Neural Machine Translation for African Languages
"Ukuxhumana" means "Communicate" in Zulu. This project is aimed at exploring ideas for using Neural Machine Translation for low-resource languages - specifically for the official languages of South Africa.
Our parallel corpuses are from the Autshumato project. The datasets contain data that was translated by professional translators, data that was sourced as translated file pairs from translators and data obtained from Government websites and documents
Two main architectures are used throughout this project, namely Convolutional Sequence to Sequence by Gehring et. al. and Transformer by Vaswani et. al. Fairseq(-py) and Tensor2Tensor were used in modeling these techniques respectively.
Results are given in BLEU.
English -> Language
|Convolutional Seq2Seq||27.77 (24.18)||0.62 (0.28)||15.35 (7.41)||36.96||16.17|
|Convolutional Seq2Seq (40K BPE)||23.83||1.44||4.89||34.28||21.06|
|Convolutional Seq2Seq (8K BPE)||2.19||15.45||26.78|
|Transformer (40k BPE) (uncased)||4.29|
|Transformer (40k BPE) (cased)||4.14|
|Transformer (8k BPE) (uncased)|
|Transformer (8k BPE) (cased)|
Language -> English
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size & hash||File type||Python version||Upload date|
|ukuxhumana-0.0.1-py3-none-any.whl (36.4 kB) View hashes||Wheel||py3|
|ukuxhumana-0.0.1.tar.gz (10.4 kB) View hashes||Source||None|
Hashes for ukuxhumana-0.0.1-py3-none-any.whl