Auditing a 28-Year-Old Japanese RPG

Most of my working days are spent on contracts and clinical documents. So it surprises people when I mention that one of my larger recent projects was a role-playing game.

The project was a pro bono, end-to-end localization QA of a Japanese RPG first released twenty-eight years ago. I audited its English localization against the Japanese original, with three things in mind: translation accuracy, consistency, and overall localization quality.

The size of it

On the Japanese side there were 548,421 source characters. On the English side, 219,330 words. Those words sat in 15,987 text boxes, spread across 595 field files.

For scale: my whole project history on ProZ.com records 303,822 words. The English side of this one game is about 72 percent of that figure.

Why games are hard to check

A contract is read from start to finish. A game is not. A player may meet a character in one town, leave for twenty hours, and meet them again in a file that was translated weeks apart. The text arrives in boxes, and each box carries only a few lines. Context is scattered across hundreds of files.

Which means the usual reading method does not work. You cannot hold 595 files in your head. You need lists: names, places, items, recurring phrases, and the one way each of them is spelled. Then you check every file against the lists, and you note every place the game drifts.

Accuracy inside a small box

Older games were often localized under tight space limits. A translator might have had to fit a whole sentence into a short line, and sometimes meaning was trimmed to make it fit. So an accuracy check on a game is never only "does the English match the Japanese?" It is also "was this cut on purpose, and did the cut cost the player something?"

What it gave back to my day job

A project like this is a strong argument for consistency as a discipline. In legal work I keep a termbase for exactly this reason: a defined term has to mean the same thing on page two and on page forty. A game simply makes the rule visible, because a player notices a renamed sword much faster than a client notices a renamed party.

It was also a reminder that QA is its own skill. Translating well and checking well overlap, but they are not the same job. Checking needs a method that does not depend on how good your memory is on a given afternoon.