Awngi fastText Skip-gram Embeddings
This repository contains Awngi fastText skip-gram embeddings trained on a cleaned Awngi corpus.
Corpus Statistics
| Metric |
Value |
| Sentences |
153,341 |
| Tokens |
2,410,758 |
| Unique words |
194,469 |
Training Configuration
| Parameter |
Value |
| Algorithm |
Skip-gram |
| Dimension |
100 |
| Epochs |
20 |
| Window size |
5 |
| Character n-grams |
3โ6 |
Usage
import fasttext
model = fasttext.load_model("awngi_fasttext_skipgram.bin")
vector = model.get_word_vector("แ แแ")
print(vector[:10])
Intended Use
This model is designed for:
Named Entity Recognition
Low-resource NLP
Ethiopian language processing
Morphologically rich language representation
Citation
@article{andualem2026awnginer,
title={AwngiNER: A Benchmark Dataset and fastText-Enhanced Neural Models for Named Entity Recognition in Awngi},
author={Andualem, Amogne},
year={2026}
}
