Why hire a human? There’s more to translation than first meets the eye
I’ve written before about how being a translator isn’t just about speaking a language other than your mother tongue. It’s not even just about speaking second language well.
On that occasion I gave a brief overview of what Google and other machine translators can (and can’t) do compared to a human being. Here, I do the same but through examples closer to my own translation work and with a deeper look at the theory behind it.
You have to laugh
Cautionary examples of translations gone wrong abound. It might be the difference between ”like” and “love” or “a lot” and “very much”. Perhaps the result was salacious cash machine signage. These are among the many meme-worthy sloppy translation gaffes.
Delving a little deeper, though, we begin to see the more serious side: bizarre advice from the Strathclyde Fire and Rescue Service – “Never jump out of a window straight. Put yourself on a donkey and come down” – offers insight into the dangers of errors. Seemingly humorous… but a fire service message is an important one to get right.
Advances in machine translation
Truly terrible (if hilarious) translations are often blamed on machines, but good and bad translation isn’t a simple matter of machine or human. It’s more nuanced than a good-bad binary. For one, machine translation has vastly improved and will continue to do so.
There are a great deal of far more detailed and knowledgeable resources on this subject, but here’s a brief overview.
Machine translation has been around for decades. Historically it relied on two main systems: statistical MT and rule-based MT. The former uses corpora (a collection of texts used for linguistic analysis. For translation, I suppose, that’s actually two bodies of texts: sources and their corresponding targets). The latter is informed by a set of linguistic rules.
Both systems involve a lengthy creation process as they rely on manual inputs, particular in the case of the latter. They also can’t be 100% accurate, having to account for too many variables – human language has been described as “inherently” ambiguous and flexible, where there is “no one single best translation of that text to another language”.
Yet, as machine learning improves via neural networks, it can begin to clear such hurdles. NMT is the driver behind machine learning and AI, and the idea is that the machines learn to interpret the language themselves.
Today, developments in all MT systems are manifesting in a variety of ways: translation agencies are training their own MT systems and so translators are increasingly being offered jobs in post-machine-translation editing (PMTE). Translators themselves can benefit from CAT tool plug-ins, such as that offered by DeepL – after checking their T&Cs against NDAs and other client agreements, of course. Most translators and language students will be familiar with that company, having used Linguee,
Getting a “good translation” out of a machine isn’t impossible but may still require input from a human translator and/or depend on subject matter.
Not just language but subject expertise
Speaking of which… When it comes to human translation, the sliding scale of good and bad translations depends heavily on subject and target audience knowledge.
I specialise in sports translations, particularly cycling translations or running translations and, within that, products for those fields. I know, in English, the technical as well as the colloquial terms and expressions for various aspects of the sports. I know what people are looking for in their kit for different types of effort, discipline or weather. On the marketing translations side of this, I know what peoples goals are – that’s important because if the product description convincingly shows it can help them achieve those goals, they’re far more likely to buy the item.
From a more technical viewpoint, an athlete might be looking for specific sports products: compression wear, supportive footwear, nutritional products – the list goes on. In Italian, I find these products often use the more formal, perhaps medical terminology to describe their purpose. That’s less likely to be the case in less Latinate English, where your everyday runner wants to understand immediately whether it corresponds to their problem in language that makes sense to them.
That these are often marketing texts aiming to sell a product brings SEO into the mix. The intricacies of SEO content are worthy of several dedicated entries, but to give a reductive summary, SEO is about keywords and ranking in Google searches. To produce a piece of content that helps a company rank more highly in a Google search, you need to produce something relevant to what people are searching for.
Knowing that for a subsequent translation is of vital importance. You can translate a word directly into another language but there are so many ways such a translation will fall down: remember the inherent ambiguity from earlier? There are often several equivalent terms in the target language, but your translation is not necessarily the one people use most commonly overall or in your target area: the term people are actually using to search could be another (related) one entirely.
I could go on – in a legal field, for example, it’s about knowing both source and target systems, in a medical field the translator would need to have a firm grasp on the terminology target culture medics use and understand and so on. Hence, the importance of a translator’s subject matter knowledge in combination with their linguistic skills.
For whom are you translating and why?
Subject expertise and two languages still isn’t enough. You could be bilingual and trained in a specific field – say, finance. That would make you a good candidate for translation, but not necessarily a good translator. Why not?
I’ll be frank: translation theory was not my favourite Masters module. But I find myself constantly drawing on the issues studied to justify – outwardly or inwardly – my choices as I translate. Usually, it’s in answer to the question, “What is the author trying to say and how did they say it?”
Skopos is a term that sticks in my mind. Boiled down to its very basics, it’s the idea that the quality of the target translation is directly linked to the target’s purpose. We’ve already begun to see this with the SEO explanation above.
Another simplistic demonstration would be a comparison between translating an internal memo within a company against translating a product description for a high-end fashion product. The former doesn’t need to be polished (the source text probably wasn’t), but it does need to get the facts bang on. Ideally it won’t contain typos (the source, I mean), but stylish writing probably has little bearing on it being correctly received.
The latter needs to be polished until you can see your face in it. The facts still need to be there – you can’t say a jacket is made of leather if it’s actually PU – but fact is probably only a portion of the source text. It likely makes gentle appeals to the source culture in it (puns ranging from cleverly subtle to dad-joke, in line with the product and/or brand’s style). The target text has to replicate that effect for the target audience, even if that means changing things up a lot. When I work on these texts, word and sentence order is likely to change. Adjectives, metaphors and colours, even if standard, rarely have direct translations, and the translator has to find a suitable solution.
That’s before we’ve even considered whether this hypothetical bilingual might be better as an interpreter. They’re bilingual and have expertise in both languages in a certain (for argument’s sake, required) area. But they’re not given to public speaking. Would they make a good interpreter? Arguably not.
Machine translation is improving all the time but is far from being the most widespread solution and even further from being able to go it alone without a human expert to check its working – certainly for the time being. I have covered a lot of ground here, but I have managed to show the benefits of hiring not just a human translator but one who really knows their stuff.