DatarrX/myX-myanmar-to-myanglish-corpus

📝 Myanglish (မြန်းဂလိ) corpus

A parallel text corpus containing high-quality, human-curated pairs of native Burmese (Myanmar Unicode) text and its corresponding Myanglish (Burmese Romanization) transliterations.

The initial high-quality release contains 2,121 fully approved rows designed to bridge the gap between formal script and the informal romanized phonetic typing structures widely used across social media, chat applications, and digital communication in Myanmar.

---

🎯 Project Purpose & Core Intent

"Myanglish" (မြန်းဂလိ) is the colloquial term for writing Burmese words using the Latin/English alphabet based on phonetic sound rather than formal translation. Because a large portion of daily online communication within the Burmese digital ecosystem happens via Myanglish, this corpus serves as an essential stepping stone for training machine learning models in informal machine translation, social media text normalization, and intent detection.

---

⚠️ Critical Linguistic Context & Potential Biases (Read Before Training)

For international/non-Burmese NLP engineers, it is absolutely crucial to understand the inherent nature and limitations of Myanglish represented in this initial release:

1. The Absence of a Standardized Orthography

There is no official or universally accepted standard for writing Myanglish. Spelling variations depend entirely on the individual typist's preference, regional accents, or intuitive phonetics.

• Phonetic vs. Orthographic Shifting: In formal Burmese, words often shift sounds when spoken. For example, the greeting "မင်္ဂလာပါ" (Mingalabar) is spelled with the character "ပါ" (par), but phonetically pronounced with a heavier sound as "ဘာ" (bar). Some users will type it exactly as it is spelled (Mingalapar), while others type it exactly as it sounds (Mingalabar). Both variants are perfectly understood by native speakers.
• Multiple Latin Mapping Paths: A single Burmese syllable can often be mapped to Latin text via multiple phonetic spelling choices (e.g., the syllable "ဆ" can be written as sar, hsar, or tsar).

2. Dataset Bias & Author Characterization

• Single-Author Style: The current batch of 2,121 approved rows was curated predominantly based on the typing habits and intuitive phonetic system of a single native author.
• Model Training Warning: Because it represents a singular, personalized writing style, this dataset does not capture the full, erratic spectrum of spelling variations found across the broader internet. It should be treated as a highly consistent baseline rather than an exhaustive representation of all internet slang.

3. Future Roadmap

This is a living corpus. The engineering roadmap includes expanding the volume of data rows and systematically injecting multi-author variations to reflect different online typing habits, subcultures, and colloquial spellings across the web.

---

📊 Dataset Structure & Features

The data is delivered in clean .csv format containing two simple string features:

• myanmar: The source text written in standard Myanmar Unicode script.
• myanglish: The target phonetic representation written using the Latin script.

Note on Content: This corpus focuses on pure phonetic transliteration (sound mapping). It does not contain direct English semantic translations or truncated text shortcuts (e.g., it maps the full pronunciation instead of text-speak abbreviations).

---

💡 Practical Applications

• Transliteration Models: Training sequence-to-sequence (Seq2Seq) networks to automatically convert informal Myanglish messages back into standardized Burmese script for indexing and search optimization.
• Social Media Text Normalization: Serving as a preprocessing dictionary for downstream sentiment analysis or toxic text filters that need to decipher romanized slang.
• Chatbot Intention Mapping: Helping localized conversational AI understand user inputs when they type Burmese phrases on English keyboards.

---

⚖️ License & Open-Source Terms

This dataset is released under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

You are free to copy, redistribute, adapt, and build upon this material for any purpose (including commercial applications), provided you acknowledge the original author/publisher and license your derivative works under identical terms.

---

🧑‍💻 About the Author & Curator

Khant Sint Heinn (Burmese: ခန့်ဆင့်ဟိဏ်း), working under the professional moniker Kalix Louis, is a Machine Learning Engineer specialized in Natural Language Processing (NLP), data foundations, and open-source AI infrastructure. His primary engineering focus centers on elevating low-resource languages—specifically Burmese (Myanmar)—into data-rich assets capable of powering next-generation language models and scalable linguistic tools.

Currently serving as the Lead Developer at DatarrX, Khant Sint Heinn architects robust data pipelines, manages large-scale dataset curations, and builds the open-source building blocks necessary for accessible machine learning applications.

Connect with the Author:

• GitHub
• Hugging Face
• Kaggle

---

🏛️ Hosted by DatarrX

This dataset is proud to be maintained and distributed by DatarrX (Burmese: ဒေတာ-အက်စ်), a non-profit open-source foundation dedicated to building a robust digital and data foundation for the Burmese language in the AI era.

Rooting Burmese into AI.

---

📜 Citation & Academic Reference

If you incorporate the myX-myanmar-to-myanglish-corpus dataset into your machine learning workflows, transliteration experiments, or linguistic research pipelines, please acknowledge the author and publisher via the following official BibTeX citation:

@misc{datarrx_myanglish_corpus_2026,
  author       = {Khant Sint Heinn},
  title        = {myX-myanmar-to-myanglish-corpus: A Parallel Text Dataset Mapping Native Burmese Unicode to Romanized Phonetic Script Strings},
  year         = {2026},
  publisher    = {Hugging Face},
  organization = {DatarrX},
  howpublished = {https://huggingface.co/datasets/DatarrX/myX-myanmar-to-myanglish-corpus},
  note         = {Published under DatarrX. Open-source community asset released under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)}
}