AfriHG: News headline generation for African Languages

TL;DR

This paper presents AfriHG, a new dataset for African language news headline generation, and demonstrates that specialized models outperform general multilingual models, with fine-tuned smaller models rivaling large LLMs.

Contribution

The creation of AfriHG dataset and comparative analysis of African language headline generation models, highlighting the effectiveness of Africa-centric models over multilingual models.

Findings

AfriTeVa V2 outperforms mT5-base in headline generation.

Fine-tuned AfriTeVa V2 with 313M parameters rivals Aya-101 LLM with 13B parameters.

Africa-centric models show superior performance in African language tasks.

Abstract

This paper introduces AfriHG -- a news headline generation dataset created by combining from XLSum and MasakhaNEWS datasets focusing on 16 languages widely spoken by Africa. We experimented with two seq2eq models (mT5-base and AfriTeVa V2), and Aya-101 LLM. Our results show that Africa-centric seq2seq models such as AfriTeVa V2 outperform the massively multilingual mT5-base model. Finally, we show that the performance of fine-tuning AfriTeVa V2 with 313M parameters is competitive to prompting Aya-101 LLM with more than 13B parameters.