AI for translations – a tempting solution that requires thought
AI for translations – a tempting solution that requires thought
More and more companies are considering AI for their translations. That’s not surprising. The speed is impressive and the price for the translation step itself can drop significantly. But there are also risks with AI translation that are easy to miss, and there are many more steps in the process than just the translation.
Recent years have seen the launch of various AI services that companies can use for translating texts themselves. However, without the right expertise, it’s better to choose a proficient translation partner, and this applies to both traditional translation and AI translation.
This post discusses what you should have in place for your AI translations to be cost-effective, provide strategic benefit and not give rise to risks and costly surprises, whether you translate in-house or via a partner. And it’s about what Språkbolaget has to offer as a partner in this regard.
AI translations require good source texts and careful post-editing
It’s easy to focus on the process when talking about AI translation, but the process matters little if the source text is flawed.
With their years of experience, traditional human translators spot flaws in the source text, and they report them and correct them. AI translation places much higher demands on the source text as errors and ambiguities tend to be amplified, and AI does not hesitate to guess the answer if something is not crystal clear. An ambiguous sentence in Swedish can become even more so in English, German or French, thereby multiplying the time and cost taken to fix the errors.
So you should review source texts intended for AI translation especially carefully, or let Språkbolaget take care of it for you.
- Is the terminology consistent?
- Is there an in-house glossary, and has it been made available to the AI translation step?
- Is there a style guide, and has it been applied?
Needless to say, AI translations must be reviewed by experienced human translators in a process known as post-editing. Be aware that AI tends to create unexpected errors that are presented with perfect syntax and are thus very difficult to find, unless you know what you’re looking for. Språkbolaget does.
The Swedish Institute for Language and Folklore (Isof) states that a good quality source text is absolutely essential for a good translation, and that a source language terminology list with definitions is one of the most effective tools you can give a translator, whether human or AI-based.
In a small business, this might involve documenting ten to fifteen key terms. For a larger company with multiple departments, a more structured approach will be necessary, preferably with the support of a terminologist. The investment pays off in lower overall costs and fewer review rounds later.
AI translations and data security – what applies?
Whether they manage translations themselves or outsource the task to a language service provider, many companies underestimate the security aspects when they begin using AI for their translations.
The most important thing to bear in mind is that in the case of AI translations, the texts are sent to a third party – the AI service.
- Where are the AI service's servers located – in the EU, the USA, or somewhere else?
- Is the transfer to and from the AI service's server encrypted end-to-end?
- Is the AI service certified according to ISO 27001:2022, the information security standard?
- Is the AI service paid or free of charge? In the latter case, your texts will without a doubt be used as input for the AI service's Large Language Model (LLM).
Many popular AI services process data on servers outside the EU, and with terms that are not always transparent. This can be either utterly unproblematic or a clear business risk, depending on what you upload. A product description is one thing, a quarterly report or an agreement with a strategic partner is something entirely different.
The level of risk is one of the most important factors when choosing a translation process. A good way to gain an overview is to categorise the company's texts according to how sensitive the information is. Most businesses can sort their texts into three levels:
Open information – Website copy, product descriptions and certain press releases. Low risk. Often suitable for AI translation.
Internal information – Process descriptions, training material, communications with partners. Medium risk. Review the AI service's guarantees and security levels carefully.
Confidential information – Contracts, pricing strategies, customer data. High risk. Should not be processed using AI solutions, but by certified suppliers who engage human translators and apply confidentiality agreements.
The Swedish Agency for Digital Government (Digg) is clear on this point: if an employee uses an AI service in their work and processes personal data, the employer may become the data controller, even if no agreement has been concluded with the supplier. The responsibility lies with your organisation, not with the AI service. This applies to organisations of all sizes.
If you’re unsure about which documents you can translate using AI, get in touch with Språkbolaget. We know our stuff!
Preserving your brand voice in AI translations
AI translations provide texts that are grammatically correct and fully comprehensible. But that’s not the same thing as texts that sound like you.
Every brand has a tone of voice, its own way of communicating which helps customers recognise the company and trust it. AI tends to flatten that tone. The result is a neutral, correct text that functions, but which does not speak with your company's voice.
This becomes very clear when the context changes. For a local company, it can mean the personal tone that makes you unique in your domestic market is lost when you begin communicating in English. For a national company, it’s about maintaining a consistent language internally and externally even when the volume increases. For an international company, the challenge is in making sure that its tone of voice is consistent and recognisable across all markets, rather than simply having texts translated into their respective languages.
These are all aspects that even advanced AI tools cannot handle. They require human translation, or AI-assisted translation with stringent human post-editing to check consistency and assure quality.
Professional translators use their own or clients' databases (known as translation memories) to reuse approved text, and term bases to ensure your internal terminology is applied. Professional language specialists make sure the end result actually sounds like you. Combining the speed of AI services with human expertise delivers both volume and quality. Finding the right balance and the most effective collaboration between human and AI translation to achieve the best value for money, requires knowledge and experience, which is where Språkbolaget can help.
Three questions before choosing AI translation
Before deciding on an AI translation process, it’s worth pausing for a moment. Usually, these three questions will quickly reveal which solution actually suits your business best.
- How does the text look right now; is it completely finished and without any ambiguities?
- Which of the company’s texts contain information that must not be made public?
- How important is it to preserve the company’s voice and tone in the translated text?
The answers to these questions will usually determine the translation strategy you should choose.
Discuss your company's translation strategy with us!
At Språkbolaget we combine advanced translation technology with human expertise and many years of experience. We help assess what’s suitable for your texts, your industry and your data security requirements, whether yours is a small company taking its first step into a new market or an established business looking to streamline an existing workflow.