ChatGPT has been everywhere since its launch at the end of 2022, not only on professional content creation and translation forums, but also in the general media. We read extensive articles about its impact on the education system and the issues faced by teachers in distinguishing whether a piece of work has been written by a student or comes from the popular chatbot. Some people even believe that it poses a threat to the business model of the Californian giant Google.

What is ChatGPT?

ChatGPT is an artificial intelligence chatbot created by OpenAI. It was trained using an enormous text corpus to generate responses expressed in natural human language to questions on almost any subject. It belongs to the Generative Pretrained Transformer (GPT) family of language models, and falls into the category of generative artificial intelligence, which consists of algorithms based on automatic learning, designed to create content —not just text, but also images or music— based on the information used to train it.

The impact of ChatGPT on translation professionals

Firstly, it is important to point out that artificial intelligence is not a newcomer to the translation world. Machine translation systems based on artificial intelligence have been used extensively for many years. With or without human post-editing, with translation engines trained specifically using a customer's content or general engines, machine translation has been in use for decades. Its integration, to a greater or lesser extent, in the workflow of a translation professional allows various approaches, and is determined by factors such as the language combination or the type of content and the tone used.

Scientific studies have been published comparing the translation provided by ChatGPT and that of machine translation tools designed specifically for translation, such as Google Translate. The result of this comparison of quality metrics indicates that machine translation tools perform better than ChatGPT, with less of a quality difference in language combinations where a larger content corpus exists, but with a very noticeable difference when it comes to lesser-used languages, or languages that are very different linguistically, such as a European language translated into an East Asian language. However, these studies indicate that there is a type of content where ChatGPT provides either the same quality or outperforms standard online translation engines, and that is text written by forum users, chats, etc., which often contain grammatical errors or spelling mistakes. This content appears better adapted to the content used to train ChatGPT.

There are two other aspects where neural machine translation systems —the main type of machine translation used currently— outperform ChatGPT and other Large Language Models (LLMs): data confidentiality (zero in the case of text submitted to ChatGPT) and integration in existing platforms.

Using ChatGPT to generate content ready for publication

In its current form, ChatGPT is the equivalent of a person that does not know anything about a subject performing a search in Google and using the initial results in the list of pages found to combine them in a written summary in a way that sounds natural. It's not bad! In the medium-term, the cost of content creation will undoubtedly reduce significantly, and therefore the amount of content created is expected to increase. This content must be checked and validated, giving rise to a new type of language service, which in a certain sense will expand the current scope of machine translation post-editing.

OpenAI is clear about the fact that the content produced by ChatGPT, despite being written in very fluid language, may contain inaccuracies or even be completely incorrect, or that it may contain unacceptable biases due to cultural or other reasons. It may also not be fully up-to-date, as ChatGPT is not connected to the internet and therefore cannot take the latest information that has just been published into account, making the review and validation process absolutely essential.

STAR, always at the forefront of technological innovation

To remain at the forefront of language service provision, at STAR, we believe that a hybrid approach combining the vast potential of AI-based applications and human talent is the best way to tackle the present and the future. This involves offering innovative language solutions in which technology with AI plays its part, and people with their cognitive abilities play a key role in ensuring the final quality required for professional content.