Transcreation and automation: Unthinkable, or newly possible opportunity?

It’s exactly as the title says. Is transcreation, the strongest and most treasured bastion of traditional language services, finally ready to embrace the power of machine translation?

There’s no need to riot just yet—this is definitely not a scenario where machines are coming to take over translators’ jobs. In fact, it’s just the opposite: they’re here to make it easier.

How? Well, first check out this recent episode of SlatorPod with Intento CEO Konstantin Savenkov. In this episode, Konstantin and host Florian Faes discuss a variety of topics such as Intento’s 2022 State of Machine Translation report, Google’s Translation Hub, large multilingual models, and generative AI.

Incidentally, you may also check out our Summary of the Intento State of Machine Translation report 2022 here.

Towards the end of the interview, Konstantin brings up an intriguing concept called Source Text Improvement.

What is source text improvement?

Now, the principle behind the concept is nothing new. It’s a well-documented fact that one way to improve machine translation performance is to feed it a source text that is optimized for machine translation. This means using simple language, breaking up long sentences into shorter ones, removing ambiguities, and the like.

But what *is* new behind the idea of source text improvement is the use of AI to do it. In this case, the use of a large language model like GPT-3 to automatically edit the source text into something more machine-translation friendly.

Large language models today are capable of an amazing range of feats. They are the product of research in a branch of AI called natural language processing, or NLP.

To learn more about NLP, read our article: Is natural language processing (NLP) taking AI into sci-fi territory?

GPT-3, in particular, has gotten so good and opened up a startling number of different use cases, from creating text summaries and automated reports to writing programming code and even full novels!

In fact, even before GPT-3 the technology’s already been out there for a while with the likes of Grammarly and the Hemingway app, which do the work of correcting grammar and simplifying text for readability. This is fairly close to what source text improvement is expected to do.

Where does transcreation come into the picture?

Now you might be thinking, what does all this have to do with transcreation? Going back to the podcast, Konstantin made a particularly sharp observation. Working with Japanese and German translators on American marketing material, he noted: “What those guys were doing was actually transcreating from the American locale rather than to other locales.”

Which is to say, there’s a hidden extra step in the transcreation process—the process of internationalizing the American language product before it is localized into a different language.

That hidden extra step is what could be automated by source text improvement. To make it easier to visualize, Konstantin provides an analogy: “When a sculptor makes a statue from rock, it’s much easier to make one from a big, solid rock rather than from another statue.” The internationalized text is much easier for creative language professionals to mold into an appropriate form in their own language.

Neat, isn’t it? This process means that machine translation can have a place in the transcreation workflow that aids human language professionals with next to zero drawbacks. Transcreators would no longer have to untangle foreign language quirks before letting their own creativity flow.

So far it seems that Intento is running some experiments with GPT-3 in this direction so hopefully we will hear more about their work with source text improvement in the near future.