The latest LocWorld event had a lot to say about AI in localization, reflecting the industry’s evolving relationship with all things AI.
Recent conferences have left attendees with more questions than answers, but LocWorld53 Malmö proved to be a turning point, as sessions and talks offered real-world case studies illustrating a clear push toward practical implementation tempered by measured skepticism.
Here’s a breakdown of what stood out and why.
Contents
AI adoption: From experiment to infrastructure
Several sessions focused on how organizations have been integrating AI into their localization workflows. While pilot programs utilizing AI were not uncommon, there is now a clear shift toward establishing AI as core infrastructure. This is especially true for organizations that work with high-volume content and fast turnarounds, where custom MT engines, hybrid translation setups and integrated LLM workflows are becoming increasingly common.
This change stands out because it underlines the growing trend of weaving AI into the very fabric of localization operations to effect greater cost-savings. Worth noting, however, is the fact that scalability and quality can vary widely depending on the types of content and the target market. Organizations will need to be strategic in developing custom solutions for their particular needs and objectives.
Not all MT or LLMs are created equal
As the industry gains more familiarity with the various tech solutions on offer, comparative testing is beginning to confirm what many users have come to understand at least anecdotally—that there is a clear variation in performance among NMT engines and LLMs across dozens of languages.
Some NMT engines excel at transactional or structured content, but struggle with tone, nuance, and voice. While LLMs have demonstrated a high degree of fluency, they often lack predictability and consistency, which proved particularly clear in enterprise-grade workflows where contextual accuracy is highly weighted.
For the time being, businesses looking to scale global content need to carefully consider their NMT and LLM options and treat the process of adopting them into workflows as ongoing optimization projects rather than once-and-done solutions.
End-to-end automation is possible…
This edition of the conference featured sessions highlighting organizations that have successfully built fully integrated AI-first workflows. The systems described connect content sources directly to translators, with AI handling intake, routing, and QA processes. Notable results included faster time-to-market, substantial operational savings, and increased bandwidth for strategic planning.
While such AI-first systems offer a significant reduction in localization costs and turnaround times, they require careful alignment across business, tech, and linguistic teams. Poorly executed automation still results in inefficiencies or requires work to be redone.
You still need human translators
While there is a lot of excitement around automation right now, many presenters emphasized the need for a layered strategy in which AI assists but does not replace human experts. Although there were many stories of organizations implementing highly automated systems, their success has been hinged on human supervision. Content that is complex or highly creative requires human sensibilities to refine tone, maintain brand voice, and ensure cultural integrity.
No systems are truly AI-only—they’re AI-assisted. Stakeholders need to recognize that human-in-the-loop models aren’t going anywhere, as they are essential for maintaining quality, especially when nuance and regulatory compliance are paramount.
The role of the linguist is evolving
Another theme that persisted at this edition of the conference was the ever-growing collection of hats that linguists are being asked to wear. Language professionals are being asked to shift away from traditional translation work toward new roles such as quality trainer, prompt designer, cultural advisor, and LLM reviewer. The strategy seems to be moving from “fixing output” to “teaching the machine.”
The upshot here is that LSPs and localization teams need to invest in upskilling and redefining metrics for success. While fluency and fidelity are important, readability, impact, and brand resonance can’t be ignored when it comes to content crafted with AI.
Emerging challenges
How organizations address challenges is always a topic of interest at LocWorld. The advent of AI has offered promising solutions, while casting challenges into stark relief. One issue faced by organizations trying leverage AI to expand the number of languages they cover is the clear gap in performance with low-resource or culturally complex languages. Not surprisingly, the quality and security of AI output for low-resource languages drops considerably when compared to high-resource languages like English, Spanish, or French.
This means there should be plenty of opportunities for specialized providers and linguists fluent in low-resource languages. Specially tuned models, rapid iterations cycles, and real-time evaluation systems should aid in addressing these long-tail language markets.
Final thoughts
Overall, LocWorld53 Malmö painted a picture of an industry growing with technology and beginning to better grasp the potential benefits and pitfalls of its new powers. Despite the shine that AI still has, many industry concerns remain evergreen—quality, ethics, teamwork, value, and accessibility.
