Over the last few years, there has been a growing trend towards using automated tools (such as generative AI or neural machine translation) for translation tasks. Some universities and publishers now even use them for translations of academic books and articles.
It’s not hard to see why: having a text machine-translated and then ‘post-edited’ can seem like an attractive option for those who don’t have the budget to have their text professionally translated. In other words: it’s much cheaper. And unfortunately, many university departments are currently being forced to tighten their belts.
As for the quality: machine-translated output is (depending on language combination) much better than it was a few years ago, when automated tools would often produce ungrammatical, unreadable nonsense. For many purposes (low-risk, low-importance, low-complexity texts), machine translation post-editing or MTPE may seem to be ‘good enough’: combining the speed and affordability of machine translation with the understanding and judgement of a human expert.
And yet speaking from my own personal experience, many clients who’ve experimented with MTPE to save money have returned to me because they weren’t satisfied with the results. I’ve experimented with some of the same tools myself and come to the same conclusion: machine translation is simply not suitable for translating complex academic texts to a level of quality I am satisfied with. For clients who value quality, and for whom ‘good enough’ isn’t good enough, there is simply no substitute for having your text translated by a human, not a machine.
So in this post, I’ll make the case for why, even in the age of DeepL and ChatGPT, it is worth investing in the services of a professional human translator.
Firstly, a few general reasons against using AI and other automated tools:
- Environmental: Many of these tools have extremely high energy and water use. Can we justify using them in the midst of a mounting climate crisis?
- Ethical/legal: Part of the reason these tools are so cheap is that they often use content created by humans without paying them for their work. This has led to widespread concerns about copyright violations and plagiarism.
- Social: There are also concerns about the wider social impact of delegating complex cognitive tasks to machines: how will future generations develop their creative and critical thinking skills?
I won’t go into detail about these points here, as they’re discussed extensively elsewhere. But they are all big factors in why I now completely avoid using such tools (even if Google, Microsoft et al. seem determined to foist them on me). And they’re likely to be important considerations for many academic authors, too, particularly those working on issues of sustainability and social justice.
But what about the specific case of machine translation?
As I mentioned above, I have in the past dabbled with using machine translation in my own workflow, to see if it would help with my productivity and quality. I purchased a one-year licence for DeepL, which is one of the most popular tools. However, I was very unsatisfied with the results for academic translations in particular: while in theory it speeds up the ‘easy bits’ of the texts – that is, the ones that are straightforwardly written and can be transposed more or less directly into English – you still have to redo all the ‘hard bits’ from scratch. And in German academic texts, there are a lot of hard bits.
Moreover, because the machine-generated text can be so smoothly plausible, it takes a lot more mental effort to spot where it has phrased things unnaturally or introduced serious errors. The time savings I gained from the ‘easy bits’ were offset by having to go slowly through sections I hadn’t directly translated myself and make sure I hadn’t been misled by the machine output. Rushing through certain sections of the text is not compatible with carefully and conscientiously crafting a coherent translation. With academic translation especially, the bulk of the time in any case goes not into typing the words in English, but rather into researching terminology and into thinking carefully about how to structure ideas and arguments in a way that works for a new audience. Crucially, DeepL and other tools do not understand the texts they are working on: they will not identify factual or argumentative errors, they will not realise that someone or something being referred to one way in one sentence is the same person or thing being referred to using a different expression elsewhere, they will not be sensitive to granular context, they will not ask questions of the author if something is unclear or be able to explain and justify their translation choices.
Overall, it took just as long to complete the translations as if I’d done them without machine translation, and I was less happy with the results until I’d done even more tweaking and editing. So much for productivity gains!
And something that’s important to emphasise here: what I was doing was not MTPE but ‘MT-assisted translation’, where I was fully in control of the translation process and was paid appropriately to take my time carefully crafting the text – and even then, I wasn’t wholly satisfied with the quality. In MTPE, where the human post-editor is expected to work much faster. It’s less like translation and more like ‘damage control’ to remove the most serious errors from the MT-generated text. It is much cheaper for the client, but it comes at the expense of quality. Machine tools may be able to produce a text that is, say, 50/100 on the quality scale – but that doesn’t mean the post-editor can get to a text that’s 100/100 on the scale in half the time as if they’d done it from scratch.
In summary: it’s understandable why budget-conscious clients may find MTPE an attractive option. Just as long as they understand that the cost saving comes at the expense of quality. But if you do have the budget, and you want someone to work on your book or article with the same care, thought and attention that you did, if you want it to be shared with new audiences in a form that is not merely ‘good enough’ but engaging, accurate, comprehensible and well written, it is still worth paying for a professional human translator who can do the job properly.
Written by one of our longest-standing academic translators, Andrew G.-C.