Back to the Conference Program

Contents

Modern MT and LLM-based translation is now linguistically fluent. That era is over, as is the era of post-editing as the primary quality lever. When F5's Japan in-country team describes their LLM translations as "child language," the problem is cultural intelligence: register, tone, in/out-group framing. Approximately 85% of their rework has nothing to do with glossaries or terminology. This talk argues the new localization stack lives upstream - in structured authoring discipline, enriched terminology, and AI-ready style guides - and that technical communicators are already building it. It closes by naming and defining the emerging role that owns this function: the Loc Eval Engineer - the practitioner who brings evaluation discipline, drift detection, and cultural quality governance to the AI translation pipeline.

Takeaways

Your authoring discipline is already localization infrastructure, whether you've framed it that way or not: the structured authoring, terminology governance, and controlled language you practice in your CCMS are exactly the pre-editing discipline AI translation needs, and the shift is to invest deliberately in terminology records that carry register, context, and brand voice as machine-readable instructions rather than treating them as a documentation afterthought. That reframing matters even more once shadow localization pipelines enter the picture, because the instinct is to treat unsanctioned AI translation as a quality problem — flagging inconsistent output — when the argument that actually reaches the CFO is that there's no audit trail for AI-generated content shipping in your name, and under the EU AI Act transparency obligations and the Digital Product Passport regime (live since January 2026), missing provenance is a regulatory liability, not just a process gap. And once that governance case is made, the next move is realizing the emerging Loc Eval Engineer function — building quality estimation, LLM-as-judge calibration, and drift detection into a localization pipeline — doesn't require hiring from AI-land: documentation managers with terminology governance experience already cover roughly 60% of that skill set, so the lateral move is simply adding reference-free QE frameworks and calibration methodology to what you already own.