CB Multilingual at 2023 Words to Deeds Conference on legal translation and risk
“Machine translation is not intended for widespread use.”
“Post-editors still translate since the translation process is not complete after a text has been processed from the source language to a target language by a machine translation software.”
Very enlightening statements by Prof. Felix do Carmo at Words to Deeds, the conference on legal translation and risk held at Jesus College in Cambridge (UK), which we had the pleasure to attend on 28 and 29 January 2023.
Top-notch speakers and professionals from all over met in Cambridge to discuss different issues pertaining to the risks of legal translations, in particular when using machine translation tools.
Different positions were presented and discussed:
- The pessimistic viewpoint: translators in fact are not being replaced but earn less for doing the same (or more) work.
- The optimistic viewpoint: there is a great potential for delivering in a shorter time span a “better” service when always having a trained legal translator to revise the output of a neural machine translation system.
One of the speakers highlighted an array of errors that a legal translator and post-editor should pay particular attention to when post-editing/processing a machine translated legal text, such as:
- Hallucinations (part of the content that is not in the source text and is included in the translation)
- Deletions (part of the content of the source text that is not included in the translation)
In addition, she elaborated on the fact that, in practice, frequent users of machine translations are judges and lawyers, who seem to rely quite a bit on the output of such tools and believe that they “can fix all errors” and skip proofreading and post-editing by professional legal translators/post-editors.
Human translations still do better
Scholars teaching at different major universities also stressed the fact that human translations are still more accurate and less prone to errors. In a machine translated text there is, statistically speaking, at least one error every 10 words and 10 deceiving pages out of 100 pages to be found.
In a few words, post-editing by a human legal translator, specifically trained and qualified to spot and correct these errors, some of which are inconspicuous at first glance, is necessary.
Risks in legal translation
Particular attention was paid to the question of liability: in case of errors, who is responsible if the translation results from processing the source text with machine translation tools?
Risks of a different nature arise with the ever-increasing trends of using machine translation tools, even in a very specialised field such as legal translations. And are enormous.
The following key questions are to be addressed:
- Who is liable for any such potential copyright violations, discriminating language, content deformation or misinterpretation, additions or deletions or inaccuracies? In fact, for the time being, this is not as clear cut as it seems.
- What is the maximum level of risk that can be tolerated?
- Are the providers of machine translation tools responsible for such errors, or the specialist who uses the machine translation that still contains many errors?
- Could we use a stamp to trace back who did what in the process?
- Can an individual alone be held accountable or is it a matter of collective liability?
New rules are to be drafted and enforced to face this new situation and protect consumers’ rights.
As it is, there are only very few cases that all participants at Cambridge knew of in which disputes regarding translation results have been brought to court. So insurers have no actual basis to assess the potential risks of such activities.
Again Prof. de Carmo stressed that a major risk is the basis on which the new generation of students will learn how to translate. We could not agree more.
Take the example of a young pupil at school who is never trained to perform simple mental calculations but uses only a calculator from the very beginning. They’ll become dependent on such devices and be deprived from acquiring fundamental skills.
Post-editors can only be efficient and reliable if they have enough experience as translators first.
Many risks, much potential, many challenges. Time will tell, showing the best answers to be the right ones.