Guidance on the use of Machine Translation

Machine translation is the process of using a system or software to automatically translate text from one language to another without a human translator. Microsoft Translate or Google Translate are examples of machine translation resources.

University staff have access to Microsoft Translate which is integrated into Microsoft Office programmes (e.g. Word). Microsoft Translate (in Welsh) was originally developed through a partnership between the then Welsh Assembly and Microsoft over a decade ago, and the standard has improved over time. The University's Translation Unit uses it as a resource, and Microsoft's confidentiality and data protection arrangements are robust (see 'Confidentiality' below). Microsoft Translate is the package that we would recommend for staff to use for their work at the University rather than other resources such as Google Translate or AI resources such as ChatGPT.

It can be an extremely useful tool to get a quick rough translation automatically. Many factors influence how successful the machine translation’s attempt is, including the context and wording of the original text. Like any resource such as artificial intelligence (AI), automatic translations must be proof-read very carefully. We intend to publish guidance for Welsh speaking staff by the summer of 2026 on how to edit MT translations effectively, and some of the main errors seen within machine translations.

In general, the most successful machine translations are created if there is enough context and if the language is neutral or corporate/official. Machine translations often struggle with signs for example because there is not enough context.

From our experience, artificial intelligence resources such as CoPilot or ChatGPT are no better than Microsoft Translate. These translations tend to be of a lower standard in our experience and may contain unintelligible words. This happens often with AI, and it is known as a 'hallucination'. Sending translations made using AI to the translation team to proof-read does not save us time. It is usually easier for the translation team to receive the original text to translate from scratch using our 'translation memory' software which includes Microsoft Translate. You can use AI programmes such as CoPilot in Welsh as part of your work in accordance with the University's guidelines, but when it comes to translation, Microsoft Translate is of better quality.

Staff who are Welsh speakers (with Welsh language skills at levels around B2+) Machine translation could be used to facilitate the drafting of bilingual content e.g.

• Internal emails.
• Responding to queries.
• Instructions or guidelines for staff.
• Content for the website.
• Internal newsletters or blogs.

In accordance with data protection regulations, personal details should not be included (see the 'Confidentiality' section below) and the translation must be proof-read carefully. Invariably, there will be sentences that will need to be edited with any machine translation.

The nature of the text also greatly influences how successful the translation is e.g. it does not produce translations as well with creative pieces or pieces that use natural language. For internal correspondence, newsletters, blogs, forms or web pages for example, MT can be a useful resource for those who work bilingually and can confidently proof-read in Welsh. Care will be required with regards to the names of departments/projects etc.

Staff with some Welsh language skills (e.g. A1-B1)

It could be used to get an idea of the Welsh text if you are unsure or to help you if you are drafting content in Welsh. You should ensure that the text has been proof-read carefully (see above)

Non-Welsh speaking staff (e.g. A0)

It could be used to get an idea of the Welsh text e.g. to identify which department/service/staff to direct an enquiry to. MT should not be used to respond in Welsh if you are not confident to proof-read the translations.

It should not be used with confidential text or text containing personal, sensitive, or special data. Similarly we would not advise using MT when translating student assessments and exams.

It should not be used to write and publish Welsh text if you are not confident in proof-reading and editing the text yourself.

The University has a translation unit which provides a translation and proof-reading service to the University's departments.

Feel free to send us text to proof-read but do not send machine translations if a Welsh speaker has not read the translation first. The translation team also has access to MT, and it does not save time to submit a 'proof-reading' request unless a Welsh speaker has already looked at the text.

Do not send translations from AI resources such as Microsoft CoPilot or ChatGPT for proof-reading, these translations tend to be of lower quality in our experience and can contain unintelligible words. This happens often with AI, and it is known as a 'hallucination’.

Microsoft Translate is integrated into a number of Microsoft programmes such as Word or Outlook.

To use MT within these programmes (with your university account) go to

Review > Language > Translate > Choose to ‘Translate Selected Text’ or 'Translate Document'

You will need to select the languages before clicking on 'Translate’.

Microsoft's website states:

"Customer data submitted for translation to Azure Cognitive Services Translator (both standard and custom models), Speech service, the Microsoft Translator Speech API, and the text translation features in Microsoft Office products are not written to persistent storage. There will be no record of the submitted text or voice, or any portion thereof, in any Microsoft data center. The audio and text will not be used for training purposes either." Microsoft

Microsoft's data protection arrangements are very robust. If you have any questions regarding using Microsoft Translate in a specific context, please contact infogovernance@aber.ac.uk for further advice.

We intend to publish a guide for Welsh speaking staff on the successful use of MT by the summer of 2026. This guide will identify issues to bear in mind when using machine translation and will provide an overview of some of the most common errors. The guide will provide specific examples and advice on how to create quality translations using MT.