münster is providing its students and staff with a tool that translates between German and English. It is currently running in a (preliminary) beta version and is being further improved. Anyone who is interested can give it a try. Jonathan Radas from the Center for Information Technology (CIT) outlines the idea behind the project: “We want to provide everyone at the University with a free tool for data-secure translation of acceptable quality.” The CIT organised the project following an initiative from the Web and Design Department, and the two departments are now carrying it out jointly.
‘uniGPT Übersetzung’ integrates various large language models. Development is taking place in consultation with the English Support Office. It is designed to enable users to comply with the University’s linguistic standards effortlessly. It takes into account the University’s German-English glossary for consistent external communication and is based on the English Style Guide, which recommends British spelling. Data security is also guaranteed. The data is processed on the University’s servers and is not used to train the models, meaning that texts classified as confidentiality levels one to three may be entered. Only data classified as level four – such as subject data from medical studies – is off-limits.
The CIT is counting on as many interested users as possible to use ‘uniGPT Übersetzung’ and provide feedback via the ‘Translation Feedback’ function. CIT staff member Patrick Thiemig explains that, particularly at the outset, it is important to gather experience in order to continuously improve the quality of the translations. He recommends, for example, that users make use of the ‘advanced settings’ field to provide additional instructions. Which academic discipline is involved, and what type of text? What is important for the translation? “You can also ask ‘uniGPT’ for recommendations on how to ensure the translation is as accurate as possible.”
‘uniGPT Übersetzung’ is currently being integrated step by step into the Imperia software. In future, it will be available to all Imperia users within the University network for the automated translation of webpages. Anyone who wishes to do so can test this function upon request (contact: or.support@uni-muenster.de).
Author: Christina Hoppenbrock
This article appeared in the University newspaper “wissen|leben”, issue no. 6, 7 October 2026