10 Languages in 3 days: field notes from localizing an AI translation product (transept.ai)

In an ambitious localization effort, the AI translation tool Transept has successfully expanded from English to eleven languages in just three days, including German, Ukrainian, Chinese, and more. This process relied on machine translation supplemented by human insights, adhering to contextual relevance and terminology decisions—reflecting the very principles it promotes to users. The project illuminated significant localization challenges, including an exhaustive examination of plural rules, contextual nuances, and register disparities among languages. For instance, it was crucial to decide between formal and informal second-person terms in German and Spanish, which impacted verb conjugations and overall user experience. The team utilized machine translation post-editing (MTPE) effectively, ensuring that every translation was contextualized with notes to aid both human translators and the machine. The localization project revealed that translation quality hinges significantly on context, emphasizing that a contextual framework is vital for coherent output. This highlights a broader implication for the AI/ML community: as machine learning tools evolve, integrating linguistic context into AI systems will be essential for producing high-quality, nuanced translations, ultimately helping to bridge the gap between different linguistic and cultural user bases.