CLRLC Logo
Langture AI

Langture-Bench

A Multilingual and Multicultural Corpus of African Proverbs

In progress

10+ languages · 10,000+ target entries · 50+ contributors

About

Proverbs carry the wisdom, humour and values of a culture, yet AI systems rarely understand them. Langture-Bench is a curated corpus of African proverbs, each with its literal and figurative meaning, validated by native-speaker linguists. It is also built as a benchmark to test whether language models understand culture, not just vocabulary.

Structure

What Each Entry Contains

1

The proverb in its original language

2

Its literal meaning

3

Its figurative meaning

4

A multiple-choice question that tests understanding of the figurative meaning

Coverage

10+ African languages, including:

Target: 1,000 proverbs per language

Process

How We Build It

01

Collection and Annotation

Native speakers collect proverbs and annotate their literal and figurative meanings.

02

Validation

A native-speaker linguist reviews each entry and writes the multiple-choice question.

Timeline

The Road Ahead

Now
Phase 1

Langture-Bench

Phase 2

Langture-Speech

Phase 3

LangtureLM

Leadership

Team

Joy Olusanya

Project Lead

Anusha Dixit

Co-Project Lead

Opeyemi Osakuade

Advisor

50+ Contributors

Data curators, annotators, validators and language leads

Get Involved

Speak one of these languages? Join Langture-Bench as a contributor or validator.