I spent fourteen months as the localization project manager on an RPG that shipped with about 1.4 million words of text across nine languages. I joined when the English script was maybe 60% finalized, which in game development means it was changing constantly. The parts that were supposedly final were about to get thrown out because the narrative director had a new idea about act three.
This is normal. That's what AAA game localization looks like from the inside. The question isn't whether the source text will change. It's whether your localization process can survive those changes without producing a product that contradicts itself in ways players will notice and post about on Reddit.
The scale problem
Main questline: maybe 80,000 words. Side quests: probably 200,000. Item descriptions: another 150,000. These are tricky because a sword described in act one might be referenced in act three, and the description and reference need to agree on name, function, and origin.
Then there's the lore. Codex entries, world history, character backstories, faction descriptions. Our game had a fictional religion with its own scripture — 30,000 words in English, 270,000 across nine languages. UI text, tutorial prompts, achievement descriptions, loading screen tips, crafting system labels...
No single person can hold all of it. Which is where things go wrong.
The item name problem
A sword called "Aethon's Edge" appears in quest dialogue, item description, codex entry, and a later side quest. In English, same name everywhere. In German: "Aethons Klinge" (quest), "Klinge des Aethon" (item), "Aethons Schneide" (codex), "Schwert von Aethon" (side quest). Four translators, four content modules, four names. None knew they were translating the same sword.
Players caught it within a week. German subreddit: "Does anyone else notice the sword has four different names?" Several hundred upvotes. Not catastrophic, but it breaks immersion.
How we fixed it (eventually)
Terminology management. The existing glossary had 400 entries for ~12,000 named elements. Building it out took months. We created a tiered system:
Tier 1 — main questline or 3+ appearances: mandatory locked translation in all languages.
Tier 2 — side quest + one other place: preferred translation, deviation allowed with justification.
Tier 3 — everything else: no requirement, but log choices in shared document.
Keeping it updated was the real challenge. An entire faction got rebranded four months before ship. Every change in English required updates across all localized files. Miss one and you've got a continuity error.
The project management layer
Tracking which files are in which state, which languages have which updates, which translators are waiting on terminology answers, which QA bugs need fixing — this is the layer that determines whether a million-word localization ships on time and intact.
At peak: ~70 translators across nine languages, plus editors, plus QA testers. A single English update might trigger updates across dozens of files per language, each needing tracking, scheduling, translation, editing, testing, and signoff.
What I wish I'd known
Start building the terminology database before translating. We lost six weeks to retroactive work. Invest in integrated QA, not bolted-on end testing. Give translators access to gameplay footage. And make sure the narrative and localization teams talk directly — not through tickets.
At Artlangs Translation, RPG game localization is where project management is half the battle. Consistency at scale requires tools, processes, and people who understand both creative game writing and logistical term management across languages. 230+ of them.
