Morphology generator systems require morphology disambiguation. I take a neural network approach in designing a morphology disambiguation process for Kinyarwanda (My native language).
Low-resource languages benefit from word-level information (Such as morphemes) to enhance information in their embeddings. Often, the morphology generation systems require morphology disambiguation (Determining) which segmentations generated are the correct ones.
In this code, I take a Neural Network approach to addressing this problem for Kinyarwanda (My native language).
This code is based on the paper: Kinyarwanda Morphology disambiguation (KinyaBERT: a Morphology-awa…)
## Improvements made:
- Included neighborhood information in feature extraction
- Embeddings were produced using a physics-inspired superposition of the morphemes embeddings (Tensorized embeddings)