How AI Translators Benefit From a Human Touch

Who wrote the acclaimed, surrealist novel One Hundred Years of Solitude? Gabriel García Márquez, of course! Well, although correct, the answer is not exactly precise. 

The Colombian-born García wrote Cien Años de Soledad, which was translated from Spanish into English by a well-respected Queens College Professor, Gregory Rabassa, who had recently translated Julio Cortázar’s Hopscotch, and had won a National Book Award for it. And the choices Rabassa made in translating García’s Nobel-prize-winning novel won him the praise of its author who famously declared the English translation superior to the Spanish original. 

With famous literary translations as a backdrop, we’ll explore in this post the art of language translation and demonstrate how that art refines the science behind machine translations. 

Language Liberties + Decisions, Decisions, Decisions

Language translation isn’t as direct as mathematics, and 2+2 doesn’t always equal 4 in the literary realm. When translators approach a text, they are confronted with a variety of decisions on how to best capture the author’s words, voice, spirit, and intent. And that job starts with the title. 

Let’s look back at Cien Años de Soledad. If you remember your grade school Spanish (don’t we all), then you know cien translates to one hundred. BUT it could also translate to a hundred. So right from the title, Rabassa had to make a decision that he believed was most fitting for García’s epic tale. And while one hundred and a hundred have the same meaning, they are indeed different. Like “azure��� and “sky blue”, they are subtle shades of the same primary color. 

We can go even further by examining the philosophical novella on the must-read list of every high school — The Stranger by Albert Camus. But, again, if we’re being precise, the French author titled his tale of senseless murder and apathy — L’Étranger. Stuart Gilbert, a British scholar, was the first to translate the book into English, which he adapted to The Outsider. 

Gilbert proceeded to translate Camus’ opening sentence — Aujourd’hui, maman est morte —  as Mother died today. In 1982, both Joseph Laredo and Kate Griffith produced English translations of L’Étranger but revised the title to The Stranger while still keeping Gilbert’s first line intact: Mother died today. 

However, many American high schoolers read a translation of The Stranger done in 1988 by the American translator and poet Matthew Ward, who changed the opening line to Maman died today, reverting Mother to the original French Maman, which is not as detached as saying Mother, yet not as childlike as Mommy. Ward made the decision to revert to the original French, even for an English translation, believing that his English-speaking readers would still comprehend it. 

It’s worth noting that none of the translators went for the more literal, dictionary translation: Today, Mother has died. Readers can even debate if this colder version might actually be better aligned with the author’s original intention, given that the book’s protagonist is himself emotionally unavailable. 

As seen from just a few words of both One Hundred Years of Solitude and The Stranger/The Outsider, translators are faced with decisions as to how best to translate each word they come across. Their choices can directly influence the essence of the piece and how it’s received by the reader. This is the art of language translation. 

Art Meets Science 

We live in a digital world full of advanced artificial intelligence and machine learning that businesses globally are adopting more and more into their workflows when seeking to enter new markets and languages.

Let’s take Unbabel, for example — an AI-powered LangOps-platform that is purpose-built for multilingual marketing and customer service interactions. What sets Unbabel apart is its recognition of the art inherent to language translations and how we apply it to the science behind machine translations. And key to these efforts are our human editors. 

We customize Unbabel’s machine translation for each brand by uploading glossary terms and localizing translations to each market so that translations account for cultural nuances and idioms which, if neglected, can make brands memorable for all the right reasons. And when machine translations are supported by human refinement, the result is fluent communications between customer and support agent. 

To achieve this, we rely on a global community of human editors and professional linguists who provide our developers with feedback and annotations on communications between customer and agent, which we use to continuously train and upgrade the AI so that it speaks both the brand’s language and the customers’. And because we incorporate feedback from many translators (rather than relying on just a single voice or perspective), our AI better reflects the way people within a specific market generally converse and maintain conversations with one hundred — or a hundred — percent accuracy.

Human-in-the-loop AI fuses the best of art and science, allowing brands to deploy language translations at scale for seamless communications and delightful customer experiences. 

Delve deeper into Why Business Translation Is Vital for the Global Customer Experience: