AfriHG: News headline generation for African Languages

2024-12-28

AI Summary

This work addresses the low-resource challenge of multilingual news headline generation for African languages. Method: We introduce AfriHG, the first high-quality, 16-language African news headline generation dataset, constructed by unifying XLSum and MasakhaNEWS. We systematically evaluate three approaches: fine-tuned mT5-base, zero-shot prompting of Aya-101 (13B+), and fine-tuning of our newly proposed Africa-centric small language model, AfriTeVa V2 (313M). Contribution/Results: AfriHG enables the first unified headline generation support across 16 African languages. AfriTeVa V2 achieves significant improvements over mT5-base across most languages and matches Aya-101’s zero-shot performance using less than 3% of its parameters—establishing new state-of-the-art results on multiple metrics. This work provides a benchmark dataset, an efficient model paradigm, and reproducible baselines for generative modeling in low-resource languages.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Language and Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

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.

Problem

• African Languages
• News Headline Generation
• Readability Enhancement

Innovation

• AfriTeVa V2
• multilingual news headline generation
• African languages