chuajeromepython/bert-spellchecker

How chuajeromepython/bert-spellchecker works

This repository is a local OCR-text correction tool: a terminal user submits text, SymSpell handles spelling and word-boundary errors, and BERT reranks candidates for contextual real-word errors before the result is displayed or saved. Separate scripts prepare domain text and fine-tune the BERT model; synthetic-case generation and accuracy scoring support evaluation. The main script’s excerpt omits some entry-point details, so those workflow links rely on the README; unsampled implementation details are not inferred.

Main entry points: BERT-spellchecker.py Correction CLI · run_accuracy_tests.py Pipeline accuracy scoring · prepare_finetune_data.py Training data preparation · train.txt Train and validation text

Components 13

Correction runtime

Correction CLI BERT-spellchecker.py
SymSpell correction BERT-spellchecker.py
BERT reranking BERT-spellchecker.py
Fine-tuned model finetuned-bert-spellchecker

Model preparation

Domain text finetune_data/raw
Training data preparation prepare_finetune_data.py
Train and validation text finetune_data/train.txt
BERT fine-tuning finetune_bert.py

Evaluation tools

Synthetic case generator gen_synthetic_cases.py
Accuracy cases accuracy_tests/test_cases.json
Pipeline accuracy scoring accuracy_tests/run_accuracy_tests.py
Score history accuracy_tests/results

Other components

Terminal user

Connections 14

Terminal user → Correction CLI: enters text README.md
Correction CLI → SymSpell correction: runs stage one README.md
Correction CLI → BERT reranking: runs stage two README.md
Correction CLI → Terminal user: prints or saves README.md
SymSpell correction → BERT reranking: passes corrected text BERT-spellchecker.py
BERT reranking → Fine-tuned model: loads model BERT-spellchecker.py
Domain text → Training data preparation: reads text prepare_finetune_data.py
Training data preparation → Train and validation text: writes splits prepare_finetune_data.py
Train and validation text → BERT fine-tuning: loads splits finetune_bert.py
BERT fine-tuning → Fine-tuned model: saves model finetune_bert.py
Synthetic case generator → Accuracy cases: generates cases
Pipeline accuracy scoring → Accuracy cases: loads cases accuracy_tests/run_accuracy_tests.py
Pipeline accuracy scoring → Correction CLI: loads and calls accuracy_tests/run_accuracy_tests.py
Pipeline accuracy scoring → Score history: saves scores accuracy_tests/run_accuracy_tests.py

Drawn by GitDiagram on November 12, 2026 from the repository's file tree, README and source files. Read as Markdown ·