license: apache-2.0
datasets:
- castorini/wura
language:
- afr
- amh
- arz
- eng
- fra
- hau
- ibo
- kin
- mlg
- nya
- orm
- por
- sna
- som
- sot
- swa
- tir
- xho
- yor
- zul
AfriTeVa V2 Base
AfriTeVa V2 Base is a multilingual T5 Version 1.1 model pretrained on Wura with a vocabulary size of 150,000. The model has been shown to improve over existing baselines on Text Classification, Machine Translation, Summarization and Cross-lingual Question Answering. The model has 428M parameters.
Paper: Better Quality Pretraining Data & T5 Models for African Languages
Authors: Akintunde Oladipo, Mofetoluwa Adeyemi, Orevaoghene Ahia, Abraham Toluwalase Owodunni, Odunayo Ogundepo, David Ifeoluwa Adelani, Jimmy Lin
NOTES:
- Dropout was turned off during pretraining and should be re-enabled for finetuning.
- Other checkpoints are available here.
Abstract
In this study, we highlight the importance of enhancing the quality of pretraining data in multilingual language models. Existing web crawls have demonstrated quality issues, particularly in the context of low-resource languages. Consequently, we introduce a new multilingual pretraining corpus for African languages, designed by carefully auditing existing pretraining corpora to understand and rectify prevalent quality issues. To compile this dataset, we undertake a rigorous examination of current data sources for thirteen languages within one of the most extensive multilingual web crawls, mC4, and extract cleaner data through meticulous auditing and improved web crawling strategies. Subsequently, we pretrain a new T5-based model on this dataset and evaluate its performance on multiple downstream tasks. Our model demonstrates better downstream effectiveness over existing pretrained models across four NLP tasks, underscoring the critical role data quality plays in pretraining language models in low-resource scenarios. Specifically, on cross-lingual QA evaluation, our new model is more than twice as effective as multilingual T5. All code, data and models are publicly available at castorini/AfriTeVa-keji.
Citation Information
@article{OladipoBQPD2023EMNLP,
title = "Better Quality Pre-training Data and T5 Models for African Languages",
author = "Oladipo, Akintunde and
Adeyemi, Mofetoluwa and
Ahia, Orevaoghene and
Owodunni, Abraham and
Ogundepo, Odunayo and
Adelani, David and
Lin, Jimmy
",
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
publisher = "Association for Computational Linguistics",
year = "2023",
}