In one of my recent blogs I looked at how AI-driven translation technology is now taking hold in the world of journalism. I pointed out the danger of “homogenizing” news and ironing out cultural differences. Here, I’d like to say something about the homogenizing effect that MT is exerting on translators’ work – that aspect which in academic terms is referred to as the “voice of the translator”, or the “discursive presence”.

Translators, just like writers, have or used to have a recognizable style that can be discerned across their work by those sensitive to the finer points of language use and style. A kind of pattern of linguistic features that they favour and therefore use more frequently; syntactical constructions they like, or avoid, preferred vocabulary and stylistic flourishes, even punctuation (I, for example, like to use a semi-colon from time to time… which I see rarely in anyone else’s translation). Such idiosyncracies would be present somewhat independently of the source text.

Admittedly, with “commercial” translations, the individual translator’s voice is not exactly prominent. On the contrary. Think of website content for example, or massive projects, where hundreds of thousands of words need to be translated with a quick turnaround, meaning that 5 or 6 translators will be sharing the work. The aim here is to make the translations as homogenous as possible, with no one “sticking out”.

In fact, this was one of the main selling points when CAT tools first made their appearance. Not only would they ensure that the same glossary terms were used by everyone involved, but also impose a uniform style.

Now that everyone’s relying on MT for their starting point, what impact do we see on the voice of the translator?

I was searching around a little to see if anyone else had given this some thought. In vain: I did not find any opinions on the matter, until I came across a very interesting study entitled “Machine translation, ethics and the literary translator’s voice”*. The Google engine had hit on the introductory remark: “Recent work in translation studies has established the literary translator’s voice as an ethical concern, but there has been little empirical research so far into how the translator’s voice is affected in workflows involving machine translation.”`

In their study, Dorothy Kenny and Marion Winters explore the idea of how the translator’s voice may be impacted by the use of machine translation. They set up an experiment with a well-known translator into German, Hans-Christian Oeser**.

In the experiment, Oeser was asked to post-edit an excerpt from a novel he had previously translated (without MT). The researchers then compared the post-edited version to both the MT output (DeepL), but also to Oeser’s earlier translation (dating back about a decade). Their findings indicate that the translator’s voice, or textual presence, is “somewhat diminished” when MT is used. In other words, the translator leaves less of his personal stylistic preference, less of a “thumbprint” and accepts a more “generic” translation.

Criteria included Oeser’s stylistic preference for a higher register, for a wide vocabulary, for subject-verb inversion, etc. It was found that while the changes he applied when post-editing the MT output, were entirely in line with his usual preferences, they were somewhat less numerous, i.e. his final output was less different from the DeepL version than his original translation. Let’s remember also that there were some 10 or more years between his original translation and the post-edit, so he was not able to remember his earlier translation (he confirmed this).

I would conclude from the results that what happens is that the post-edited version is somewhat “less personalized”, “more DeepL-ish”, than a translation done in free-hand style, and slightly less recognizably a translation by the specific translator, Oeser. Which in the case of a renowned literary translator is of course a loss.

While the study is focussed on the literary branch of translation, I believe that it has some relevance to what we are doing in our everyday work as well, and confirms our own experience. When time is of the essence and deadlines and budgets have to be met, the temptation is no doubt to change as little as possible, leaving more “as is”, whether it comes from a TM, or from an MT engine. In fact that is very often what the clients want.

My perception and fear is that over time translators working as post-editors may become more and more tolerant of what the machine churns out – and perhaps clients too. Linguistic lethargy or over-tolerance may start to set in.

I believe it is true therefore to say that the translator’s voice is less manifest in her/his post-editing work than when we work from scratch. In other words that it is compromised in some ways: as the machine becomes better and better, we edit less and less. This results in texts that are more boring and devoid of personal style. The translators voice – and also the authors – gets submerged somewhat.

The aim as stated by the two authors of the study is to offer a “methodological intervention and represents an initial attempt to design studies in literary machine translation that put the focus on the human translators, allowing their voices to be heard …”.

Perhaps this is the time to re-think the entire translation process and the translator’s role, both for literary and technical purposes. Not along the lines envisaged by Jaap van der Meer of TAUS and his Data-Enhanced Machine Translation, but in the opposite direction, with more emphasis on the human-in-the-loop. Can we have a bit more of the Voice of the Translator again, please!

*https://researchportal.hw.ac.uk/en/publications/machine-translation-ethics-and-the-literary-translators-voice (Publisher Rights Statement: This article has been published in final form in Translation Spaces http://doi.org/10.1075/ts.00024.ken © John Benjamins. The publisher should be contacted for permission to re-use or reprint the material in any form