TRANSLATION & TECHNOLOGY
Local and international restrictions implemented during the COVID pandemic affected how both individuals and organisations related to and navigated the world. While machine translation (MT) seemed a useful tool in communicating information to culturally and linguistically diverse (CALD) communities, issues arose regarding quality, accessibility and ethics. Specialist legal interpreter Christine Le is investigating how T&I professionals will deal with these dilemmas, and the scope of their ethical responsibilities.
‘… ethical issues began to emerge alongside the developing language technologies.’
As COVID took hold and the need to deliver essential health messages promptly to CALD communities became evident, MT assisted by integrated AI technologies seemed to many a sensible solution.
However, the quality and accessibility of the resulting translated materials proved limited and disappointing. In addition, ethical issues began to emerge alongside the development of language technologies. These include issues around data security and usage, lack of professional guidelines, and transparency of the translation modalities adopted – human translation, MT with postediting, fully automated MT and so on.
As T&I professionals, how will we deal with these ethical dilemmas, and what is the scope of our ethical responsibilities within the whole ecosystem of MT stakeholders?
Ethical impacts of MT
With the advance and adoption of digital technologies in the workflow of T&I processes, the professional codes of conduct that are designed to help practitioners navigate ethical decisions need to be extended to cover related ethics, such as data ethics and business ethics.
One common misconception is that technology is ethically neutral. It’s true that human beings tend to be subjective; however, it doesn’t follow that technological systems are objective, since each incorporates the values of both its creator(s) and the human subjects studied to generate the data on which it operates.
Nowadays, translation work is often managed through web-based platforms on which translation companies and their clients can access the input of individual translators and instantaneously monitor their activities. This automatically grants permission to access data and potentially reuse it in the future. Given that companies and translators may not have a unified position on data protection and usage, what ethical solutions and standards can be applied to prevent data errors, bias, misuse, or reuse without permission?
On reviewing several government guidelines for providing translations to the public, we found that their position on this topic is more often ‘sitting on the fence’ than clearly defined. Use of machine or automated T&I tools is generally not recommended, although integration of AI tools such as text-to-speech into websites to improve accessibility is mentioned. Many of these tools integrate MT to support multilingual content (although not all community languages are available).
Unfortunately, neither the ISO standard for translation (ISO 17100) nor that for MT postediting (ISO 18587) covers the risks of MT use, nor do either set out who is ethically and legally accountable when translation work is fully or partially automated.
For example, suppose a language service provider (LSP) is awarded a government contract to deliver multilingual content to the public. The individual translators and posteditors subcontracted by the LSP will have little – or more likely no – control of whether and how AI technologies are incorporated into the overall process. Furthermore, there is little room for these individuals to negotiate the ethical and legal responsibilities they share. Yet under the current legal framework, ‘liability’ is always attributed to a human or group of humans. In other words, a person or an organisation is ultimately responsible for any injury or loss caused by a faulty translation process that fully or partially incorporates MT.
Another prominent concern around MT use that runs through the reports and academic research examined is that of quality. In critical contexts such as health and legal settings, the consequences of mistranslation or misinterpreting can be significant. However, although MT use isn’t widely recommended, medical and legal professionals sometimes turn to it for a fast solution, even if it’s not a better or reliable one.
MT use in medical emergencies
Some research into the assessment of MT output quality when used in health settings has concluded that to ensure accuracy, web-based translation tools such as Google Translate and Microsoft Translator still need heavy postediting. For example, one scenario-based study investigating the quality of communication between emergency medical personnel and individuals with limited English proficiency via translation tools (such as QuickSpeak and Google Translate) in a mock emergency situation in the United States confirmed the potential usefulness of the MT in this context. However, numerous errors and usability issues did impact the user experience in this particular setting.
MT use in legal settings
Legal rulings made in similar cases involving MT use can be poles apart. For example, in the following two cases that occurred in the USA, the defendants had both had drugs found in their cars after police had used Google Translate to convey search requests to them, and both argued that the request had been imprecisely rendered. In one case, the court invoked the ‘good faith exception’ – in other words, even if the search request was not appropriately or accurately rendered, any evidence obtained under it was still valid. In the other case, however, the court granted the defendant’s motion to suppress the evidence gained in the search, based on uncertainty around their comprehension of – and therefore consent to – the request.
Ethical problems involving AI technologies in real-life crises are not easy to address, not only because the technologies are still in the early phase of development, but also because of the complex correlations between them and their developers, corporate and individual users, policymakers and so on.
Conclusions on ethical implications
In conclusion, more systematic investigations into the ethical use of MT in critical situations need to be conducted, and backed up by comprehensive industrial guidelines, before it should be introduced into real crises.
As part of this process, the ethical responsibilities of all stakeholders in the chain of an automated machine translation system – including MT developers, policymakers, translators, posteditors and MT users – must be determined, specifically:
- MT creators must ensure data security and acknowledge data ownership
- policymakers must create appropriate guidelines on the ethical use of MT through consulting with relevant industrial codes and associated laws and regulations
- human translators and posteditors need to learn about MT and be familiar with potential issues for them
- users such as LSPs are responsible for informing both their clients, and the translators and posteditors who they employ, of the translation modalities they adopt and their potential risks.
To ensure the healthy development of MT and the sustainability of the T&I industry, ethical considerations should never be seen as ‘the icing on the cake’; instead, they should be treated as critical in reducing the risks involved in T&I projects. Language specialists must be involved in the process of MT development, alongside professional bodies responsible for industrial codes, to ensure high-quality products with minimum risks to their developers and users.
Christine Le holds an MA from the University of Queensland and an MEd from Queensland University of Technology, and is currently a PhD candidate in media translation at Western Sydney University. She has sat on NAATI’s Technical Reference Advisory (TRAC) and Ethics committees; is a specialist legal interpreter (SLI) with over 10 years’ experience in providing T&I services in legal settings for various Australian courts and law firms; and is also a trainer at RMIT and a subtitling translator and reviewer for SBS.
Advertisement:
![Fmhhs72119 Translation And Interpreting Ad 2024 199.5x95.5mm Fa Digital 700dpi[88] Copy](https://ausit.org/wp-content/uploads/2024/07/FMHHS72119_Translation-and-Interpreting-Ad-2024_199.5x95.5mm_FA_DIGITAL-700dpi88-copy-scaled.jpg)
