Finding hard-coded strings, text that is cut off, and encoding problems before real translations exist
Pseudo-localization changes every translated string in a way that is easy to spot but still readable. Accents test character encoding and fonts. Extra characters make strings longer, to show layouts that cut off text. The brackets show where a string is truncated. Any plain English left on the screen is a hard-coded string that skipped the translation system.
"Machine translation" is the tempting wrong answer. A pseudo-locale doesn't translate anything. Its value is that developers find i18n bugs on day one, with no translators needed. Some teams also add a right-to-left pseudo-locale to test RTL layouts.
Read more