Scaling localization at Atlassian: keeping translation at the pace of AI-era development

This post details the expansion of Atlassian's AI-Assisted Localization capabilities by introducing a two-pronged approach: AI drafting of translations and tooling to fix source code issues before translation. The AI drafting process involves gathering context (developer notes, similar strings, glossary terms, style rules) to generate initial translations, which are then reviewed and approved by professional translators. The system tracks Translation Edit Rate (TER) to measure draft quality and provide feedback for model improvement, aiming for up to 50% cost savings. The internationalization (i18n) validation system catches and guides developers to fix source code problems like untranslated text, fragmented sentences, and missing plural rules. This includes providing richer context in string descriptions for both AI and human reviewers. Over 20,000 existing issues were cleared by AI-driven bulk fixes and code review guidance.