We Need to Talk About MT-Mediated Communication

timelapse foto of people crossing a street

When language industry professionals think of machine translation (MT) they probably think of post-editing. There has been a wealth of research in the past couple of decades on how MT might change professional translation. Meanwhile, the use of MT tools in society at large remains largely understudied, even though understanding or conveying information across languages nowadays often means using an MT system.

Facebook alone provides over a billion machine translations a day.(1) By March 2021, there had been a billion installs of the Google Translate app.(2) Like in its use by professional translators, as a communication tool MT presents challenges and opportunities.

In emergency situations, MT is potentially lifesaving. In the current circumstances, MT may mean the difference between accessing monolingual public health materials or falling victim to online misinformation about Covid-19 vaccines. MT may also have a role to play in education. It may eventually be more successful than English as a medium of communication by enabling what has been called ‘uniform multilingualism’(3), a vision for how MT might allow users to communicate freely across languages they might not speak or know well.

On the less rosy side, the risks posed by MT use are not negligible. A Facebook error has in one case prompted a mistaken arrest after ‘good morning’, in Arabic, was machine-translated into Hebrew as ‘attack them’.(4) In a review I conducted with two colleagues (5), we found numerous other examples of how MT use can have serious consequences. A transport police officer in the US used MT to ask for consent to search a vehicle. The search led to criminal charges, but the motorist’s consent was later nullified in court because Google Translate was not considered enough to break the language barrier between the motorist and the officer.(6)

There are other potential legal ramifications of MT use. For example, MT may jeopardize immigration applications, or lead to challenges to spousal communications privilege when online MT tools are used by couples.(7) These examples show how understanding the limitations and use implications of MT is increasingly important. Researchers have treated this type of understanding as a matter of ‘MT literacy’. Being MT-literate means, among other things, understanding the limits of what MT can provide and being able to mitigate its risks by adapting its input and/or output.(8)

Promoting MT literacy is not a trivial task, however. In another recent study, I analyzed how MT was portrayed in the English-language press.(9) In my experience as a professional translation researcher, the perceptions of MT I encountered have often been nuanced and considered. The treatment of the technology in the news was quite different. When MT developers make questionable claims of ‘human parity’(10), this can filter through to the wider press with great fanfare.(11) What I found in my analysis of the news is that MT was often framed uncritically positively, and in ways that could give users a false sense of security.

In the future, mitigating risks may involve technological solutions such as quality estimation and interactive MT designs that alert users to potential problems.(12) But working out the limits of what MT can do for everyday communication involves several other questions. In high-stakes settings this is an issue of ethical governance. Society needs clearer guidelines on the use of MT in courts, hospitals, police stations and other settings where communicative acts may be personal and short-lived but nonetheless consequential.

There is also the more specific issue of accountability. If in an extreme turn of events Facebook MT sends you to jail even though you might never have meant for your post to be machine-translated at the other end, who is accountable for that? If healthcare workers or immigration officers turn to their smartphone for convenience, who responds for any problems associated with MT inaccuracies? Do we blame consumers for their low MT literacy? Do we blame developers for their overstatements? Do we scapegoat the technology or the cost of professional language services? These slightly provocative questions do not do justice to the complexity of the issue. But they do illustrate something less controversial: as the Internet becomes more mobile and online populations more linguistically diverse (13), everyday use of MT is a social phenomenon that deserves more attention.

 

 
(1) Way A (2018) Quality expectations of machine translation. In: Moorkens J, Castilho S, Gaspari F, et al. (eds) Translation quality assessment. Springer, pp.159-178.
 
(2) Pitman J (2021) Google Translate: One billion installs, one billion stories. In: The Keyword. 
 
(3) Ramati I and Pinchevski A (2018) Uniform multilingualism: A media genealogy of Google Translate.  20(7): 2550-2565.
 
 
(5) Vieira LN, O’Hagan M and O’Sullivan C (2020) Understanding the societal impacts of machine translation: a critical review of the literature on medical and legal use cases. Information, Communication & Society. DOI: 10.1080/1369118X.2020.1776370. 1-18.
 
 
 
(8)Bowker L and Buitrago Ciro J (2019) Machine Translation and Global Research: Towards Improved Machine Translation Literacy in the Scholarly Community. Bingley: Emerald Publishing Limited.
 
(9) Vieira LN (2020) Machine translation in the news: A framing analysis of the written press. Translation Spaces 9(1): 98-122. 
 
(10) Hassan A, Aue A, Chen C, Chowdhary V, Clark J, Federmann C, Huang X, et al. (2018) Achieving Human Parity on Automatic Chinese to English News Translation.  
 
(11) See Vieira LN (2020).
 
(12) Liebling DJ, Heller K, Mitchell M, Díaz M, Lahav M, Salehi N, Robertson S, Bengio S, Gebru T and Deng W (2021) Achieving Human Parity on Automatic Chinese to English News Translation.
 
(13) Wu AX and Taneja H (2016) Reimagining internet geographies: A user-centric ethnological mapping of the world wide web. Journal of Computer-Mediated Communication, 21(3): 230–246.