A button labeled “Close” sits next to a door in the English build. In the Chinese strings file it becomes “靠近” — approach, draw near. Technically the dictionary supports it. On screen, with the door graphic and the interaction prompt, the choice is nonsense. Players stop, re-read, and the immersion snaps. The same spreadsheet can assign masculine forms to a female NPC because the name “Alex” or a gender-neutral source string gave no visual cue. Both mistakes survive careful desktop review. Both surface the moment a native speaker plays the actual build.
This is the gap linguistic testing, or LQA in game localization, is designed to close. Translation performed against isolated strings or even a well-maintained glossary still operates without the full scene: the character model, the UI layout constraints, the preceding dialogue branch, the emotional tone of the voice line, the exact moment a tutorial tip appears. Context arrives only when the text is live inside the game.
Industry experience has repeated the pattern for decades. Early Western ports of Japanese titles produced lines such as “All your base are belong to us” and “This guy are sick” not because translators lacked dictionaries, but because the work happened far from the playable product. Later projects still stumble on gender agreement, register mismatches, and polysemous UI verbs precisely when context notes or screenshots are thin. One documented case involved a protagonist named Alex whose lines were fully gendered male until LQA revealed the character model and narrative were female. The error was invisible in the table.
Native-speaker testers running the localized build catch what editors working from files cannot. They see whether a line fits the facial expression and body language on screen. They notice when a quest objective refers to an object that looks different from the one described. They feel when formal language appears in a casual banter scene or when a joke dies because the cultural trigger is missing. They also flag truncation, awkward line breaks, and font failures that only appear under real rendering conditions. Functional QA can confirm that text displays; linguistic testing confirms that the displayed text still works as language and as gameplay instruction.
The practical method is straightforward once the build is ready. Provide the LQA team with a test plan organized by systems and narrative beats, debug tools that jump to specific strings or states, the style guide and glossary, and — critically — enough play time to experience the text in sequence rather than as isolated tickets. Screenshots and video of problem areas accelerate reporting, but the decisive step remains sitting inside the game as a player of that language would. Severity ranking helps: critical issues that block progress or reverse meaning, major ones that break characterization or consistency, and minor stylistic notes that can be deferred if schedule is tight.
Data from player feedback reinforces the cost of skipping the step. Analyses of large review corpora show localization comments appearing in a meaningful share of posts, and negative remarks about awkward or incorrect text correlate with lower overall ratings and shorter session times in non-source languages. Studios that treat LQA as optional often discover the problems only after launch, when patches and public explanations become more expensive than the original testing would have been.
The deeper insight is that “correct” translation is not the same as playable translation. A string can match the source meaning, respect the glossary, and still produce the wrong player action or the wrong emotional response once visual and systemic context is restored. Linguistic testing restores that context before the audience does. It is the stage that converts a technically accurate localization into one that feels native to the people who will actually play it.
Artlangs Translation has spent more than twenty years refining exactly this process across game localization, short-drama subtitles, video and audiobook dubbing, and multilingual data work. With coverage of over 230 languages and a network of more than 20,000 professional linguists, the company routinely combines in-context linguistic testing with full localization pipelines so that context errors of the “Close / 靠近” variety are caught while they are still inexpensive to fix.
