Software Localization Time by Strings

Software localization time estimate

Localization (l10n) goes well past plain translation. It's the cultural and technical work of fitting software to a target locale: UI strings, date/number/currency formats, plural rules, right-to-left scripts, image and colour conventions, even the legal copy. For a ballpark on capacity, use hours = (strings × avg_words) / words_per_hour. An experienced translator gets through something like 500–1,000 strings per day when the UI text is short, or around 2,000–3,000 words/day for longer prose.

The common exchange formats you'll run into are GNU gettext (PO/POT), XLIFF (XML Localization Interchange File Format, an OASIS standard), structured JSON and YAML, plus Android strings.xml and iOS .strings/.stringsdict. Plurals and gendered messages tend to go through ICU MessageFormat, which captures the rules each language needs (English has 2 plural forms, Russian has 4, Arabic 6).

Applications

An estimate like this lets you budget l10n sprints, brief a Language Service Provider (LSP) such as Lionbridge or TransPerfect, size up TMS platforms like Crowdin, Lokalise, Phrase, Transifex or Smartling, set up continuous localization in CI/CD, and quote freelance translators by the word or by the hour. Product managers also use it to weigh machine translation post-editing (MTPE) against full human translation.

FAQ

What is the difference between i18n and l10n? Internationalization (i18n) is the engineering side: getting a product ready for any locale by externalising strings, using Unicode, and relying on formatting APIs. Localization (l10n) is what happens afterward, adapting the product for each specific locale.

How many strings can one translator handle per day? Figure on 500–1,000 short UI strings, or 2,000–3,000 words of running text. Technical or legal material slows things down. Repetitive UI that gets good leverage from translation memory (TM) speeds them up.

Should I use machine translation? Modern neural MT (DeepL, Google, GPT-class models) does a great job on first drafts and on strings nobody scrutinizes. For marketing copy, legal terms and anything tied to brand voice, you'll still want a human reviewer in the loop, which is the MTPE workflow.