BSSpell is a sentence-level Bangla spelling correction model built on a Transformer-based encoder-decoder architecture, enhanced through Knowledge Distillation from a grammatical error correction task. It achieves 97.02% accuracy on a custom 3.8M sentence corpus.
Figure: BSSpell Architecture – Pretrained on BGEC, fine-tuned on BanglaSSECorpus.
Title: BSSpell: A Monolingual Transformer-Based Sentence Level Spelling Error Correction Method for Bangla
Authors: Ajoy Dey, Muhammad Abdullah Adnan
Affiliation: Department of CSE, BUET
Link: [Coming soon]
| Model | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| T5-Small | 94.09% | 94.5% | 93.33% | 93.56% |
| BSSpell | 97.02% | 96.5% | 98.7% | 96.01% |
| Dataset Size | Accuracy | F1 Score |
|---|---|---|
| 1.3M | 92.09% | 0.92 |
| 2.9M | 94.09% | 0.93 |
| 3.8M | 97.02% | 0.96 |
BanglaSSECorpus
# Clone this repository
git clone https://github.com/ajoydey00001/BSSpell.git
cd BSSpell
# Install requirements
pip install -r requirements.txt
# Train the model
python train.py --config configs/bsspell.yaml
# Evaluate the model
python evaluate.py --model checkpoints/bsspell_best.pt
BSSpell is a sentence-level Bangla spelling correction model built on a Transformer-based encoder-decoder architecture, enhanced through Knowledge Distillation from a grammatical error correction task. It achieves 97.02% accuracy on a custom 3.8M sentence corpus.
Figure: BSSpell Architecture – Pretrained on BGEC, fine-tuned on BanglaSSECorpus.
Title: BSSpell: A Monolingual Transformer-Based Sentence Level Spelling Error Correction Method for Bangla
Authors: Ajoy Dey, Muhammad Abdullah Adnan
Affiliation: Department of CSE, BUET
Link: [Coming soon]
| Model | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| T5-Small | 94.09% | 94.5% | 93.33% | 93.56% |
| BSSpell | 97.02% | 96.5% | 98.7% | 96.01% |
| Dataset Size | Accuracy | F1 Score |
|---|---|---|
| 1.3M | 92.09% | 0.92 |
| 2.9M | 94.09% | 0.93 |
| 3.8M | 97.02% | 0.96 |
BanglaSSECorpus
# Clone this repository
git clone https://github.com/ajoydey00001/BSSpell.git
cd BSSpell
# Install requirements
pip install -r requirements.txt
# Train the model
python train.py --config configs/bsspell.yaml
# Evaluate the model
python evaluate.py --model checkpoints/bsspell_best.pt