Multilingual NLG — Generate Natural Text From Your Data

I will build a multilingual NLG system generating natural, grammatically correct text in multiple languages from a single data input — covering language-specific template systems for languages with complex grammatical agreement (gender, case, number), LLM-based generation for languages where templates are impractical, quality evaluation by native speaker criteria, and a unified data pipeline serving all language outputs from the same source data. Generating natural text across multiple languages requires more than machine translation of English output — grammatical agreement rules, idiomatic expression conventions, and cultural appropriateness vary substantially across languages in ways that translation of English-generated text consistently fails to preserve.

The system covers language-specific generation strategy (template vs. LLM per language), grammatical agreement handling for inflected languages, quality evaluation methodology per language, a unified data pipeline, and output delivery in your required format for each language.