AI and machine translation can save both time and money, but they also entail risks

AI and machine translation – opportunities, risks and best practice

Artificial intelligence (AI) is present in most industries today, and the field of translation is no exception. Machine translation (MT) has very quickly gone from delivering substandard results to forming an important part of the translation strategy of many companies.

Used appropriately, AI can save both time and money. But because it also comes with financial, linguistic and confidentiality-related risks, it’s important to know when machine translation will work, and when human expertise is crucial.

What is machine translation?

Machine Translation (MT) is the application of artificial intelligence (AI) to translate text from one language into another using large language models (LLMs) and vast amounts of training data. We can see MT at work when we type a sentence into an online translation tool and get a finished text back.

At first glance, the result can often look clean and flawless, but it’s important for us to realise that AI itself does not ”understand” the text. It doesn't know the context, it cannot decide what tone or feel will be appropriate, and it cannot judge whether a translation is misleading. It can only mimic understanding based on patterns in previous data.

Raw MT – i.e. direct machine translation without editing – can work in private contexts or when you want to quickly grasp the gist of a text. But in a professional setting, it’s risky. What’s more, free online tools involve sharing your data, which in the worst case may be saved or disseminated. Thus free of charge does not mean it comes at no cost to you.

MTPE – machine translation post-editing

For professional use, raw machine translation is not enough. The text must always be reviewed and improved by a human. This is known as Machine Translation Post-Editing (MTPE).

Here’s how it works:

Step 1. AI does the heavy lifting
The text is translated quickly and cheaply by an MT engine.

Step 2. A translator takes over
A professional translator reviews the text, sentence by sentence. This usually takes place in a CAT tool (Computer Aided Translation). Errors are corrected here, nuances fine-tuned, and tone and terminology tailored to the customer's wishes.

Step 3. Quality control
Once the text has been edited, a systematic check is carried out to ensure spelling, grammar, numerals and units are correct, and that the text is consistent and follows established term lists.
Used properly, MTPE can save both time and money. In the case of translating simpler texts such as data sheets or uncomplicated user manuals between two major languages, the cost can be reduced by around 20% compared to traditional translation. However, in texts where style, tone or accuracy are crucial, post-editing can be more time-consuming – and sometimes even more expensive – than translating from scratch.

MTPE compared to traditional translation

In traditional translation, experienced translators – whose native language is the target language – take care of the entire text from start to finish. This means the result is not only linguistically correct, but also naturally tailored to the target audience and context. Språkbolaget has always worked this way, and our experience of using this method has been positive; the results are very good and maintain top quality.

In the case of MTPE, AI generates the first draft, and the translator – whose native language is again the target language – then steps in to edit and correct the text. This can be effective for simpler types of text such as basic user manuals, data sheets or internal documents, especially between major languages where there are large volumes of training data for the AI to draw upon.

It’s important to know that MTPE does not work equally well for all language pairs. If the volume of training data is small or the subject matter very complex, an AI translation often becomes so flawed that the translator must in effect rewrite almost the entire text. When this happens, the time and cost benefits of MTPE are lost.

Språkbolaget also has extensive experience in this type of translation, and we always provide a recommendation on whether the text or language pair in question is suitable for MTPE.

When it comes to texts that demand high standards of accuracy, style and trust such as annual reports, sustainability reports, legal documents, marketing material or certain types of user manuals, traditional translation is the only option. Here, the choice of every word, nuance and turn of phrase must be flawless.

Benefits and risks of MTPE

Machine translation post-editing (MTPE) can provide several benefits when used correctly:

✔️ Faster process – the translation of certain types of text is quicker
✔️ Lower costs especially when dealing with large volumes of text

However, there are also important risks, both with MTPE and with raw machine translation:

✔️ Lack of context – AI often translates sentence by sentence, causing a text to lack a logical thread
✔️ Inconsistent terminology – the same term can be translated differently in different parts of the text
✔️ Hallucinations – AI can insert completely irrelevant words or skip important information
✔️ Apparent perfection – a text may look correct superficially but contain hidden errors, making MTPE time-consuming
✔️ Confidentiality risks – machine translation almost always takes place in the cloud, which means data can be disseminated. This poses a major risk for sensitive material.

The roles of the translator and project manager

Today’s translators need so much more than a feel for the language. In addition to expertise in the target language, translators must also possess knowledge of the subject matter, experience in working with CAT tools, and have the ability to handle databases, term lists and quality assurance measures. 

In order to deliver a text that maintains the right register, a professional translator must also judge when machine translation can be an aid or when it becomes a risk.

Språkbolaget’s role is to make sure all these elements work together. We provide you with guidance by analysing your needs as our customer, recommending the right method and putting together a team of translators and reviewers that matches your specific project. Our project managers supervise the workflow, quality assurance and deadlines and take care of any unforeseen events. So you can be confident that regardless of document type, your translation will be accurate, consistent and delivered on time.

When will a particular method work best?

The factors that come into play are quality, time and cost.

Classic translation
Essential when a text must be completely accurate and build trust, and when the requirement for confidentiality is high. Used for legal, medical and financial documents; annual reports and sustainability reports, marketing texts and user manuals with high safety requirements, as well as other texts where the human touch is crucial.

MTPE – machine translation post-editing
Combines AI and human review. Suitable for simpler texts where time and cost are important, such as certain instruction manuals, data sheets, price lists and internal documents.

Raw machine translation (MT)
Provides a rough translation quickly without editing. Can be used for private use or for an overview when quality and confidentiality are not important.

AI or human? Let us help you choose the right method

While machine translation and AI are powerful tools, they’re not a one-size-fits-all solution. With the right strategy, they can help make projects faster and cheaper, but they can never replace human linguistic expertise.

Unsure which method is suitable for your next project? Contact us Språkbolaget and let us help you find the right solution, the right workflow and the right balance between AI and human expertise.