Can you trust AI with your translations and copywriting?

Since generative AI became widely accessible in late 2022, companies have been experimenting with it and using it for inventory, forecasting, marketing, customer support and other functions. AI is also being used for writing emails and preparing documents for internal communication. There have been attempts to use it for translations as well, with varying degrees of effectiveness. In this article I will provide you a wider perspective on whether it’s a good idea, or whether not to trust AI with your translations. Teaser: you shall not trust it fully. But why so – let’s dive deeper to find out! 

A wild card in your box

Let’s take, for example, the most popular – ChatGPT. It has been trained on a vast amount of data and can understand requests in over 100 languages, including English, French, Italian, Spanish, Dutch, and German, and translate text between these languages. At first glance, this might seem like a threat to the translation industry. But is it really?

On one hand, AI processes information quickly. In its latest releases, its speed approaches that of machine translation. However, there’s a catch—it hallucinates, and this is not a temporary glitch; it’s a feature. AI was designed as a creative tool, not a translation tool. Accuracy is often crucial in translations. In recent industry experiments, AI has translated documents occasionally inserting text in the wrong language without any warning to the user. In our experiments, when asked to translate a bullet-pointed text, AI rearranged paragraphs and added content on its own discretion without being asked to do so. Would you want such a wild card when translating software or medical documents? The answer is obvious.

Another point to consider is confidentiality. If you’re translating sensitive information (such as a new product release), the only way to ensure confidentiality with AI is to purchase a private cloud subscription. Does it make financial sense to incur these costs if you still need professional linguists to oversee / proofread / confirm AI’s work? Ultimately, someone must confirm quality and take responsibility for the result, and AI won’t provide that assurance.

Another issue is energy consumption. AI’s energy use is significantly higher than that of traditional machine translation. In Europe, where businesses are installing solar panels to cut costs, any responsible company claiming to run a sustainable business should consider the costs of using AI.  Was there mention of free versions? Now you know, they are not quite ‘free’.

I am not saying that using AI for translations or copywriting is a bad idea. But I am sure that to do it, you need to build a professional human team around the technology anyway. So, for some companies coming to professional providers who already have all the necessary instruments and AI tools might make more sense and be more cost-efficient.

Can AI replace all the functions of translators?

We incorporate artificial intelligence into our work and continuously explore and implement the latest technologies to enhance our products and stay current. Last year alone, we examined over 30 platforms and international articles on AI trends. Our current use-cases span from terminology work to simultaneous interpretation tools). We are confident that AI will not replace humans, but will greatly simplify and accelerate processes. Our commitment to innovation allows us to offer the best solutions and ensure high-quality services.

AI bots are suitable for providing basic translations (for short phrases, paragraphs, as normally bots have limitations in terms of number of symbols to be inserted). It can even generate output in multiple languages from one input text. But the output still requires careful checking because AI often distorts meaning (hallucinates). 

For large texts (100, 200 pages, etc.), AI faces limitations in character count, struggles with images, many tools (still) distort formatting, and translate tables as plain text. The AI of today seems more ‘artificial’ than ‘intelligent’ and requires meticulous oversight.

As mentioned, ChatGPT understands over 100 languages but performs best with English.

“Hey ChatGPT, are we sure our translations in Poland or Ethiopia won’t fail?” There are languages lacking datasets for training large language models, making generative AI results in these languages either non-existent or very limited. Have you read anything translated into Ukrainian by ChatGPT? It’s not great. Polish translations aren’t much better. German translations have issues. Generative AI is heavily geared towards English, producing more-or-less acceptable results, but they still need verification.

AI through the eyes of linguists

There is an increasing volume of content in the world, as generating it (now with AI) is becoming easier and faster. Yet, often pure AI content is homogeneous and uninteresting. The linguistic industry is changing and preparing to digest larger amounts of content, which needs to be proofread and verified by humans.

Pure ‘handmade’, quality content (as well as its translation or localisation) will become a rarity and increasingly in demand, thus becoming more expensive. 

Humans remain at the core of translation technology (not in the loop, but in the core) and all technologies, as humans bear responsibility for the final result and its verification. Our experiments with AI have shown that it is a tool for linguists, not a replacement. 

Can AI actually help?

AI helps project managers in the projects of generating and translating subtitles. Correcting machine-generated data is easier than creating it from scratch. For the companies who do not use special translation software, AI can be helpful with certain templated data. However, you may need to manually reformat tables, as AI typically sends translated material as plain text. When asked to return text in its original format, it creates the appropriate code but doesn’t always handle it well.

Even with seemingly simple tasks, AI may not perform as well as a human. You might spend considerable time and resources prompting and trying to make AI work for you, but is it worth it? Will it become a subject matter expert? The answer is obvious. AI exists in every field, but specific tasks require specific solutions. A designer will likely use Adobe and similar tools, not ChatGPT, which can’t even compile basic images without errors. Building robots will also involve specialised solutions, not generative AI.

In translation, AI has been used for over 10 years in the form of machine translation. It functions thanks to large language data sets, known as large-language models (LLMs), which enable generative AI to operate by predicting the next word in response to a query. Generative AI (ChatGPT and its analogues) is suitable for translating relatively short texts (without formatting and tables). But more complex tasks present challenges. Need to translate PDFs, scans, or handwritten text, or 500 pages for the court? Ordinary AI won’t suffice. You’ll need specialised tools or specialised companies who help you deal with it quickly and as cost-efficient as possible.

Experts believe AI could be used to create SEO-optimised articles. If there’s little competition in your field or few search results for a query, AI-generated content might give you an advantage. Having content is better than none. AI can be a valuable tool for generating text in more competitive niches. However, the process becomes more complex, requiring more time. Less time than full manual content creation, but even so, SEO specialists recommend conducting thorough research, identifying search-engine expectations, analysing competitor content, specifying key queries, and creating detailed technical specifications. When these parameters are met, whether a human writes it or AI generates it, success is possible. Of course, human-authored content will be better, but not by a large margin. SEO experts advise caution. While AI can generate vast text volumes for pennies, its utility depends on proper query preparation and text finalisation.

Takeaways

We see increasing automation and AI-tool integration into the translation industry serving as an aid to professionals, not a replacement. Combining machine intelligence and human expertise enables high-quality translation.

AI translations aren’t universal or practical solutions. Some industries require caution. Medical recommendations, legal documents and user manuals require experienced linguists with best translation practices to act as editors.

Human involvement in complex technical localisation projects involves platforms, developers and local experts. Humans are operators and inspirers, managing translation projects, coordinating participants, and ensuring deadlines are met. In short, humans control the result. You need people for translations and communication with global audiences. You need people again to orchestrate any project. Someone must take responsibility for the outcome, assembling project components.

At Task Force, we believe AI is incredible, and tools like ChatGPT can help. However, staying current with trends and comparing AI work with reality is crucial. These tools are far from perfect, requiring human linguists to guarantee quality and accuracy. It’s too early to talk about a machine uprising or AI fully replacing humans.

Task Force is a linguistic services provider, one of the Top 20 in the Eastern Europe and fourth fastest growing in the world.