[Part 5 of my series on language and service design]
How do you translate something exactly? As someone on Twitter said recently, translated works are basically rewritten. My favourite example of this, and also my favourite Wikipedia page, is how you might translate an expression indicating heavy rain. In Welsh, we say that “it’s raining old ladies and sticks” – mae hi’n bwrw hen wragedd a ffyn. If any English speakers reading this think that’s a bit strange, then you should probably take a moment to reflect on the fact that this makes equally as much sense as cats and dogs falling out of the sky. Or trolls. Or wheelbarrows. Or frogs. Or snakes and lizards. Or husbands.
If you want to say something, there’s more than one choice in how to say it: words matter, choice matters, the care that you apply to the process matters. Not all translations of Anna Karenina are equal. The recent CDPS seminar on language featured an example of how the use of a word for divorce in Welsh on the courts system had confused users until it had been replaced with another word.
I once worked for a Government agency that had some Welsh language content online, and also Welsh language content displayed on signs around their headquarters. The problem was that the actual name of the agency had been translated differently across the two. Could anything be more fundamental than the name of the organisation? Yet you could see how the problem had emerged: two different translation agencies, engaged at two different times, had come up with two different approaches.
A few years ago I worked on a website that had a page designed to help people find their local GP. The problem was that most people don’t say ‘GP’, they say ‘doctor’. When they search on the web, they say ‘doctor’ there too. Users weren’t finding the right page on the website because it didn’t have the word ‘doctor’ on it. I changed the term and the page went from being in the bottom 50 for the whole site, to one of the top pages.”Sarah Richards “Content Design”
If words matter, then content design matters a lot. Content Design is the term used to describe the approach to words in service design. It often involves the move of responsibility for content out of a comms or marketing area more oriented around print into a specifically digital area, better able to cope with an iterative approach to the provision of services. I’m going to finish this post by giving a couple of examples of how content design can help provide the correct consistency and tone to services, but please don’t think that this is all that content design does: it’s far more than that.
In the past, I’ve had to manage the daily translation of content for online usage to the point where I’ve used between two and five translation agencies simultaneously. The only way that I could make using multiple translators across multiple agencies work was to contractually mandate that they used, and contributed to, a dictionary of common terms that we held. Any other approach would mean that common terms, names, or descriptions could end up differently in adjacent content.
Different terms translated differently might sound like it makes for an inconsistent user experience, but it’s worse than that. For a start, it makes search problematic if you can’t search on common terms. Beyond this, language choices are not always neutral and can sometimes be political or offensive. Say, for example, that you are describing a role that had been restricted to men in the past. The choice of a gendered word to describe that occupation (e.g.fireman instead of firefighter etc.) is problematic, but can easily slip through translation if you don’t have a process in place to ensure consistency.
At a more nuanced level, translated content in any language is often criticised for being overly formal. Modes of address can be different in different languages, and translators without in-depth knowledge will usually default to the formal option rather than risk informality, particularly where the public sector is concerned. This can have major consequences.
Translators are also users in the translation process, and many professional translators that I’ve used have been older than average; partly, I think, because many undertake translation as part-time work. Many are also more used to translating formal language. (Who pays for translation? It’s often company reports, or legal documents, or government: none of them famous for informal, straight to the point tabloidesque immediacy.) Providing glossaries, and setting expectations for them is critical.
That using a robot translated version of text  without checking it shows a lack of respect for users is a well rehearsed debate, but a once-and-done translation with no check for consistency is almost as bad. GOV.UK has a style guide. What if the Welsh public sector had an open source Welsh language version of this that could be used (and updated) by Welsh translation agencies as a default?
 This is not to say that ML doesn’t have anything to add: it clearly does, can form a valuable part of a translation process and is the area of some exciting future work. This notwithstanding, cutting and pasting text into Google translate and expecting it to do anything apart from turn factories into vegetables and supporters into waste pipes is not a wise idea, at least in the short term. If you want evidence for this beyond alleged and amusing mistranslations, then this research looking at use of Microsoft Word concludes [MT=machine translation]
however, their [users’] eﬃciency (time for task completion) and self-reported satisfaction are signiﬁcantly higher when working with the released product as opposed to the unedited MT version, especially when participants are less experienced. The eye-tracking results show that users experience a higher cognitive load when working with MT and with the human-translated versions as opposed to the English original. The results suggest that language and translation modality play a signiﬁcant role in the usability of software products whether users complete the given tasks or not and even if they are unaware that MT was used to translate the interface”