Gemini Afro MT Bench

Gemini × African Languages Corpus-Grounded MT Benchmark
GitHub Code & Data

This dashboard evaluates how accurately Google Gemini translates vocabulary into 645 African languages. Rather than relying on model self-grading or synthetic tests, each translation is directly verified against human-authored parallel literature from the YouVersion corpus (AfriSpeech/africa-corpus) across 877 African language Bible editions.

🎯 How It Works Gemini is asked for the everyday translation of 204 concepts (nouns, adjectives, numerals). A translation is marked verified only if it appears in actual written sentences for that word in the language's parallel corpus.
📊 Common Core Score The Core Score standardizes rankings across 195 universally attested concepts present in virtually all African Bibles, ensuring large national languages and smaller dialects are compared fairly.
🏷️ Capability Tiers
Strong (≥60%) Established vocabulary
Medium (30–60%) Everyday coverage, some gaps
Weak (<30%) Emerging or fragmented knowledge
Unsupported (0%) No verified written vocabulary

Macro patterns across 645 African languages: performance across word types (Parts of Speech), tier distributions, geographical regions, and word frequency bands.

Browse and compare all 645 languages. Search by language name or ISO 639-3 code, sort by standardized Common Core Score, or filter by African region and language family.

Compare language performance across linguistic categories: basic concrete nouns, physical adjectives, and cardinal numerals.

Select any language to inspect its performance profile, linguistic metadata, and individual translations side-by-side with verified parallel Bible verses.