23/04/2026

More and more companies are using AI to translate their content quickly and cost-effectively. If an in-house AI solution is used, translations are often merely a ‘by-product‘. This is because the technology is usually designed for other usage scenarios and considered in isolation from existing language processes – which can have serious consequences for quality, consistency and effort. So it’s important that AI is integrated into translation processes right from the start, whilst ensuring that language experts and translators remain at the heart of the process. You can find out exactly how this process works in this blog post.

It is not uncommon for decisions on how to use AI to be made by the IT department or management. Translation is quickly regarded as an additional use case – though without taking the existing translation process into account or involving those responsible for multilingual content. This is because many companies interpret ‘AI-powered translation‘ to mean that the AI system they use can produce a file in the desired foreign language.

Those who bear the brunt of this expectation are often the people within the company responsible for multilingual content. Instead of involving them when designing the process, they are presented with a fait accompli. Translation is thus losing its status as a specialist field in its own right and is increasingly being viewed as a by-product that can be handled as an add-on by the chosen AI tool.

So it‘s hardly surprising that this strategy has not gone down well with the employees concerned. The main concerns include:

• Scepticism regarding quality: Language experts know just how much expertise, time and energy go into professional translations – and how substandard or simply incorrect AI output can sometimes be.
• Loyalty to traditional translation approaches: Many employees in language departments are translators themselves and are reluctant to let AI take over a craft they have often spent decades perfecting.
• Fear of losing one‘s job: Understandably, there is considerable concern about being completely replaced by AI tools. This can undermine trust within the company and, in the worst-case scenario, lead to a brain drain.
• Lack of appreciation: People want to be valued for their work. If people get the impression that machine translation output is considered sufficient, they will be less motivated.

The reasons we‘ve outlined may lead to a resistance to change, which means the AI system is not given a chance. What’s more, many companies assume that they can simply pass on the AI output to branches or employees with foreign language skills for review and approval.

However, this approach is risky as the staff tasked with this work often lack the necessary resources or competence in the source language. Furthermore, the translation is produced as a source-language file in isolation and independently of any tools, meaning that existing review workflows and the integration of resources such as terminology databases no longer function. It‘s only when the poor quality of texts begins to show up in KPIs that decision-makers realise that high-quality translations really do require the input of language experts.

So how can AI be successfully and professionally used in translation, and what do companies need to bear in mind when doing so?

Work involving language data, files and quality checks is complex. Professional translations usually require specialised computer-aided translation tools that separate the content from the file format whilst preserving the formatting. Through the integration of translation memories and terminology, all language resources are utilised, which ensures greater consistency and lower costs. The process is complemented by terminology work, automated quality checks and, in the case of machine translation, post-editing by translators.

Large Language Models can be used as an additional language resource in this process as most LLMs are based on multilingual training material. Existing resources such as translation memories (TMs) can then be used as additional training material for fine-tuning in order to tailor the machine output to the specific needs of the company. However, this does not automatically lead to correct or consistent results.

In practice, it‘s clear that:

• If TM data is to be used for AI training, it usually needs to be cleaned up and curated
• Without integration into computer-aided translation tools, LLMs operate independently of the translation memory Corrections to the AI output are therefore not fed back into the system
• There is no reproducibility, meaning that errors can occur repeatedly or in different forms
• Language data accumulated over many years may be lost if integration is not possible or if the volume is too small to have a significant impact on the system

Large Language Models (LLMs) should be viewed as one of many resources within existing translation processes.

Computer-aided translation (CAT) tools remain the specialised environment for professional translation, particularly when it comes to data handling and process management, whilst AI is integrated as an additional resource. The integration of all available language resources, such as translation memories and terminology, is of crucial importance. It‘s only through the combined use of all resources that consistency and accuracy can be ensured, whilst at the same time reaping the time and cost benefits of AI.

It is crucial that the translation process remains firmly rooted where it can be managed most effectively from a professional and technical perspective – with AI taking on clearly defined tasks within this framework.

The roles can be clearly distinguished as follows:

The tasks of AI:

• Rapid production of draft translations
• Assistance with standard and routine content
• Processing large volumes of text in a short space of time
• Ensuring the use of specified language, e.g. with the help of glossaries

The tasks of language experts and translators:

• Checking and ensuring quality from a content and linguistic standpoint (post-editing)
• Assessment of context, tone and meaning
• Correction of errors, omissions and stylistic discrepancies
• Ensuring consistent corporate language and terminology
• Evaluation of error sources and identification of potential for optimisation

This clearly shows that AI should not replace the translation process, but merely complement it. Language professionals should therefore play an active role in shaping this process. Only with their linguistic and process expertise can AI be meaningfully integrated into existing processes.

AI offers the greatest benefit when used together with existing translation processes, language data, post-editing and linguistic expertise. It should not replace established workflows, but it can support them effectively. When introducing AI for translation work, companies must therefore take their existing translation process into account and use AI to usefully complement human expertise. This allows translations to be produced more quickly without compromising on quality or exasperating the language services department.

Would you like to use LLMs in your business, or do you already have your own system? The experts at oneword would be happy to show you, in a no-obligation consultation, how you can best integrate the tool into your processes or incorporate one of our company-specific LLMs.