Spaces:
Sleeping
Download README.md from shah-bakhsh/BalPOS: direct link, hf CLI and curl.
- Browser
- Download file 4.68 kB
-
https://huggingface.co/spaces/shah-bakhsh/BalPOS/resolve/main/README.md
- Command line
-
hf download hf://spaces/shah-bakhsh/BalPOS/README.md
-
curl -L -o README.md https://huggingface.co/spaces/shah-bakhsh/BalPOS/resolve/main/README.md
A newer version of the Gradio SDK is available: 6.29.0
title: BalPOS - Balochi Part-of-Speech Tagger
sdk: gradio
emoji: π
colorFrom: yellow
colorTo: blue
pinned: true
---
title: BalPOS - Balochi Part-of-Speech Tagger
emoji: ποΈ
colorFrom: teal
colorTo: blue
sdk: gradio
sdk_version: 4.36.0
app_file: app.py
pinned: true
---
# ποΈ BalPOS - Balochi Part-of-Speech Tagger
**BalPOS** is an interactive Balochi Natural Language Processing system for automatic **Part-of-Speech (POS) tagging**.
Enter a Balochi sentence and the model predicts the grammatical category of each word using the **Universal Part-of-Speech (UPOS)** tagset.
## π Try BalPOS
### How to use
1. Enter or paste a Balochi sentence.
2. Click **Tag Sentence**.
3. Review the predicted POS tags.
4. Check the token-level confidence scores.
5. Try the example sentences provided in the Space.
The interface is designed for **desktop, tablet, and mobile devices**.
---
## π§ Model
**Model:** `shahbakhsh/BalPOS`
**Task:** Balochi Part-of-Speech Tagging
**Backbone:** BalBERT
**Tagset:** Universal Part-of-Speech (UPOS)
BalPOS is part of a broader research project focused on developing NLP resources and models for the Balochi language.
---
## π Evaluation Results
The final held-out evaluation reported the following results:
| Metric | Score |
|---|---:|
| Accuracy | **87.26%** |
| Balanced Accuracy | **78.81%** |
| Precision (Macro) | **79.56%** |
| Recall (Macro) | **78.81%** |
| F1 (Macro) | **78.99%** |
| Precision (Weighted) | **87.41%** |
| Recall (Weighted) | **87.26%** |
| F1 (Weighted) | **87.22%** |
| MCC | **0.8526** |
These results are from the project's final held-out evaluation set. They do not guarantee the same performance on every new Balochi sentence.
---
## π·οΈ UPOS Tags
BalPOS predicts 16 Universal POS categories:
| Tag | Description |
|---|---|
| `NOUN` | Noun |
| `PROPN` | Proper noun |
| `PRON` | Pronoun |
| `VERB` | Verb |
| `AUX` | Auxiliary |
| `ADJ` | Adjective |
| `ADV` | Adverb |
| `ADP` | Adposition |
| `CCONJ` | Coordinating conjunction |
| `SCONJ` | Subordinating conjunction |
| `DET` | Determiner |
| `NUM` | Numeral |
| `PART` | Particle |
| `PUNCT` | Punctuation |
| `INTJ` | Interjection |
| `X` | Other / unknown |
---
## π What You Can Test
You can test BalPOS with:
- Individual Balochi sentences
- Different grammatical structures
- Different vocabulary
- Punctuation
- Longer Balochi sentences
- Real-world Balochi text
- Different writing styles
The Space provides:
**Token β POS Tag β Confidence**
Example:
```text
ΪΫ β ADP
ΩΨͺΫ β PRON
ΩΩΨ³ΩΫ β NOUN
Ψ‘Ω β ADP
Ψ§Ψ³ΨͺΨ§Ψ― β NOUN
Ψ‘Ω β ADP
π§ͺ Testing and Feedback
If you find a potentially incorrect prediction, please record:
Sentence:
[complete Balochi sentence]
Word:
[problematic word]
Predicted POS:
[model prediction]
Expected POS:
[expected POS tag]
Additional note:
[optional explanation]
This type of feedback can help identify weaknesses and guide future improvements.
β οΈ Limitations
BalPOS is a research model and may produce incorrect predictions, particularly for:
- Rare words
- Unseen vocabulary
- Ambiguous words
- Different Balochi dialects
- Code-switched text
- Unusual spelling
- Text outside the training distribution
The displayed confidence represents the model's prediction confidence. It should not be interpreted as a guarantee that a prediction is correct.
For research or linguistic annotation, predictions should be reviewed by a qualified Balochi speaker or linguist.
π Balochi NLP Research
BalPOS is one component of a larger Balochi NLP development pipeline:
Balochi Corpus
β
Balochi Tokenizer
β
BalBERT
β
BalPOS
β
BalNER
β
BalMorph
β
BalParser
β
Balochi Language Model
The broader goal is to develop reusable NLP resources and models for Balochi.
π Resources
π€ Hugging Face Model
BalPOS
https://huggingface.co/shahbakhsh/BalPOS
π» GitHub Repository
BalPOS
https://github.com/shah-bakhsh/BalPOS
π Citation
If you use BalPOS in research, please refer to the model and project repository for the latest citation information.
π Acknowledgements
This project is part of ongoing work toward improving computational resources and NLP tools for the Balochi language.
Thank you to the Balochi speakers, researchers, annotators, and contributors supporting the development of Balochi NLP.
ποΈ BalPOS
Balochi NLP Β· Part-of-Speech Tagging Β· Low-Resource Language Technology