Uhura: A Benchmark for Evaluating Scientific Question Answering and Truthfulness in Low-Resource African Languages

TL;DR

This paper introduces Uhura, a benchmark for evaluating the performance and truthfulness of large language models in six low-resource African languages across scientific and safety-related tasks, highlighting significant performance gaps and the need for improved multilingual NLP in these languages.

Contribution

The paper presents Uhura, a new benchmark created through human translation for low-resource African languages, addressing the lack of datasets and evaluating models on scientific and safety tasks.

Findings

Models perform significantly worse in African languages compared to English.

Proprietary models outperform open-source models in the benchmark.

All models show increased false claims in low-resource languages.

Abstract

Evaluations of Large Language Models (LLMs) on knowledge-intensive tasks and factual accuracy often focus on high-resource languages primarily because datasets for low-resource languages (LRLs) are scarce. In this paper, we present Uhura -- a new benchmark that focuses on two tasks in six typologically-diverse African languages, created via human translation of existing English benchmarks. The first dataset, Uhura-ARC-Easy, is composed of multiple-choice science questions. The second, Uhura-TruthfulQA, is a safety benchmark testing the truthfulness of models on topics including health, law, finance, and politics. We highlight the challenges creating benchmarks with highly technical content for LRLs and outline mitigation strategies. Our evaluation reveals a significant performance gap between proprietary models such as GPT-4o and o1-preview, and Claude models, and open-source models like Meta's LLaMA and Google's Gemma. Additionally, all models perform better in English than in African languages. These results indicate that LMs struggle with answering scientific questions and are more prone to generating false claims in low-resource African languages. Our findings underscore the necessity for continuous improvement of multilingual LM capabilities in LRL settings to ensure safe and reliable use in real-world contexts. We open-source the Uhura Benchmark and Uhura Platform to foster further research and development in NLP for LRLs.