---

In one regard, the localization industry is quite used to this. Over the past 30 years, we’ve seen numerous technologies arrive on the scene that were guaranteed to completely upend the industry, take everyone’s job, and annihilate vendors once and for all. Rules-based machine translation (MT), translation management systems, statistical MT, automated LQA and I18N testing, neural MT, and massively multilingual MT engines were all expected to coldly pummel the human element out of localization. The news of our demise was greatly exaggerated. Welcome to the party, generative AI!

It’s August, 2023. The generative AI (GenAI) fanfare is mercifully ebbing. A recent webinar alluded to the “tide of hype receding” — a great and useful visual. It means we can all begin to search for what this most recent wave left on the beach: to piece together how we’ll effectively use GenAI as the tool that it is.

In addition to some practical thinking around GenAI and localization, this article contains two predictions. The predictions don’t have to be correct; they’re made in the hope that you’ll be better prepared for the GenAI possibilities that eventually land at your company. As every company has a unique set of requirements that fulfill its global content needs, be playful as you read this. Think about how the discussion points can apply to you and your company’s current and future work.

GenAI captures the imagination more than MT ever did

While both included their own type of hype, the difference between the advent of MT versus GenAI is striking. Apart from translating song lyrics from English into Bulgarian and back into English, or using Google Translate to cheat on high-school language essays, the general public didn’t spend a lot of energy on MT. In the business world, a new and still problematic security challenge — employees using free MT engines to better understand company-internal communications — was born. The most substantive (and never-ending) conversations about what MT means for the future, however, have always taken place within the hallowed halls of the localization industry.

That’s not GenAI.

GenAI has captured way more public mindshare than MT ever did. Everyone is talking about it. Prompt engineering was suddenly a thing. Hundreds of millions of dollars have been added to the valuations of companies allegedly poised to cash in. For goodness sake, my wife used GenAI to write limericks to differentiate her demand gen emails. The widespread talk, first usage, and speculation are mind-numbing and ubiquitous.

Your manager and C-suite are caught up in this hype, suffer FOMO, and want answers. How might a localization professional who has lived through other stages of hype talk to them about GenAI? Below are some practical statements that lower the temperature of this overly hot topic and help you speak to the inevitable questions from management.

Talking points for GenAI

GenAI is a tool. GenAI is NOT a panacea.

We will all learn how to best use this latest technology tool in the next few years. While GenAI means great things for content authoring and localization, no one knows its precise impact today.

I can say one thing for sure about GenAI: Humans will always be in the loop. GenAI will always have limitations and its output will always need to be reviewed. Please don’t fire our legal, marketing, product, and localization personnel who are currently responsible for creating and reviewing content. Humans will have the hard job of figuring out the relevancy of generated content and where/how it can be best applied.

I look forward to learning how we can continue to become even more efficient by using GenAI. Similar to MT — something we have a lot of experience with — our success with GenAI will be iterative. Remember that it took years to learn how to properly leverage the promise of MT. Though the velocity of GenAI improvement is pretty staggering, it’s still going to take time to get the use cases right.

For localization, it’s likely we’ll find that MT and current workflows still work best for certain use cases. Other use cases will be best served by GenAI, and they’ll require new and different workflows.

Prediction: The tool everyone will use

This prediction is less than outrageous because multiple companies are developing similar systems as I type.

In a year or three, most companies will have a single-platform GenAI tool. Naturally, the tool can generate copy in any language.

In the process of generating the content, the tool will scan previously approved content to better inform the generated output. The tool will also reference the company’s linguistic assets — translation memories, style guides, and terminology databases.

Within the platform, reviewers will review/edit the generated content.

Integrated with publishing systems, the platform will push the post-edited content (in any language) for publication.

In the review/edit phase, the tool highlights non-conformances of the GenAI output with previously approved content or linguistic assets. Reviewers will correct any non-conformance. The corrected content will then become part of the reference data for future generations (that’s “future generations” as in what will be generated as content in the future, not your children’s children). This process probably looks extremely familiar to you — MT post-editors working with adaptive engines have done this type of work for several years.

An interesting aspect of this new process is that the concept of the global content workflow — source creation in a content management system (CMS) → translation management systems → CMS → publication of all languages — will be eliminated or drastically changed. The punchline is that both localization technology and CMS providers must ensure seamless management and publication of “sourceless” content in multiple languages.

A more speculative prediction about translators’ work and GenAI

For the translators reading this, the answer to the question you think I will discuss is this: No, GenAI will not take your job. I know some of you still can’t believe that neural MT didn’t replace you, but here we are. Here you are.

This speculative prediction pertains to how translators might expand their roles in the GenAI world. Take as given the continued need for translator expertise in translation, reviewing, editing, transcreation, data annotation, neural algorithm scoring, etc. This prediction is a potential “plus one.” Translators have my permission to love it, hate it, or ignore it.

With linguistic expertise still at the core, I can envision a GenAI world where translators specialize in a language and choose to develop a deeper understanding of a product-specific legal, business, or commercial aspects of their locale. A simple example:

It’s 2026, and a US-based company selling baby formula worldwide has the GenAI tool described above. The Brazilian Portuguese translator/reviewer/editor has developed some expertise in Brazilian commercial law pertaining to information that must be included when marketing baby formula. In their queue, they see the tool’s latest generated marketing blurb (marked “urgent”) for resellers in Brazil. The translator judges the content to be linguistically sound.

The translator recognizes, however, that some of the wording is not in compliance with a recently passed Brazilian law. The translator contacts legal, marketing, and the department responsible for ensuring the AI engine is trained with recent, important legal data. The translator’s relative expertise means the matter was resolved more quickly than it would have been had they waited for everyone to complete their individual reviews.

Even though the translator in the example does not have a legal degree, their deeper understanding of Brazilian law clearly helps the entire process. While a translator’s “domain expertise” can be seen as a linguistic parallel, the concept I’m suggesting falls outside the scope of pure language work. It’s best thought of as highly specific “locale expertise,” which combines linguistic expertise with some semblance of legal, marketing, or other commercial knowledge. Translators possessing such knowledge ensure that generated content is — from both a linguistic and a commercial/legal business standpoint — correct.

Before you object, remember that this content is being generated. Unlike the old days, August 2023 for example, the content is not being authored by marketing, reviewed, edited, reviewed, sent to product for review, returned with comments, sent to legal for review, approved, sent to localization, returned, and published simultaneously in 20 languages. In the GenAI world, translators who develop practical expertise in one of the many locale-specific business areas will create more value. They will make the entire review process of generated content more efficient.

One extremely smart and practical sentence

I recently attended a webinar — one of hundreds, it feels like — about the future of localization and GenAI. The webinar started with a large panel discussion where a panelist dropped the comment that “Everyone should start working to make their style guides (in all languages) as good as possible.” It was almost treated as a throwaway comment, missed as others rushed to speak about topics more speculative.

The brilliance of this simple comment is that it informs everyone — the localization industry in particular — how to best prepare for a GenAI future. To efficiently create the most useful content possible, GenAI needs to know what you have said and how you like to say it. It must reference your company’s tone, style, preferred usage, and terminology. Clean, reliable linguistic assets are now more important than ever. Why not start by making your style guides better for all languages?