Why Your Translations Can't Yet be Fully Automated

This guest blog post has been brought to you by Kayleigh Alexandra from MicroStartups.

There’s no denying that the rise of automation has achieved some incredible things. It’s similarly inarguable that machine learning in the translation field is making major strides. For example, semantic mapping is already widely in use and there is every chance that the coming years will see further advancements.

All that said, translation remains one of the biggest challenges for AI systems, and it’s a hurdle that seems unlikely to be overcome for a long time yet. If you want to pick up a vague understanding of what a menu item involves, or you’re trying to identify a bathroom, then an automated translation will surely suit your needs. However, if you want to translate meaningful long-form website copy, you’ll need manual assistance. Here’s why:

Human language is highly illogical

The English language has long been regarded as the language of business, and it dominates culture in the Western world. This position leads many to assume that it’s a carefully-crafted entity with a consistent set of rules and a sensible structure. That assumption is incorrect. The English language of today, like most languages (barring some exceptions such as esperanto), was cobbled together through the disjointed efforts to millions of people over hundreds of years.

Human language, in general, is similarly clumsy, featuring rules with nigh-arbitrary exceptions, confusing constructions, and glaring inefficiencies. And while automated checkers can readily flag up basic grammatical issues, that ability only goes so far before it becomes guesswork. If you’ve ever used Microsoft Word to write something with any degree of informality, you’ll be familiar with how often it mistakenly deems things incorrect.

Until machines can think as humans do, something that might never happen, they’ll struggle to grasp all the awkward intricacies of ever-evolving language structures. That makes them far from optimal for handling translations.

Context is extremely difficult to account for

Context is important for a digital copy. The same thing referred to on two distinct pages might need two distinct terms to make sense. That’s before you even factor in the value of SEO. Keywords are essential for getting pages to rank, and if you fail to include them, you’ll see your level of traffic suffer.

Yet context is difficult for machines. This is partially because they can’t really stay apprised of popular culture in the way that a person can. Regular everyday interactions give us subconscious cues about how language is changing and how regional trends are progressing. Every day, we hear slang terms come and go, and find concepts being given new names.

All a machine can do is bring in as much data as it possibly can, and make an effort to meaningfully parse it to find trends. This is a process that’s comparatively ineffective and inefficient. Would you trust such a process with finding the right wording for a major category name? It takes time and effort to handle SEO-suitable translation.

Tone and brevity are complicated

Social media is riddled with arguments. A lot of that is due to the relative safety and anonymity of posting on the internet. However, there’s another major contributing factor: the lack of intuitive tonal markers. The reason that emojis have become so popular is that many people struggle to get tone across in text alone. It’s certainly possible, but tricky.

And when it comes to translation, the delicate nature of tone can so easily be destroyed. You can preserve the literal meaning of a sentence while leaving behind any of the tonal markers that determine how it should be interpreted. This is why having a non-literal translation is often preferable. If the translator knows what the writer intended to communicate, they can write something that gets the same idea across.

There’s also the matter of brevity. Digital copy often has character restrictions, or at least space restrictions. If a literal translation comes out four paragraphs too long, it’s not too useful. Take SaaS, for instance. Often a layout designed for a mobile device won’t be able to shrink and enlarge text as needed (at least, not legibly), so it must all be worked through manually.

Consider that Shopify still has various languages for its store-setup software in beta testing (openly acknowledged in 2018). Given its evident determination to corner the market, it’s likely going as quickly as it can. It’s far better in such a circumstance to rework the content until it fits the available space (just as anime and cartoon animations must fit dialog to mouth flaps).

Using automation for first drafts

For all these reasons and more, you shouldn't rely on automated translation for your copy translation. In fact, it might not ever be suitable for that. I’m not saying it’s useless, though, because it has a solid role in the manual translation process: providing first drafts. It’s likely to be much easier for a translator to get things done quickly if they already have much of the literal translation done for them. They just need to make the appropriate corrections.
The important thing is getting the right translators on the case. The right team will know when to use technology and when to step in to course-correct. Thus they are ensuring that the work gets done as efficiently as possible without sacrificing any quality.