---
license: mit

datasets:
  - DatarrX/Myanmar-Written-Spoken-Parallel-Corpus

language:
  - my

metrics:
  - bleu
  - chrf
  - ter
  - bertscore

base_model:
  - facebook/nllb-200-distilled-600M

pipeline_tag: text-generation

library_name: transformers

tags:
  - burmese
  - myanmar
  - myanmar-language
  - burmese-nlp
  - style-transfer
  - text-rewriting
  - formal-to-informal
  - written-to-spoken
  - seq2seq
  - nllb
  - lora
  - peft
  - low-resource-language
  - text-generation

model-index:
  - name: myX-TransStyle-W2S
    results:
      - task:
          type: text-generation
          name: Burmese Style Transfer (Written to Spoken)
        dataset:
          name: Custom External Test Set
          type: csv
          config: default
          split: test
        metrics:
          - type: bleu
            value: 19.6381
            name: BLEU
          - type: chrf
            value: 78.3975
            name: chrF
          - type: ter
            value: 50.7353
            name: TER
          - type: bertscore
            value: 0.9693
            name: BERTScore F1

---

# 📝 myX-TransStyle-W2S: A Transformer-based Style Transfer for Myanmar Written (ရေးဟန်) to Spoken (ပြောဟန်)

**myX-TransStyle-W2S** is a specialized Sequence-to-Sequence (Seq2Seq) model developed by **Khant Sint Heinn (Kalix Louis)** under **DatarrX**. It is specifically designed to transform formal **Written Burmese (ရေးဟန်)** into its natural colloquial **Spoken Burmese (ပြောဟန်)** counterpart. This model ensures that formal documents or news can be converted into fluid, human-like dialogue while maintaining 100% semantic integrity.

## Model Details

- **Developed by:** [Khant Sint Heinn (Kalix Louis)](https://huggingface.co/kalixlouiis)
- **Organization:** [DatarrX | ဒေတာ-အက်စ်](https://huggingface.co/DatarrX)
- **Model Architecture:** Fine-tuned NLLB-200 (600M Distilled) with merged LoRA adapters
- **Language:** Burmese (Myanmar)
- **Task:** Text Style Transfer (Written → Spoken)
- **License:** MIT
- **Trained on:** [Myanmar Written-Spoken Parallel Corpus (MWSPC)](https://huggingface.co/datasets/DatarrX/Myanmar-Written-Spoken-Parallel-Corpus)

---

## Linguistic Context: The Diglossia Challenge

Burmese is a **diglossic language**, featuring a major linguistic gap between two functional registers:

* **Written Style (ရေးဟန်):** Used in news, law, textbooks, and officialdom. It relies on formal grammatical markers such as **"သည်"**, **"၏"**, and **"၍"**.
* **Spoken Style (ပြောဟန်):** Used in daily life, verbal communication, and social media. It uses colloquial markers like **"တယ်"** (tense), **"ရဲ့"** (possessive), and **"နဲ့"** (conjunction).

**myX-TransStyle-W2S** addresses the "robotic" nature of modern AI by allowing formal text to be localized into the natural, warm tone used by native speakers every day.

---

## Training Methodology

The model was trained using an efficient adaptation strategy optimized for the unique structural shifts of Myanmar style.

### 1. The Dataset ([MWSPC](https://huggingface.co/datasets/DatarrX/Myanmar-Written-Spoken-Parallel-Corpus))
The model was trained on **5,555 high-quality, unique parallel text pairs**. This dataset provides a direct mapping from formal literary structures to their informal colloquial equivalents, filtered to ensure maximum diversity.

### 2. Parameter-Efficient Fine-Tuning (PEFT)
To capture nuanced stylistic shifts without overwriting the base model's linguistic depth, we utilized **Low-Rank Adaptation (LoRA)**:
* **Target Modules:** `q_proj`, `k_proj`, `v_proj`, `out_proj`.
* **Rank (R):** 32 | **Alpha:** 64.
* **Learning Rate:** 8e-5 with a Cosine scheduler.

### 3. Merging Strategy
The LoRA adapters were merged into the base `nllb-200-distilled-600M` model using `merge_and_unload()`. The resulting standalone **2.8 GB** model provides high-speed inference without requiring the PEFT library.

---

## Evaluation Results

The model was validated on **100 unseen test sentences** and showed superior performance compared to its S2W sibling.

### Performance Metrics
| Metric | Score | Interpretation |
|---|---|---|
| **BERTScore F1** | **0.9693** | Indicates near-perfect meaning preservation during style transfer. |
| **chrF** | **78.40** | Exceptional character-level accuracy, specifically in converting formal suffixes. |
| **BLEU** | **19.64** | Higher than S2W, reflecting a more consistent conversion pattern into spoken style. |

### Qualitative Analysis
Manual review by native speakers confirms the model's ability to not only swap particles but also adjust vocabulary (e.g., converting *“အလွန်ပင်”* to *“သိပ်”* or *“အကယ်ပင်”* to *“တကယ်လို့တောင်”*) in a way that feels authentic and human.

---

## 🔗 Related Models in the DatarrX Ecosystem

Explore other specialized models for Myanmar linguistic styles:

* **[myX-TransStyle-S2W](https://huggingface.co/DatarrX/myX-TransStyle-S2W):** The sibling model for converting Spoken Style to formal Written Style.
* **[myX-StyleClassifier](https://huggingface.co/DatarrX/myX-StyleClassifier):** Use this to automatically detect the style of your input text before processing.
---

## How to Use

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# 1. Load the Merged Model
model_id = "DatarrX/myX-TransStyle-W2S"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)

# 2. Prepare Input
prefix = "Rewrite Burmese formal written sentence into spoken Burmese: "
written_text = "ပုဂံခေတ်သည် မြန်မာနိုင်ငံသမိုင်းတွင် ပထမဆုံးသော အင်ပါယာနိုင်ငံတော်ကြီး ဖြစ်ခဲ့သည်။"
input_text = prefix + written_text

# 3. Generate Spoken Style
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(
    **inputs, 
    forced_bos_token_id=tokenizer.convert_tokens_to_ids("mya_Mymr"),
    max_length=160,
    num_beams=5
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))
# Output: ပုဂံခေတ်က မြန်မာနိုင်ငံသမိုင်းမှာ ပထမဆုံး အင်ပါယာနိုင်ငံတော်ကြီးဖြစ်ခဲ့တယ်။
```

---

## Intended Use & Limitations

### Use Cases
- **Natural AI Personalities:** Converting formal bot responses into natural-sounding speech.
- **Content Localization:** Making formal news or articles more accessible for audio/podcasts.
- **Creative Writing:** Assisting authors in converting narrative descriptions into natural character dialogue.

### Limitations
- **Dialectal Focus:** Primarily focuses on the standard Yangon/Mandalay dialect; regional slang may be less represented.
- **Contextual Nuance:** While meaning is preserved, the "warmth" of the spoken style may vary depending on the complexity of the input.

## Citation

### BibTeX
```BibTeX
@misc{myx_transstyle_w2s_2026,
  author = {Khant Sint Heinn (Kalix Louis)},
  title = {myX-TransStyle-W2S: A Written to Spoken Burmese Style Transfer Model},
  year = {2026},
  publisher = {Hugging Face},
  organization = {DatarrX},
  howpublished = {https://huggingface.co/DatarrX/myX-TransStyle-W2S}
}
```
---

## About the Author

**Khant Sint Heinn**, working under the name **Kalix Louis**, is a **Machine Learning Engineer focused on Natural Language Processing (NLP), data foundations, and open-source AI development**. His work is centered on improving support for the Burmese (Myanmar) language in modern AI systems by building high-quality datasets, practical tools, and scalable infrastructure for language technology.

He is currently the **Lead Developer at DatarrX**, where he develops data pipelines, manages large-scale data collection workflows, and helps create open-source resources for researchers, developers, and organizations. His experience includes data engineering, web scripting, dataset curation, and building systems that support real-world machine learning applications.

Khant Sint Heinn is especially interested in advancing low-resource languages and making AI more accessible to underrepresented communities. Through his open-source contributions, he works to strengthen the Burmese (Myanmar) tech ecosystem and provide reliable building blocks for future language models, search systems, and intelligent applications.

His goal is simple: to turn limited language resources into practical opportunities through clean data, useful tools, and community-driven innovation.

**Connect with the Author:**  
[GitHub](https://github.com/kalixlouiis) | [Hugging Face](https://huggingface.co/kalixlouiis) | [Kaggle](https://www.kaggle.com/organizations/kalixlouiis)

---
*Developed with ❤️ by [DatarrX](https://huggingface.co/DatarrX) to empower the Myanmar AI ecosystem.*