myTokenize

myTokenize is a Python library that tokenizes Myanmar text into syllables, words, phrases, and sentences. It supports multiple tokenization techniques using rule-based, statistical, and neural network-based approaches.

Features

• Syllable Tokenization: Break text into syllables using regex rules.
• BPE and Unigram Tokenization: Leverage SentencePiece models for tokenization.
• Word Tokenization: Segment text into words using:
  • myWord: Dictionary-based tokenization.
  • CRF: Conditional Random Fields-based tokenization.
  • BiLSTM: Neural network-based tokenization.
• Phrase Tokenization: Identify phrases in text using normalized pointwise mutual information (NPMI).
• Sentence Tokenization: Use a BiLSTM model to segment text into sentences.

Installation

1. Clone the repository:

   git clone https://github.com/ThuraAung1601/myTokenize.git
   cd myTokenize

2. Install dependencies:

   pip install -r requirements.txt

3. Install the library:

   pip install .

Usage

Syllable Tokenizer

from myTokenize import SyllableTokenizer

tokenizer = SyllableTokenizer()
syllables = tokenizer.tokenize("မြန်မာနိုင်ငံ။")
print(syllables)  # ['မြန်', 'မာ', 'နိုင်', 'ငံ', '။']

BPE Tokenizer

from myTokenize import BPETokenizer

tokenizer = BPETokenizer()
tokens = tokenizer.tokenize("ရွေးကောက်ပွဲမှာနိုင်ထားတဲ့ဒေါ်နယ်ထရမ့်")
print(tokens)  # ['▁ရွေးကောက်ပွဲ', 'မှာ', 'နိုင်', 'ထား', 'တဲ့', 'ဒေါ်', 'နယ်', 'ထ', 'ရ', 'မ့်']

Word Tokenizer

from myTokenize import WordTokenizer

tokenizer = WordTokenizer(engine="CRF")  # Use "myWord", "CRF", or "LSTM"
words = tokenizer.tokenize("မြန်မာနိုင်ငံ။")
print(words)  # ['မြန်မာ', 'နိုင်ငံ', '။']

Phrase Tokenizer

from myTokenize import PhraseTokenizer

tokenizer = PhraseTokenizer()
phrases = tokenizer.tokenize("ညာဘက်ကိုယူပြီးတော့တည့်တည့်သွားပါ")
print(phrases)  # ['ညာဘက်_ကို', 'ယူ', 'ပြီး_တော့', 'တည့်တည့်', 'သွား_ပါ']

Sentence Tokenizer

from myTokenize import SentenceTokenizer

tokenizer = SentenceTokenizer()
sentences = tokenizer.tokenize("ညာဘက်ကိုယူပြီးတော့တည့်တည့်သွားပါခင်ဗျားငါးမိနစ်လောက်ကြာလိမ့်မယ်")
print(sentences)  # [['ညာ', 'ဘက်', 'ကို', 'ယူ', 'ပြီး', 'တော့', 'တည့်တည့်', 'သွား', 'ပါ'], ['ခင်ဗျား', 'ငါး', 'မိနစ်', 'လောက်', 'ကြာ', 'လိမ့်', 'မယ်']]

Folder Structure

./myTokenize/
├── CRFTokenizer
│   └── wordseg_c2_crf.crfsuite
├── SentencePiece
│   ├── bpe_sentencepiece_model.model
│   ├── bpe_sentencepiece_model.vocab
│   ├── unigram_sentencepiece_model.model
│   └── unigram_sentencepiece_model.vocab
├── Tokenizer.py
└── myWord
    ├── phrase_segment.py
    └── word_segment.py

Dependencies

• Python 3.8+
• TensorFlow
• SentencePiece
• pycrfsuite
• Numpy

License

This project is licensed under the MIT License. See the LICENSE file for details.

Authors

• Ye Kyaw Thu
• Thura Aung

References

• myWord: Syllable, Word and Phrase Segmenter for Burmese, Ye Kyaw Thu, Sept 2021, GitHub Link: https://github.com/ye-kyaw-thu/myWord
• sylbreak: Syllable segmentation tool for Myanmar language (Burmese), Ye Kyaw Thu, GitHub Link: https://github.com/ye-kyaw-thu/sylbreak
• mySentence: Corpus and Models for Burmese (Myanmar language) Sentence Tokenization, GitHub Link: https://github.com/ye-kyaw-thu/mySentence