File size: 2,767 Bytes
cbd649b 4cded35 cbd649b 4cded35 cbd649b 4cded35 cbd649b 4cded35 cbd649b 4cded35 cbd649b 1eec673 cbd649b 1eec673 cbd649b 4cded35 cbd649b 4cded35 cbd649b 4cded35 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 | ---
language: tt
license: apache-2.0
datasets:
- TatarNLPWorld/tatar-morphological-corpus
metrics:
- accuracy
- f1
pipeline_tag: token-classification
tags:
- tatar
- morphology
- token-classification
- mbert
---
# Multilingual BERT (mBERT) fine-tuned for Tatar Morphological Analysis
This model is a fine-tuned version of [`bert-base-multilingual-cased`](https://huggingface.co/bert-base-multilingual-cased) for morphological analysis of the Tatar language. It was trained on a subset of **80,000 sentences** from the [Tatar Morphological Corpus](https://huggingface.co/datasets/TatarNLPWorld/tatar-morphological-corpus). The model predicts fine-grained morphological tags (e.g., `N+Sg+Nom`, `V+PRES(Й)+3SG`).
## Performance on Test Set
| Metric | Value | 95% CI |
|--------|-------|--------|
| Token Accuracy | 0.9905 | [0.9898, 0.9913] |
| Micro F1 | 0.9905 | [0.9897, 0.9913] |
| Macro F1 | 0.5563 | [0.5954, 0.6387]* |
*Note: macro F1 CI as reported in the paper.
### Accuracy by Part of Speech (Top 10)
| POS | Accuracy |
|-----|----------|
| PUNCT | 1.0000 |
| NOUN | 0.9905 |
| VERB | 0.9718 |
| ADJ | 0.9718 |
| PRON | 0.9918 |
| PART | 0.9986 |
| PROPN | 0.9779 |
| ADP | 1.0000 |
| CCONJ | 1.0000 |
| ADV | 0.9948 |
## Usage
```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch
model_name = "TatarNLPWorld/mbert-tatar-morph"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
tokens = ["Татар", "теле", "бик", "бай", "."]
inputs = tokenizer(tokens, is_split_into_words=True, return_tensors="pt", truncation=True)
outputs = model(**inputs)
predictions = torch.argmax(outputs.logits, dim=2)
# Get tag mapping from model config
id2tag = model.config.id2label
word_ids = inputs.word_ids()
prev_word = None
for idx, word_idx in enumerate(word_ids):
if word_idx is not None and word_idx != prev_word:
tag_id = predictions[0][idx].item()
if isinstance(id2tag, dict):
tag = id2tag.get(str(tag_id), id2tag.get(tag_id, "UNK"))
else:
tag = id2tag[tag_id] if tag_id < len(id2tag) else "UNK"
print(tokens[word_idx], "->", tag)
prev_word = word_idx
```
Expected output (approximately):
```
Татар -> N+Sg+Nom
теле -> N+Sg+POSS_3(СЫ)+Nom
бик -> Adv
бай -> Adj
. -> PUNCT
```
## Citation
If you use this model, please cite it as:
```bibtex
@misc{arabov-mbert-tatar-morph-2026,
title = {Multilingual BERT (mBERT) fine-tuned for Tatar Morphological Analysis},
author = {Arabov Mullosharaf Kurbonovich},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/TatarNLPWorld/mbert-tatar-morph}
}
```
## License
Apache 2.0 |