Labels in Translation – a matter of ethics

Our profession has always been in evolution – from how we recorded our work in the olden days to how we research and produce translations currently.

People have long tried to develop automatic translation systems. The earliest references date to the 17^th century, but most developments took place in the 20^th century. W. J. Hutchins published Concise history of the language sciences: from Sumerians to the cognitivists  (Pergamon Press, 1995) and, in the chapter titled MACHINE TRANSLATION: A BRIEF HISTORY, he takes us down machine translation memory lane.

Enter the 21^st century and change has been at a breakneck pace. Recent technology developments such as neural machine translation (NMT) using transformer-based architectures, multimodal machine translation (MMT), self-learning algorithms (or Adaptive MT Models), large language models (LLM), and natural language processing integrated with GenAI have brought to our attention the need for discussions about ethics in their use, especially where consumers are concerned.

Ethics must come into play because the general public often projects human-like qualities and capabilities to technology and tends to accept their output at face value. Regardless of what process is used, MT output still produces hallucinations (inaccuracies, misinterpretations, biases), and requires validation by a qualified professional to ensure that it properly corresponds to the source.

Therefore, we ask is it ethical for an organization or individual to publish raw machine translation output without some kind of warning label?

Raw machine translation is acceptable in some situations, for example in social media. Platforms actually have a “see translation” button available for posts not in the language of their target public. And users of the feature are aware that what they are reading is raw machine translation output. However, in medical, legal, engineering, and other specialized settings, errors may result in risks of harm and users should be warned.

The recently published ASTM F2575:2023 Standard for Quality Assurance in Translation and ISO 11669:2024 Translation projects, General guidance introduce the concept of labels for translations, similar to consumer labels we find in various products, aiming to offer translation consumers transparency as to the origin of the material they are reading (or listening to, in the case of audiobooks, for example), providing them with a means of risk management.

These labels will allow translation consumers to easily identify if a qualified professional was involved in the process of translating the material they are enjoying. The main labels used by ASTM are BRT, which stands for bilingually reviewed translation, and UMT, which stands for unedited machine translation.

You can learn more about labels and their use by reading Labels: an ethical approach to Translation and Technology (updated), by Prof. Alan Melby.