Recommendations for the translation with AI
Translations from the Language Service for University Administration (ZUV)
The FAU Language Service is responsible for translating the content of the fau.eu website and for translating documents related to studying and administration at the University into English. We translate using a professional computer-assisted translation (CAT) tool and have been storing all our translations in an internal database since 2012. This enables us to work extremely productively while ensuring our translations meet FAU’s high standards of quality. Specialist terminology and FAU terminology is thus translated consistently and inserted automatically.
The Language Service translates documents free of charge for University Administration and can also translate documents for faculty administration and institutions free of charge in certain circumstances. Please feel free to get in touch.
Which AI system does the Language Service recommend for translations into English?
DeepL uses neural machine translation or NMT and specializes in translation. In our experience, this tool currently provides the most convincing translation results at a quality that is satisfactory for several purposes. However, there are several risks and aspects of the tool that are problematic and which users must be aware of, which is why you must always check the translation results it provides very carefully (post-editing).
Please note that the Language Service does not check the quality of texts translated using machine translation tools.
Are you considering using AI for a translation? Please ensure you check the following aspects:
Confidentiality and data protection: Are you able to comply with data protection requirements for sensitive data?
The free version of DeepL stores any texts you enter and uses them to train its system. You should keep this in mind, especially when it comes to unpublished research results and confidential documents. You are not permitted to enter personal data. Even in DeepL Pro, personal data may only be entered if a data processing agreement is in place. [1]
Do you have the language skills necessary to critically assess the output generated by AI and to find any incorrect translations and omissions (see examples below)?
Even though the performance of machine translation tools has improved considerably during the last few years, AI continues to make errors in certain circumstances that may be difficult to detect at a superficial level. Typically, these errors include missing negations, omissions or incorrect translations.
Do you have the time to check all the specialist terminology in the document and adapt it to conform with FAU standard terminology if necessary?
AI tools often translate the same term in German with different terms in English within the same document. Consistency with terminology is extremely important in academic texts in particular, both to ensure they are comprehensible and conform to professional standards. It is also important that FAU standard terminology is used (e.g. for “Studiengang” in German: degree program instead of study program). The Language Service Lookup database contains standardized FAU terminology.
Is making your text sound “natural” and idiomatic not that important to you?
Depending on the purpose of the text and your expectations regarding quality and style, a machine translation can often suffice, for example for internal communication purposes. However, if you require a text with perfect style that will be published for a wider audience, a machine translation will probably not be good enough. AI translates texts very literally, which is not idiomatic in some contexts or makes the style unnatural (see examples below).
4. Tips for writing in English (for non-native speakers)
Tools such as DeepL Write or even ChatGPT can provide assistance if you need to write texts in English:
- The AI finds errors and improves the style or adapts it to match the requirements you have put in the prompt.
- Ideally, you should only use these tools for individual sentences and not for entire texts.
If you have any questions or are interested in taking part in a DeepL training course, you can contact us at sz-sprachendienst@fau.de.
Examples of problematic results of AI-generated translations
(From DeepL, on May 15, 2024)
| Original | KI-Übersetzung | Anmerkung |
| Die naturwissenschaftliche Fakultät besetzt eine Professur für XY im Beamtenverhältnis auf Zeit für die Dauer von 6 Jahren. | The Faculty of Natural Sciences is filling a professorship for XY as a temporary civil servant for a period of 6 years |
Terminologiefehler (aufgrund von fehlendem Glossar): an der FAU lautet der korrekte Name Faculty of Sciences Kontextfehler civil servant: Die Verbeamtung ist ggf. unzutreffend für internationale Bewerber*innen |
| Eignungsfeststellungsverfahren | Aptitude assessment procedure | Terminologiefehler: Qualification assessment process |
| Das Tenure-Komitee unterbreitet anschließend dem Präsidenten einen begründeten Entscheidungsvorschlag. | The Tenure Committee then submits a reasoned decision proposal to the President | Zu wörtlich übersetzt: reasoned decision proposal ist nicht idiomatisch. Besser: The final decision is made by the President based on a detailed proposal put forward by the tenure committee o.Ä. |
| Die Aufnahme einer großen Zahl von Geflüchteten zeigt, dass die gesellschaftliche Einbindung und soziale Teilhabe von Neuankommenden eine kontinuierliche Aufgabe für die Gesellschaft sind. | The admission of a large number of refugees shows that the social integration and social participation of new arrivals is an ongoing task for society. | Nominalstil. Idiomatischer und stilistisch besser wäre eine verbale Konstruktion. Zudem Tempusfehler: Im Englischen sollte hier Present Perfect verwendet werden, obwohl im Deutschen Präsens steht: Accepting large numbers of refugees has shown that helping new arrivals integrate into and participate in society is an ongoing task for society. |
|
Aus Helferkreisen entstehen häufig Patenschaftsmodelle, zum Teil auch Bekanntschaften und Freundschaften. |
Sponsorship models often develop from volunteer circles, and in some cases also acquaintances and friendships. | Fehlübersetzung: In diesem Kontext ist die wörtliche Übersetzung von „Patenschaftsmodell“ falsch und unverständlich. Volunteer circle ist nicht idiomatisch. Hier ist eine sinngemäße Umschreibung notwendig, z.B.: What starts out as a group of helpers often leads to a closer and more caring relationship, with those involved sometimes becoming good acquaintances or even friends. |
| Änderungen sind im Rahmen des Einstellungsprozesses nicht mehr möglich. | Changes are no longer possible as part of the recruitment process | Unklar. Besser verbal: Changes are no longer possible once the recruitment process has started. |
| Liebe:r Studierende:r | Dear:r students:r | Umgang mit Genderzeichen ist generell schwierig. Genderstern funktioniert meist |
| Seitz&Braun | Seitz&Brown | Vorsicht bei Eigennamen! |
| Ich bin flexibel |
I am flexible |
Kontextfehler: Terminplaner-Auswahlliste My schedule is flexible |