Quo vadis, human?

By Maria Wolf
Translated by Richard Peters

Development work on AI-based language models is advancing at an exponential rate. The fourth generation of ChatGPT can write computer code, create images, complete text-based tasks, invent stories, write poetry, and even crack jokes; it writes, translates, learns from user feedback, and adapts its output to the style and taste of its clients. Speech models are becoming more and more powerful, and their output is becoming ever less distinguishable from human speech. Will companies still have any need at all for copywriters, translators, and interpreters in the future?

ChatGPT itself responds to this question by stating that AI-based language models “still struggle with nuance, cultural aspects, and contextual meanings – despite impressive advances – so human translators often remain superior in terms of high-quality and accurate translations.” It also states that there’s little chance a replacement will appear anytime soon for the combination of linguistic, technical, and cultural competence – and indeed empathy – that human interpreters bring to the table.

These statements from ChatGPT sound comforting. It must be noted, however, that the ChatGPT chat assistant I asked, which was released in December 2022, remains at the knowledge level of 2021 – and development work hasn’t exactly been standing still in this area. With increasing amounts of more recent training data, as well as a growing volume of feedback, language models are becoming more powerful – as ChatGPT-4 already demonstrates. So we language professionals aren’t in the clear yet.

AI will change our job profiles, there’s no question about it. And that process has long since got underway. There seem to be no limits to AI-based systems’ capacity for learning, and this raises many questions regarding safety, ethics, and law. The challenge for everyone, and particularly for the language services profession, will be to employ technological advances – especially in AI – sensibly, responsibly, and in a way that makes the most of the advantages on offer. Anyone using AI-based language models must ensure any content is truthful and must check any translations for accuracy. The key here is not to be blinded by a pleasing flow of language.

Countless examples of sometimes funny, sometimes frightening translation errors committed by translation programs such as Google Translate or DeepL show just how essential it is (and indeed always has been) to have a meticulous human reviewer in the loop – especially when it comes to topics of a sensitive nature, questions of safety and security, or when a company’s reputation is at stake. In translation, assessing whether a word or phrase accurately conveys meaning, is culturally the most appropriate rendering, or best serves the author’s intention relies not just on experience and on linguistic, technical, and cultural skills – incidentally, in both the source and target language – but also calls for coordination with the customer as well as a deeper understanding of their objectives, products, and target audiences. This is the only way to create precise, high-quality communication. This is easier to work out from person to person than from person to machine.

What’s more, when we review and edit machine translations, which is known as postediting, we often find that while the output from the better translation programs is thorough and – apart from a few (often very subtle) mistakes – comprehensible, there’s always plenty of scope for us to make the content more attractive and engaging. In the worst case, the damage done by not going to the effort of carefully reviewing a text becomes apparent only after the event. Good advice in advance from a professional pays off and may obviate embarrassing rework.

Moreover, not all copyright and data protection issues of AI-based tools have been fully, reliably, and transparently clarified yet.

Expert language service providers can offer companies added value and peace of mind, especially if they are approachable and keen to speak to customers in person. The best way to work out precisely which services are required, and to what extent, is by talking it through with someone. Personally, for all the efficiency that modern technologies and AI bring to our lives, I’m happy when I’m dealing with people and don’t have to communicate with machines. A person who tells me, “I see what you mean,” even if I haven’t necessarily put my question to them in the clearest terms, is far preferable to some automated solution that presents me with a list of standard problems and obliges me to choose one of them by pressing a certain key, then leaves me hoping it will put me through to a competent person at some point. It’s so much more efficient to find the right person right away, receive good advice, and enjoy excellent service – from person to person.