Copywriting
Artificial intelligence … It has become a hot topic in recent years, glowing hot even. Everyone is talking about artificial intelligence (AI). In the media, at school or at work, it cannot be missed. But are we always talking about the same thing when we talk about AI? And above all, will AI soon take over our jobs, or are we expecting too much? Following a report in Terzake¹, the Untranslate team asked itself these questions. In this article, we look for the answer and give our take on AI in the language industry.
AI, a recent discipline … or not?
AI is gaining considerable momentum. In recent years, it has become indispensable in the daily lives of students, professionals and researchers. Yet it is not a new discipline. The term ‘artificial intelligence’ was first introduced by John McCarthy at a 1956 conference attended by researchers from various fields. That conference and previous successful experiments with different forms of artificial intelligence, such as the Georgetown-IBM experiment, sparked interest and investment in AI research.
From the very beginning, expectations for the knowledge and ability of artificial intelligence were high. For example, it was claimed that perfect machine translation would be completely possible within three to five years. Ten years later, that did not turn out to be the case. Because of these sky-high expectations that couldn’t be met – not just for machine translation, but for artificial intelligence in general – interest and investment in AI research declined sharply. A real ‘AI winter’ arrived, lasting from the 1970s to the end of the 1980s.
What is AI?
In the introduction to this article, we questioned whether we are always talking about the same thing when we talk about ‘AI’. In the paragraphs that followed, we ourselves were guilty of using the term as a general denominator to name all kinds of technologies. Time to rectify that.
In general, the term ‘artificial intelligence’ refers to all technologies that try to mimic human thinking and learning². Whereas traditional computer programmes follow a fixed script to perform a specific task, AI systems use machine learning, learning from data containing relevant examples and recognising patterns in it. In other words, traditional computer programmes blindly and meticulously follow a roadmap, over and over again. AI, on the other hand, processes data and learns in the meantime to improve itself.
AI in the translation industry
AI is thus as complex as the human thinking it tries to emulate and has a wide range of applications by now. Hence, using the term as an umbrella term can be misleading and cause disappointment. It is therefore important to know which form of AI is best suited for particular tasks. Here are some examples of specific forms and AI applications used daily in the translation industry:
- Neural machine translation: Neural translation machines are trained on datasets via machine learning to translate texts. The machine uses neural networks to make a prediction of the correct translation. To do so, the machine looks at the patterns in the data, which may be stronger or weaker, and determines the most probable word order.
- Generative AI: Unlike neural machine translation, generative AI does not make predictions based on patterns, but creates new content (texts, images, etc.) based on the training data. Our copywriters and prompt engineers use generative AI tools as intelligent assistants for brainstorming or quick inspiration.
- Speech recognition software: Natural language processing (NLP) and neural networks make it possible to convert speech to text based on algorithms. For example, our captioners use speech recognition software to quickly get a rough transcription of spoken text.
AI, a friend or a foe?
Will AI soon take over our jobs, or are we expecting too much? It would be naive to say that new technologies and developments in artificial intelligence will not affect the language industry at all. The 2024³ European Language Industry Survey (ELIS) report shows that dealing with AI and machine translation is seen as one of the biggest challenges for language service providers.
At Untranslate, we are convinced that AI tools in all their forms are neither a friend nor foe, but tools. This is why Untranslate’s project managers always analyse your application thoroughly to come to an informed decision on which tools are useful and which are best avoided. Below, we give a brief overview of some strengths and pitfalls of AI tools.
Why use AI in language projects?
Stubbornly refusing to use AI as a tool may result in your company falling behind. Truth compels us to admit that computers simply outperform humans in some areas:
- Speed and efficiency: Machine translation makes it possible to quickly obtain understandable translations in many languages. Speech recognition, for example, enables smooth subtitling of videos. Generative AI can generate countless ideas. In short, AI tools can process a lot of information in no time.
- Cost savings: A limited budget? Because AI is so fast and efficient, it can perform some time-consuming tasks at a lower cost.
- Repetitive tasks: AI tools recognise patterns in the data they are trained on and perform tasks accordingly. So one could say that they are the ideal interns to handle your boring, repetitive tasks in seconds.
Why avoid AI in language projects?
In other areas, however, computers perform less well. “AI doesn’t understand humour”, “AI doesn’t make funny puns” … These are all popular arguments to avoid AI in language projects. They are certainly valid, but what if humour or puns are not an issue in a text? Does that mean translators and copywriters are redundant after all? No way, there are plenty of other arguments to avoid AI:
- Confidentiality: AI tools such as DeepL or Google Translate can use data to learn and reuse for other translations. It is therefore best not to have sensitive information translated by a translation engine.
- Bias: Research⁴ has shown that AI tools are often biased by the data they are trained on. Due to the so-called AI bias⁵, systems can sometimes produce results that hurt or disadvantage minority groups.
- Misinformation: Generative AI looks for the most plausible results according to the data it is trained on, not the most correct ones. So due to AI bias, errors in the training data or incorrect correlations, the output is not always correct.
AI at Untranslate
According to Untranslate, using AI is not a with-or-without issue. We always make informed decisions whether or not to use AI tools to enhance efficiency during a project. Our team is not only linguistically proficient, but knows the strengths and weaknesses of linguistic AI tools inside out. This allows us to always use the most appropriate tools in the most efficient way for your projects.
Questions about how the Untranslate team can assist you?
Click the button below and contact us to discuss the approach to your language project.
References
- “Mensenwerk wordt ondergewaardeerd”, vertalers voelen sterke concurrentie van AI – Terzake
- What is AI? – Cole Stryker, Eda Kavlakoglu, IBM
- European Language Industry Survey-rapport (ELIS)
- Humans Are Biased. Generative AI Is Even Worse – Leonardo Nicoletti & Dina Bass, Bloomberg Technology + Equality
- What is AI bias? – James Holdsworth, IBM