Future Roadmap¶
This document outlines planned features for NepaliKit releases.
v1.0.5: Lemmatizer ✅ Completed¶
Dictionary-based Lemmatizer¶
Status: Released (July 14, 2026)
Dictionary-based lemma lookup with rule-based fallback. 340+ dictionary entries, morphological rules for common Nepali suffixes.
v1.0.6: ML POS Tagger ✅ Completed¶
ML-based POS Tagger¶
Status: Released (July 14, 2026)
Context-aware POS tagger using n-gram statistics and Viterbi decoding. Bigram transition model, emission probabilities, suffix rules, and context pattern boosting. Zero-dependency pure Python implementation.
v1.0.7: Named Entity Recognition¶
NER Module¶
Priority: High Gap: No NER capability at all.
Implementation:
- Create nepalikit.ner module
- Optional torch/transformers dependency
- Bundle pre-trained NER model (PERSON, LOCATION, ORGANIZATION, DATE)
- Lazy-load to keep zero-deps default
v1.0.8: Word Embeddings¶
Word Embeddings¶
Priority: Medium Gap: No word embedding support.
Implementation:
- Create nepalikit.embeddings module
- Optional fasttext/gensim dependency
- Lazy-load pre-trained Nepali fastText/Word2Vec
- Support similarity, nearest neighbors, vector ops
v1.0.9: Translation¶
Translation¶
Priority: Medium Gap: No translation support.
Implementation:
- Create nepalikit.translation module
- Optional deep-translator or transformers dependency
- Support Nepali to English and English to Nepali
v1.0.10: Synonym Generator¶
Synonym Generator¶
Priority: Medium Gap: No synonym support.
Implementation:
- Create nepalikit.synonyms module
- Use pre-trained embeddings for nearest-neighbor synonyms
- Return ranked suggestions
v1.0.11: Spell Checker v2¶
Spell Checker Improvements¶
Priority: Medium Gap: Current checker is basic. Could add n-gram frequency model and context-aware corrections.
Implementation: - Add n-gram frequency model - Add context-aware suggestions - Add batch spell check for documents
v1.1.0: Ecosystem¶
Dataset Loaders¶
Priority: Low
Implementation:
- Create nepalikit.datasets module
- Download/cache from Hugging Face Hub
- Support: News, Sentiment, NER, QA, POS datasets
Model Hub Integration¶
Priority: Low
Implementation:
- Create nepalikit.model_hub module
- Lazy-load Hugging Face models
- Support: NepNewsBERT, Sentiment, Sentence Similarity
Benchmark Suite¶
Priority: Low
Implementation:
- Create nepalikit.benchmarks module
- Evaluate on Nepali NLP tasks
- Standardized evaluation metrics
Summary¶
| Version | Feature | Priority | Status |
|---|---|---|---|
| 1.0.5 | Lemmatizer | High | ✅ Completed |
| 1.0.6 | ML POS Tagger | High | ✅ Completed |
| 1.0.7 | Named Entity Recognition | High | Planned |
| 1.0.8 | Word Embeddings | Medium | Planned |
| 1.0.9 | Translation | Medium | Planned |
| 1.0.10 | Synonym Generator | Medium | Planned |
| 1.0.11 | Spell Checker v2 | Medium | Planned |
| 1.1.0 | Dataset Loaders | Low | Future |
| 1.1.0 | Model Hub | Low | Future |
| 1.1.0 | Benchmark Suite | Low | Future |
How to Contribute¶
We welcome contributions for these features!
- NER: Build transformer-based NER with lazy loading
- Embeddings: Create fastText/Word2Vec loaders
- Translation: Wrap translation APIs/models
- Datasets: Add download helpers for public datasets
Guidelines:
- Use lazy imports (try/except ImportError)
- Keep core zero-deps
- Document all optional dependencies
- Include examples in docs
- Write tests
Last Updated: July 2026