Recently, I had the pleasure of observing a roundtable discussion between Dr. Lucas Nunes Vieira, Professor Lynn Bowker, and Dr. Mary Nerman. The topic was machine translation use in public services, prompted by recent, groundbreaking research that Dr Vieira had published concerning use in public services in the UK. This discussion was moderated by Don Hebelthwaite, Head of Membership at the Chartered Institute of Linguists (a professional body of which I am a member). It explored the opportunities and risks of using machine translation (MT) across healthcare, social care, emergency services, and legal services.
Presentation of Research by Dr Vieira
Dr Lucas Nunes Vieira, a leading researcher at the University of Bristol, presented his study’s findings on using MT in the UK’s public services. This study was part of a larger initiative, the Critical Language Barriers project, funded by the UK’s Arts and Humanities Research Council. The findings showed that one-third of public service workers use MT tools, typically by their own initiative, and when interpreter services are not available. It showed a high reliance on commonly available tools, such as Google Translate and, more recently, ChatGPT. This is a potential concern as these platforms are not secure, and there are privacy and confidentiality concerns.
The study found vital areas where MT was used: face-to-face interactions with those unable or unwilling to speak English, frontline communication, and sometimes for translating documents such as leaflets. The choice to use MT was usually informal—based on need instead of a policy set in place by the public service organisation. The findings also found that a significant minority were recommended or encouraged by colleagues or higher-ups to use MT, especially in cases where interpreting needs were urgent.
Further data revealed that there was also little or no training on how to use these MT tools and that many of the users were unaware of the reliability and confidentiality issues surrounding MT. Dr Vieira emphasised that clear policies need to be put in place and recommended that institutions establish clear guidelines and even training on how and when to use MT, which would increase MT literacy and awareness.
Another issue found in the data was that those in public service often had a favourable outlook on and confidence in MT tools. However, they did not understand that problems could arise from using MT, such as mistranslations, additions, and omissions that someone without knowledge of both languages may not be able to spot.
Dr Vieira’s recommendations include implementing safeguards for the public sector, as banning MT use would be impractical, but these problems could be very serious if allowed to continue.
Professor Lynn Bowker – A Canadian Comparison
Professor Lynn Bowker, Research Chair in Translation, Technologies, and Society at Université Laval, was asked to provide a Canadian comparison. She started by reiterating how groundbreaking this research is; there have been no other similar studies to date in the world. However, she compared machine translation use in public services in the UK to a case in 2016 where the Canadian government unveiled a plan to implement MT specifically for its civil service. It was going to be developed in-house using only government data to improve security and accuracy. However, there was a large backlash from unions, language professionals and even members of parliament. The fear was that French would be relegated to a secondary language as a bilingual country where English is the majority language, so all government documents would be written in English and then put through this MT to create the French versions. French-speaking members of parliament, in particular, were vehemently opposed due to concerns that linguistic quality would suffer greatly; unions were worried about the translation profession losing its standing and being replaced by MT.
Professor Bowker emphasised the issue of how the government introduced this tool. It was presented as the be-all-and-end-all for communication, replacing translators in government. However, this led to people believing that standards would suffer. Ultimately, the Canadian government changed their messaging, calling the tool a “comprehension aid” for internal use only. It was to facilitate understanding, not translation, but that distinction was not made clear from the start. She stated that the lesson to be learnt from this case is that MT policies must be nuanced and transparent to avoid these sorts of misunderstandings. This could be extrapolated to both MT and AI, which need safe guidelines, transparency, and a critical view, especially when used in public services.
Dr Mary Nerman – A European Perspective
Dr Mary Nerman is a translation professor at Tampere University in Finland. She has experience interfacing with Finland’s government’s public services, which require all documents to be in Finnish and Swedish, an official minority language. She noted that limited data was available, but behind the scenes, public service agencies are discussing machine translation use in public services to bridge communication gaps, especially with non-Finnish speakers.
She also echoed the other two participants in calling for AI and MT literacy while recognising the potential benefits of communication and information dissemination. She argued that rather than creating new tools, which many governments and organisations do but then fail to maintain, we should focus on refining processes and educating laypeople that without proper training, public service workers will turn to generic MT and AI instead of using better translation resources, such as professional linguists. This could limit communication effectiveness in Finland’s non-Finnish languages and expose the process to risks, especially in sensitive situations.
Takeaway Messages
There were several common themes throughout their presentations.
Formal Guidelines and Training
Machine translation use in public services needs careful and clear guidance as well as training for public service professionals. The study especially showed that all areas included in public services use MT in an informal, ad hoc way, which raises concerns about misuse, misunderstanding, privacy, and potentially harmful and dangerous outcomes. Introducing and maintaining MT literacy training could help professionals recognise when, where and how to use MT, its limitations, and how to reduce risks as much as possible.
Balancing MT and Human Translations/Interpreters
Although MT is fast and easy to access, there are places where it cannot replace a human, especially in high-content, complex situations like healthcare and legal areas. A simple slip can cause serious consequences in these situations. The presenters advocated strongly, in their own ways, for a balance between the two, especially in cases where MT can pose a risk.
Ethics and Privacy
Dr Vieira’s study showed that public service workers usually used MT on their personal devices, such as smartphones, which raises serious questions about data security and worker-client confidentiality. Professor Bowker and Dr Nerman also highlighted that appropriate data security and handling are a must in guidelines for adopting MT.
Potential of MT for Public Access and Equity
While MT can make it much easier to access information, especially in multilingual countries, even “democratising” the process, it can also relegate languages that are not spoken by the majority to a “second tier” in terms of translation quality, as was shown in the 2016 case in Canada.
How does this affect a translator like me?
I am a translator who specialises in medical and scientific translations. Since the quality of MT has increased over the years, my job has been shifting from pure translation (with no help from MT) to something called MTPE—Machine Translation Post Editing, where a text goes through MT and then a human corrects the output. In theory, this should be faster than translation, but depending on the quality, it can take a lot longer to fix the text.
In the scientific and medical fields, there is the added issue of privacy. Scientists do not want others to be able to access their research before it is published for fear that it may be stolen from them. Medical documents usually contain sensitive information about patients that should be for privileged eyes only. However, MT solutions such as ChatGPT and Google Translate usually save the information inputted into their systems, which could include confidential or privileged information.
If they do not realise that this is happening, medical and scientific professionals may use MT to speed up the translation process—potentially while breaking privacy laws and inadvertently revealing sensitive information.
MT literacy for the masses seems to be the way to go, as ChatGPT and Google Translate are easy to access and will not be going away or replaced any time soon. This roundtable discussion also showed me that governments and public institutions also need to catch up with evolving technologies. They may be replacing translators and interpreters with a solution that they think works well, but this only masks the problems.
I think that my job security is directly proportional to MT literacy – the more people understand what MT actually does and how it does it, the more people will understand and see the need to keep translators and interpreters to combat these issues. Rather than just translating, my job is quickly becoming an educatory one as well.