Why the Localization Process Can’t Be Fully Automated
Localization is key for a business’ international expansion, and it goes beyond culturally-sensitive adaptation. It’s the translation of a digital product, website, marketing campaign, manual, or any other information architecture, with its context in mind.
For instance, the localization of a website involves the translation of text so it fits the site’s UI as neatly as the original.
One should think of localization as a long-term commitment with the international client. It's an effort to always make your products and your message as accessible to them as possible. It’s showing them that you care, and it's an essential investment for any company with global goals.
Some instances of the localization process can be automated (such as the repetitive translation of a single word appearing on several locations within an interface), but both the cultural adaptation of material per se and the quality-assessment process have to be on human hands.
But what about Deep Learning?
Neural network machines absorb parallel bodies of literature in different languages, and map its sentences in what’s referred to as a “common latent space”. Basically, a “common latent space” is an intermediary phase between the input and the output, in which the neural network extracts certain features from the input in order to properly reconstruct it.
Machine translation is improving by the day, and machines are learning to carry out certain translation tasks unsupervised and with a skimmer and skimmer body of linguistic references. But their output still has to be fine-tuned by humans. Specially if the job at hand isn’t just translation.
Localization is not translation
It’s currently impossible for a machine translation to be indistinguishable from a human-made translation.
Consider this: Localization is far more complex than translation, because it involves more than simply word-by-word or even conceptual equivalence.
When it comes to marketing messages, a translator might be forced to resign word-by-word translation in order to get a slogan that is rhythmically close to its original, that sounds similar.
Sometimes, it’s more about the function that a certain text serves than about the text itself. The localization process is conditioned by a wider design structure that contains that text, and it’s detail-oriented. For instance, when it comes to language pairs such as English and Hebrew, the localization process will require re-writing certain parts of a CSS sheet to make sure that numbers aren’t read backwards.
It’s all about the feeling
Put simply, the goal of localization is to make the material feel as if it had been originally created in the target audience’s language. It’s about making things feel natural, feel familiar, feel native. This might sound vague, but how do you build up a certain feeling?
Through details. Localization has to be carried out with the second-nature understanding of a language that only a human native can have. It also involves playfulness, flexibility, decision-making with a specialist’s criteria, priority-setting, replacing or restructuring wordplay, and tastefully including slang. It involves the creativity and uniqueness that, for the time being, machines don’t have.
That’s why, through greatly (and increasingly) assisted by technology, localization professionals are unlikely to have their jobs automated in the short-term: If your profession relies on uniqueness, it’s safe from automation.