Does the world still need translators and interpreters?

AI is rapidly changing the work of translators and interpreters. But according to VUB senior lecturer Koen Kerremans, technology will not make human language expertise redundant. On the contrary: the more AI can do, the more important it becomes to assess whether its output is accurate and appropriate for a particular situation. Tomorrow’s language professional will combine strong linguistic and cultural skills with the ability to use technology critically and help steer multilingual communication.

Is there still a place for translators in tomorrow’s labour market?

Koen Kerremans: “Absolutely. There will continue to be a need for translation. Thanks to technology, and more recently generative AI, the possibilities for translating are only increasing. For companies, for example, AI lowers the barrier to providing product information in other languages more quickly and cheaply, making it easier to reach customers abroad. Human translators will remain necessary to assess whether that information is accurate and appropriate for the language and culture of the target audience. “This changes their role. Translators have been working with translation memories, terminology databases and machine translation for some time. Today, translators are increasingly given a pre-translated text to post-edit: checking the output and correcting it where necessary. As a result, the work is shifting partly from translation itself towards checking, revising and verifying. But that still requires a very high level of language proficiency and subject knowledge.”

What makes a human translator better than AI?<
“AI can process a great deal of what is visible in language: words, sentences and increasingly the broader context. But communication is about more than that. There are communicative intentions, registers, cultural differences and specific expectations on the part of a client. These are things that are not always visible to an AI. Cultural and situational knowledge helps a language professional recognise such implicit communication needs.

“An AI translation can be linguistically correct and yet still be unsuitable for the context in which it is to be used. The human translator therefore assesses whether the translation is accurate and whether it is appropriate for the target audience.”

“An AI translation can be linguistically correct and yet still be unsuitable for the context in which it is to be used.”

When does that happen?
“Sometimes you encounter what we call false fluency: a text that reads very smoothly but nevertheless contains substantive errors. This can happen, for example, with specialised terminology. A language model can translate a technical concept very convincingly without the result actually being correct. In such cases, a translator may have more work checking the output than if they had started with a reliable translation memory or a validated terminology database. Producing text more quickly therefore does not automatically mean translating more efficiently.”

What impact is AI having on the work of interpreters?
“Here too, it is not black and white. Interpreters can use AI to prepare for an assignment, for example by gathering information about the subject and relevant technical terms. There is also a great deal of experimentation with speech translation and automatic subtitling, for example in online meetings. In well-defined situations, this can be useful, but in real, complex or high-risk conversations, you cannot simply assume that these systems are reliable enough to replace a human interpreter. “Take, for example, a doctor talking to a patient who speaks another language. If a message is translated incorrectly, the consequences can be serious. Emotions and intentions also play an important role. A patient may not say that something is unclear, but hesitate or give an evasive answer. An interpreter can pick up on such signals and ask for clarification without interpreting for themselves what the patient means. Non-verbal communication also contributes to meaning.

“AI systems have become increasingly good at taking context into account over the years, but not all the relevant information from a conversation and its surrounding situation reaches the system. A fluent automatic translation therefore does not guarantee a successful conversation.”

How reliable are current systems?
“That depends heavily on the language combination and the domain. You cannot say: this is the best AI translation system today. What works well for English–Dutch in a particular field does not necessarily work well for another language combination or another subject. Much depends on the data on which the model was trained. Garbage in, garbage out: if there is a lot of noise or insufficiently high-quality data in a model, you will see that reflected in the output. The same applies to languages. Some languages are much better represented in the training data of AI models than others. English occupies a dominant position in many models. So-called low-resource languages, such as Tigrinya and Pashto, have far less digital language data available. As a result, systems are often less reliable for these languages. Dutch has a relatively large amount of digital language data, but it is less strongly represented than English in many models. There are also differences within a single language. Think, for example, of Belgian Dutch and Dutch as spoken in the Netherlands. If one variety is much more strongly represented in the data, it can also dominate the output. Models can also reproduce cultural or social biases present in their training data.”

How is the degree programme adapting to this rapid technological change?
“Our core mission remains to train students to become highly skilled language users with a profound understanding of culture and communication. Technological skills build on strong language proficiency and cultural knowledge. They do not replace that foundation. Students must also understand how AI systems work, what their limitations are and how to assess their output critically. That is why, for example, our bachelor’s degree in Applied Linguistics includes a course on language and AI, in which students learn what happens under the bonnet of these systems. The master’s programme covers multilingual AI and language technologies. Students also apply this knowledge in translation and interpreting workshops. They learn to justify their choices: why are they using a particular application, and what checks are necessary?”

“Students must also understand how AI systems work, what their limitations are and how to assess their output critically.”

Why should I choose to train as a translator or interpreter today?
“Our students are now being trained as multilingual language professionals who understand multilingual communication, technology and the social context in which language is used. They learn not only to translate and interpret, but also to consider when technology can be used responsibly and how to safeguard the quality of multilingual communication. It is not just ‘human in the loop’, but really ‘human in control’: the human remains in the driving seat. From the 2027–28 academic year, the current master’s programmes in Translation and Interpreting will also be merged into a single Master of Arts in Translation and Interpreting. This reflects professional practice, where translators and interpreters increasingly combine different roles. We are not only preparing students for existing tasks; we are also teaching them to engage critically with new developments.”