Translating with AI

If you do a lot of translation or if you translate complex texts that carry a high level of risk, Microsoft Translator and Copilot are not suitable tools. In these cases, you should use Phrase instead.

If translation is not one of your core tasks, we recommend that you

• Use AU’s Microsoft Translator and Copilot solutions at work rather than free translation tools or chatbots
• Only use Microsoft Translator and Copilot to translate everyday, low-stakes communication
• Leave the following types of texts to language specialists:
  • Long, complex texts
  • Texts with specialist terminology
  • Texts with a high level of risk

You can translate in all Microsoft applications (like Word, PowerPoint and Outlook) by right clicking on the text you want to translate.

Microsoft’s guide to Microsoft Translator

You can access Copilot in different ways – in a browser or by clicking on the Copilot icon in the various Microsoft applications.

AU IT’s guide to Microsoft Copilot

Microsoft's guide to using Copilot for translation

This guide is primarily intended for administrative staff and managers who need to translate everyday communication into English as part of their work at AU. If you find it difficult to express your ideas in English, it can be a good idea to write in Danish first and then use machine translation. This saves time and often results in a text that is more fluent, nuanced and linguistically accurate.

Use Microsoft Translator to translate texts in low-stakes situations: where any mistakes or inaccuracies will not have serious consequences for your own, your unit’s or the university’s reputation.

Examples:

• Emails to international staff members about practical matters
• Minutes from a meeting that took place in Danish
• Slides for an internal workshop or presentation
• Web content with a short shelf life

• Never rely on ‘raw’ machine translation. As the sender, you must always read through and double-check the translation. For this reason, it often makes sense to use machine translation for shorter texts or to translate and post-edit a few sentences at a time.
• Consider aligning expectations with your recipient about your use of machine translation, or add a disclaimer if this is not possible – for example, on a webpage: This text/page was translated automatically and lightly post-edited for clarity. The translation is provided for information purposes only and may not capture the full meaning of the original. 
• Never translate individual words, phrases or sentences without any context. Microsoft Translator needs context to ‘understand’ how to translate words and expressions correctly. Always translate at least two to three sentences, even if you only need a few words.
• Remember that Microsoft Translator is a generic tool. It doesn’t recognise AU-specific terms such as unit names or job titles, nor your own subject-specific terminology. It’s always a good idea to double-check AU terms in the AU Dictionary.

You can also use Copilot to suggest translations for terms that can’t be found in a dictionary. But remember: If it’s important you get the terminology exactly right, you should reach out to a language consultant.

You should only use Microsoft Translator in situations where there isn’t much at stake: that is, where mistakes or inaccuracies won’t have serious consequences or undermine the credibility of your message.

Machine translation should also save you time and make your workday easier, not more complicated – so use Microsoft Translator when it can help you complete tasks more quickly and confidently, especially tasks that aren’t part of your core expertise.

With this in mind, your standard for quality should be ‘good enough’ when you translate texts using machine translation. So what should you focus on to achieve a ‘good enough’ standard? And what is your responsibility?

Fortunately, Microsoft Translator takes care of the grammar and spelling, so that removes one headache. Your job is to ensure that:

• Any AU-specific terminology is correct (job titles, names of units, locations, etc.)
• The translation communicates your message well enough without distorting it
• The translation is useful for the recipient

Pre-editing

If your Danish source text is convoluted, ambiguous or difficult to follow, the machine translation will reflect this. Put simply: nonsense it, nonsense out.

Pre-editing – that is, editing your Danish text before you translate it – allows you to improve the quality of the output by improving the quality of the input. And if you’re translating from your first language into an additional language, pre-editing is your trump card. It’s much easier to identify and correct issues in a text written in your own language than in a translated text, where you risk overlooking errors or introducing new ones.

You can pre-edit your texts proactively – before you run the text through Mictrosoft Translator – or reactively – after you’ve translated the text, if you’re unhappy with the result.

Tips for pre-editing

When you pre-edit your text, focus on whether the text makes sense and has a clear message – not whether it sounds ‘professional’ or ‘academic’.

You should:

• Shorten long sentences and simplify long, complex sentences
• Use active verbs
• Correct spelling errors, grammatical errors and punctuation errors
• Use keywords consistently. If you start out writing about ‘guidelines’, use the same word all the way through. Don’t create unnecessary confusion by using other words, such as ‘rules’ or ‘instructions’, for the same concept.
• Be clear and to the point. Remove filler words such as ‘e.g.’, ‘and so on’ and ‘etc.’ and use simple, everyday language.

Post-editing

Post-editing a machine-translated text has the same goal as pre-editing – to achieve a usable text of good enough quality by removing anything that distorts meaning. When you post-edit, you should focus on spotting and correcting mistakes, not making your text sound good.

How long you spend post-editing your text and what type of mistakes you focus on will depend on the level of quality you wish to achieve. But if you follow our advice and only use machine translation in low-stakes situations, the goal is to produce a text that’s good enough for its purpose – a text with a clear message that can be understood.

When you post-edit a text, you should:

• Change as little as possible
• Focus on content, not style
• Make sure key words are clear and used consistently

If you come across a sentence or a passage in the translation that is unusable and that you cannot correct yourself, you can:

• Go back to the source text, edit it and re-translate
• Ask Copilot for help (see the next section)
• Reach out to a colleague with good language skills or to a language consultant

You can of course use Copilot for translation. If you ask Copilot to ‘translate xx’, you’ll get a text that is almost identical to the output from Microsoft Translator. For most staff members, it will therefore be faster and just as effective to use Microsoft Translator for low-stakes translation tasks that you’re expected to handle yourself as a non-specialist.

You can also prompt Copilot to produce more fluent translations tailored to different target audiences. But if you use Copilot in this way there are a few pitfalls you need to avoid.

1. Copilot hallucinates

Copilot may sometimes add, distort or omit content in a translation. It takes a sharp eye to spot these errors – precisely because the text often sounds convincing and Copilot cannot express uncertainty. Microsoft Translator is much less likely to hallucinate, because it’s optimised for translation and nothing else.

2. Copilot overpromises

Copilot never admits its own limitations. It always delivers its output with great confidence. The fact that you can prompt Copilot to use a specific tone of voice or style is therefore not necessarily an advantage. You can’t be sure that the output will be useful, so you need to read the text carefully. Even then, if English isn’t your first language, you may find it difficult to judge whether the model actually captures a tone and style in English that is both appropriate and convincing.

3. Copilot can waste your time

You often have to prompt Copilot several times before you get a useable result. In most situations where it makes sense for you to do the translation yourself, it’s far more important – and entirely sufficient – to focus your valuable time on ensuring that your message is clear and understandable for the recipient.

4. Copilot consumes energy and water

Copilot and other GPT models use significantly more electricity and water to translate your email or LinkedIn post than Microsoft Translator and similar machine translation tools.

When you translate with Microsoft Translator (or your own brain!), you can use Copilot to improve the quality of your translations in many ways. You can use it to:

Refine and simplify your source text

The clearer and simpler your source text is, the better the translation will be. Copilot can help you refine and simplify your sentences, making them easier for the machine to translate. Here are a few prompts you can try:

• Can you help me make this sentence/text clearer and simpler so it’s easier to translate with Microsoft Translator?
• I need a simplified version of this text that is grammatically correct and easy to understand – without changing the content. The target audience for the translation is international students at a Danish university.

Help you understand the source text before you translate

If you’re unsure what a Danish word or phrase means, you can ask Copilot to help you. In order to assess whether Microsoft Translator has captured your message, it’s important you understand the source text.

Here are some prompts you can try:

• I’m translating a text that includes the word/phrase xx, and I’m not sure what this means. Can you explain this word/phrase to me briefly and clearly? [Insert the word or phrase here, and in include the section and any other relevant contextual details.]
• I can’t make sense of this sentence/expression. Can you help me interpret it? The context is xx. I need to translate the text, so please keep your explanation brief and clear. [Insert the sentence or expression here, and in include the section and any other relevant contextual details.]

Help you translate words and expressions that Microsoft Translator cannot handle

Sometimes Microsoft Translator falls short – especially when it comes to highly subject-specific or AU-specific terms and phrases. In these cases, Copilot can help you find a more precise and appropriate English translation. Here are some prompts you can try:

• I’ve translated a text with Microsoft Translate, but I sense that the term xx has been translated incorrectly or too literally. Is the translation accurate? If not, can you suggest an idiomatic English equivalent that would make sense for my target audience? [Insert the term here, along with the full paragraph and, if relevant, the Microsoft Translate version. Specify your target audience.]
• I’ve translated a text with Microsoft Translate, and I’m not sure whether the technical terms have been translated correctly. Please review the translation for me, focusing on subject-specific terminology. I’d like alternative suggestions if you find any terms that are not accurately translated. [Insert source text and translation, and specify the context and target audience.]

Check your terminology by uploading a term list

If you have a term list with Danish and English domain-specific terms – from your department or project, for example – you can use Copilot to check that Microsoft Translator has translated them correctly. This is particularly useful when working with texts where the terminology needs to be clear and consistent. We've also put together a list of AU-related terms with one-to-one translations (job titles, unit names and the like) that you can use to double-check AU-specific terminology. 

Here’s how to do it:

1. Upload your term list as a Word or Excel file. The term list should contain both the Danish and the English terms. Ideally, the list should consist of one-to-one translations: that is, Danish terms for which there is one approved translation.
2. Upload the text you have translated with Microsoft Translate.
3. Ask Copilot to check that the translation follows the term list and to highlight any deviations that might need closer review. Here are some prompts you can use:

I have translated this text using Microsoft Translate and would like to check whether the terminology matches my term list, which contains [xx] terms. Can you review the text and suggest corrections if any terms have been translated incorrectly? [upload term list and translated text, specify context and target audience]

Here is a term list containing Danish and English terms. Can you use it to check whether the translation of this text is terminologically correct and consistent? [upload glossary and translated text, specify context and target audience]