chuuhtetnaing/myanmar-pos-dataset

Please visit to the GitHub repository for other Myanmar Language datasets.

Myanmar POS Tagging Dataset

A token classification dataset for Myanmar (Burmese) Part-of-Speech (POS) tagging, formatted for sequence labeling tasks using BIO tagging scheme.

📓 Data Preparation Notebook: dataset-preparation.ipynb

📓 Fine-Tuning Notebook: myanmar-pos-fine-tuning.ipynb (based on the HuggingFace Token Classification Guide)

🚀 Try it out: Myanmar POS Demo

Dataset Description

This dataset contains Myanmar language text annotated with Part-of-Speech (POS) tags. The dataset has been cleaned, preprocessed, and formatted to be ready for fine-tuning HuggingFace token classification models.

The dataset uses 15 Myanmar POS tags designed to support various NLP tasks including information extraction, semantic processing, and machine translation.

Source Data

Original Dataset: LULab/myPOS

Original GitHub Repository: ye-kyaw-thu/myPOS

This is a preprocessed and cleaned version of the original myPOS dataset. Please cite the original dataset and authors if you use this preprocessed version in your research.

Processing Pipeline

1. Data Cleaning: Fixed annotation issues and typos including tag corrections, missing spaces between annotations, and cases with multiple tags for a single word
2. Token-Label Alignment: Verification of proper alignment between tokens and their POS tags
3. Deduplication: Removal of duplicate entries
4. Train-Test Split: 80% training (32,777 examples) and 20% test (8,195 examples)
5. Format Conversion: Conversion to HuggingFace token classification format

Dataset Statistics

Split	Examples
Train	32,777
Test	8,195

Data Format

{
    "tokens": [ကျွန်, တော်, မျက်, မှန်, တစ်, လက်, လုပ်, ချင်, ပါ, တယ်, ။],
    "pos_tags": [10, 25, 6, 21, 13, 8, 14, 8, 8, 9, 11]
}

Features

• tokens: Sequence[string] - Input tokens (Myanmar words/syllables)
• pos_tags: Sequence[ClassLabel] - Part-of-Speech labels

POS Tags

The dataset includes 15 Myanmar POS tags:

POS Tag	Description	Examples
abb	Abbreviation	အထက (Basic Education High School), လ.ဝ (Confidentiality)
adj	Adjective	ရဲရင့် (brave), လှပ (beautiful), မွန်မြတ် (noble)
adv	Adverb	ဖြေးဖြေး (slow), နည်းနည်း (less)
conj	Conjunction	နှင့် (and), ထို့ကြောင့် (therefore), သို့မဟုတ် (or)
fw	Foreign Word	1, 2, 3, Myanmar, BBC, Google, ミャンマー, 缅甸
int	Interjection	အမလေး (Oh My God!)
n	Noun	ကျောင်း (school), စာအုပ် (book), ဒေါ်အောင်ဆန်းစုကြည်, လွတ်လပ်ရေး (freedom)
num	Number	၁ (1), ၂ (2), ၃ (3), ၁၀ (10), ၁၀၀ (100)
part	Particle	များ, ခဲ့, သင့်, လိမ့်, နိုင်
ppm	Post-positional Marker	သည်, က, ကို, အား, သို့, မှာ, တွင်
pron	Pronoun	ကျွန်တော် (I), ကျွန်မ (I), သင် (you), သူ (he), သူမ (she)
punc	Punctuation	။, ၊, (, ), , _, ', "
sb	Symbol	?, #, &, %, $, £, ¥, 𝜆, π, ÷, +, ×, @
tn	Text Number	တစ် (one), နှစ် (two), သုံး (three), တစ်ရာ, တစ်ထောင်
v	Verb	ကူညီ (help), လိုက်နာ (observe), အားပေး (encourage)

Usage

from datasets import load_dataset
ds = load_dataset("chuuhtetnaing/myanmar-pos-dataset")

ds["train"].features["pos_tags"].feature.names
# ['B-abb', 'B-adj', 'B-adv', 'B-conj', 'B-fw', ..., 'I-ppm', 'I-pron', 'I-punc', 'I-tn', 'I-v']

Example from the dataset:

ds["train"][7]
# {
#     'tokens': [ဒီ, ကို, ၁, ၅, မိ, နစ်, အ, တွင်း, လာ, နိုင်, မ, လား, ။],
#     'pos_tags': ['B-pron', 'B-ppm', 'B-num', 'I-num', 'B-n', 'I-n', 'B-n', 'I-n', 'B-part', 'B-part', 'B-part', 'I-part', 'B-punc']
# }

id2label = {i: label for i, label in enumerate(ds["train"].features["ner_tags"].feature.names)}
label_ner_tags = [id2label[tag_id] for tag_id in ds["train"][7]["ner_tags"]]

Intended Use

• Training POS tagging models for Myanmar NLP
• Token classification / sequence labeling experiments (HuggingFace Token Classification Training Example)
• Myanmar language processing research
• Information extraction and semantic processing
• Machine translation preprocessing

Citation

If you use this preprocessed dataset in your research, please cite the original dataset:

@dataset{mypos,
  title={myPOS: Myanmar Part-of-Speech Tagging Dataset},
  author={Ye Kyaw Thu and others},
  howpublished={HuggingFace Datasets},
  url={https://huggingface.co/datasets/LULab/myPOS},
  year={2023}
}

If you want to use myPOS models or POS-tagged data in your research and we'd appreciate if you use the following two references:

Paper 1: Khin War War Htike, Ye Kyaw Thu, Zuping Zhang, Win Pa Pa, Yoshinori Sagisaka and Naoto Iwahashi, "Comparison of Six POS Tagging Methods on 10K Sentences Myanmar Language (Burmese) POS Tagged Corpus", at 18th International Conference on Computational Linguistics and Intelligent Text Processing (CICLing 2017), April 17~23, 2017, Budapest, Hungary. [Draft paper], [Poster]

Paper 2: Zar Zar Hlaing, Ye Kyaw Thu, Myat Myo Nwe Wai, Thepchai Supnithi, Ponrudee Netisopakul, "Myanmar POS resource extension effects on automatic tagging methods", In Proceedings of the 15th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP 2020), Nov 18 to Nov 20, 2020, Bangkok, Thailand, pp. 189-194. [Paper]

For more information about the original dataset, please visit:

• GitHub Repository
• HuggingFace Dataset

Acknowledgments

Special thanks to:

• Professor Ye Kyaw Thu and the LULab team for creating the original myPOS dataset
• The Myanmar NLP community for supporting language technology development