Technology and translation – how AI is having an impact

There is great and well-founded enthusiasm surrounding machine translation among both the language industry and its clients. Machine translation can in certain cases provide undeniable benefits when it comes to efficiency, but it must be handled with understanding, great expertise and a good knowledge of both the process and the technology – in combination with post-editing and thorough review. In this post, we take a closer look at AI and machine translations, and what to keep in mind for your translation projects.

AI and translations

In recent years, advances in AI have contributed to machine translation becoming increasingly common. Thanks to neural networks and large language models, AI translation can now be surprisingly good, especially when it comes to simpler texts and common language combinations. Many companies and organisations that want to streamline translation work and reach a global audience more quickly are therefore making increasing use of AI translations.

However, there are certain limitations and risks to take into account, particularly when it comes to nuances, tone and subject-specific terminology. The most reliable solution is therefore to combine modern technology with human expertise in order to achieve high quality and to ensure you feel confident with the translations.

Common pitfalls of machine translation

Although machine translation has improved significantly through AI and neural networks, the technology still carries risks. One of the primary challenges is so-called ”hallucinations”. A hallucination means that something has been added or disappeared in the translation compared with the original text. Because the text is largely correct in terms of grammar and spelling, these errors can be very difficult to spot.

Some common risks associated with AI-based translation are that

• terms are added, misused or disappear entirely
• words like ”not” are added or omitted
• or i ental is translated as plural or vice versa
• numerical data becomes incorrect, for example by a decimal separator changing to a thousand separator
• minus sign disappears
• currencies or units are confused.

These mistakes can lead to serious consequences and, in a worst-case scenario, damage your company's reputation. Even the most advanced machine translation can suffer from hallucinations, making human quality control the key to a safe and accurate translation.

How does Språkbolaget view AI?

We at Språkbolaget have naturally followed the development of AI closely. For the projects we work on, human expertise is fundamental to ensuring that the style and content are correct. This applies just as much to lively social media copy as to annual reports or medical texts. In some cases, however, machine translation combined with human translators can be an option, for example if you have large volumes of text that need to be translated quickly so you can get an understanding of the content – especially if the text is to be used internally within the company. To easily analyse what is appropriate, we have a risk assessment matrix.

Risk matrix for translations

Horizontal axis – potential consequences of a poor translation

Here on the horizontal axis we have placed the potential consequences of a translation that does not reach the right level of quality. It could be about 

• disruptions in internal communication within a company – disruptions that can be resolved with a quick phone call
• damaged reputation among the target audience – something that can take time to repair
• difficulties for the end-user in handling a product – this can lower the product's reviews and sharply reduce sales
• financial loss – may be due to poor wording in an agreement – the damage can be unlimited
• a fatality – simply cannot be allowed to happen. 

Vertical axis – potential probability of something occurring

On the vertical axis lies the probability of something on the horizontal axis occurring in the event of a translation flaw – the axis ranges from very low to very high probability.

A translation project that ends up in the matrix green the bottom-left part is certainly important, but perhaps no more critical than management's weekly newsletter to employees in another language area. A dubious phrasing here can lead to raised eyebrows around the coffee machine and perhaps a conversation to get something explained, but it stops at that. 

If the project is somewhere in it sugar In the case of the area, it may be a question of a quarterly report, where an unclear explanation of the company's future plans makes investors pull in their horns, or an instruction manual where an incorrect choice of term makes it impossible for the user to understand how the product is to be used.

A project in the red the area could be a contract where a missing ”not” or an incorrect currency designation could entail enormous costs for one of the parties – or the safety instructions for a heavy machine tool that could run amok and risk the life of the operator who has read but received an incorrect instruction. 

Green projects – low risk

In green projects, machine translation that has been post-edited to human quality can be entirely sufficient. Linguistically correct, functionally sound in content, but it does not need to be a literary masterpiece.

Amber and red projects – high to very high risk

In yellow and red projects, machine translation can possibly serve a function early in the process, but here the utterly dominant part of the work is about editing, re-editing, letting the text rest and editing again. Another pair or two of eyes go through the text with hawk-like precision, double-checking all terms, weighing every phrasing on a gold scale.

We help you make the right choice

AI and machine translation can offer significant advantages in terms of time and cost. At the same time, it is important to remember that nothing beats the linguistic finesse and contextual understanding of an experienced translator. At Språkbolaget, we work primarily with critical texts in the yellow and red zones, such as contracts, annual reports, user manuals for medical technology, high-end marketing, and the like. But every project has its own character and not everyone needs the same amount of human touch.

If you are thinking of using AI in your translation projects, we recommend that you first consult us at Språkbolaget. We can help you to

• assess the complexity and risks of the text
• choosing the right process for your text type and industry
• minimise the risk of errors, which protects your company's reputation.

In this way you avoid unnecessary mistakes and secure a translation that conveys what you want to say – in the right way and in the right language.