Last year, I attended a conference on “translation in the digital age” at the British Academy in London. Unavoidably, the discussion centred almost exclusively on machine translation and the threat it posed to human translators. Three of the four guest speakers predicted that MT would reach such a state of sophistication that it could reasonably replace human translators entirely, and in all domains: commercial, legal, technical and, most controversially for me, literary. The only guest speaker to openly reject this prophecy of impending doom was, perhaps unsurprisingly, a literary translator…
Personally, I’ve never been particularly worried about the prospect of being made jobless by machine-produced translations. That’s not just because, if it were ever to happen, it certainly wouldn’t be within my lifetime, but rather because the romantic in me holds the mechanical, binary nature of computer language to be incompatible with the organic, spontaneous reality of human tongues. And nowhere does the originality and playfulness of language become more apparent than in the novel, the short story, or the poem.
And that was my main contention at the conference last year: whilst I can accept that machines will take an increasingly more important role in the translation of generic, legal or technical texts (sub-categories of language whose formulaic constructions present few problems to the translation algorithms of a machine, although even here it’s an imperfect science), I find it almost impossible to imagine a world in which novels are translated exclusively by machines. And, if such a world were ever to exist, there would probably be very little value in writing novels in the first place.
But to state my case more clearly, I’m not against machine translation per se in non-literary contexts. In fact, I often make use of it within my own translation practice. Modern-day machine translation software – rather than simply being “taught” grammar and vocabulary in isolation like its predecessors – uses complex statistical models to arrive at their translation. The Internet means that this software can draw from vast corpora of original and translated texts, mapping concordances and probabilities to produce the most statistically likely translation. This means that it really is no exaggeration to say that machine translation has come on leaps and bounds in recent years. But even now, despite the vast amount of data they have access to, machine translations can still sometimes get it terribly wrong.
The other day, I was translating a short piece from French into English about a stately home in Nice. I was with friends as I was looking over the final draft and, out of curiosity, they asked how my version would compare to that of Google Translate. And so I entered the text into Google and was, for the most part, impressed by the result. I say “for the most part” because its translation was unfortunately marred by its rendering of petit-fils, or “grandson”. Google seemed unable to decide how to translate this and, in its confusion, gave “breakfast son” (somehow, it independently introduced the idea of petit-déjeuner – “breakfast” – and ran with it). What’s interesting is that this is an instance in which context, rather than helping the machine, actually proved disadvantageous: type petit-fils into Google on its own and it will instantly give you “grandson”. Unsurprisingly, machine translation also fares just as bad when context is absent: I remember seeing a sign in a takeaway shop in Camden Market which read “Mélange et Allumette” – no doubt a computer’s attempt to translate the English idea of “Mix and Match” but which, to the amusement of many a francophone, actually just means “Mixture and Matchstick”.
My non-translator friends were surprised to discover that as a professional translator I would admit to using machine translation. My reply was that I use it “intelligently” (read: “cautiously”), for small sections and only after I’ve arrived at a translation independently. I use Google Translate, for example, with the knowledge that it is infallible and that it’s not to be taken at face value. Every translation choice I make is informed by a near-native knowledge of the languages from which I translate, and it’s this knowledge that means I can spot when machine translations have went awry. I use machine translations on the off chance that they might offer me a solution that no dictionary or online forum could. And sometimes they do precisely that.
This article from Stanford University makes for an interesting read: http://news.stanford.edu/news/2014/october/translate-human-machine-10-29-14.html. Whilst I’m not entirely a supporter of the post-editing of machine translations (depending on the languages involved, such editing is likely to take longer than it would to translate the text from scratch), the move towards integrating human and machine translation to improve productivity is a positive one. Arguably, this move has largely already happened with the introduction of CAT tools; translation memories register a translator’s input and use it to inform future suggestions for similar segments later in a text. By incorporating corpora-based machine translation from the Internet into existing tools, however, the human translator will be given a much-needed helping hand in meeting tighter and tighter deadlines. There’s even the suggestion in the article that Stanford’s new software can detect a translator’s style and adopt it when offering matches – an improvement that will help to avoid the cold, uniform translationese of automated translation.
I’ll conclude with a quote from the article by Martin Kay, author of “The Proper Place of Men and Machines in Language Translation“:
“”When I wrote [The Proper Place of Men and Machines in Translation], I was convinced of two things concerning computers and translation. One was that there would surely be many roles that computers would be able to fill; second, that automating the whole process would not likely be one of those roles.”
And so relax, technical and institutional translators; your job is safe, and what’s more, machines are going to make it easier for you. As for literary translators, machines can’t help you much, and I bet you couldn’t be more delighted.
L.S. TRANSLATIONS 