YAD: Leveraging T5 for Improved Automatic Diacritization of Yor\`ub\'a Text

TL;DR

This paper introduces YAD, a new Yorùbá diacritization benchmark, and demonstrates that a specialized T5 model trained on Yorùbá data outperforms multilingual models, with larger datasets and models improving accuracy.

Contribution

The paper presents a dedicated Yorùbá diacritization dataset and fine-tunes a T5 model specifically for Yorùbá, achieving superior performance over multilingual models.

Findings

YAD dataset enables effective evaluation of Yorùbá diacritization.

Yorùbá-specific T5 model outperforms multilingual T5 models.

Larger datasets and models lead to better diacritization results.

Abstract

In this work, we present Yor\`ub\'a automatic diacritization (YAD) benchmark dataset for evaluating Yor\`ub\'a diacritization systems. In addition, we pre-train text-to-text transformer, T5 model for Yor\`ub\'a and showed that this model outperform several multilingually trained T5 models. Lastly, we showed that more data and larger models are better at diacritization for Yor\`ub\'a