Datasets:

freococo
/

1_pattern_10Kplus_myanmar_sentences

license: mit
task_categories:
  - text-generation
  - text-classification
language:
  - my
tags:
  - MyanmarSentences
  - Burmese
  - BurmeseSentences

🧠 1_pattern_10Kplus_myanmar_sentences

A structured dataset of 11,452 Myanmar sentences generated from a single, powerful grammar pattern:

📌 Pattern:

Verb လည်း Verb တယ်။

A natural way to express repetition, emphasis, or causal connection in Myanmar.

💡 About the Dataset

This dataset demonstrates how applying just one syntactic pattern to a curated verb list — combined with syllable-aware rules — can produce a high-quality corpus of over 10,000 valid Myanmar sentences.

Each sentence is:

Grammatically valid
Syllable-tokenized
Pattern-consistent
Cleaned and filtered

🔁 Pattern in Use

Examples:

ချစ်လည်း ချစ်တယ်။
ကစားလည်း ကစားတယ်။
ကံကြီးလည်း ထိုက်တယ်။
ခေါင်းချင်းဆိုင်လည်း တိုက်တယ်။

📏 Rules in Use

Syllable Count	Rule Name	Sentence Format	# of Sentences Generated
1	Rule_1_Syllable	`Aလည်း Aတယ်။`	1
2	Rule_2_Syllable	`Aလည်း Bတယ်။` and `Aလည်း ABတယ်။`	2 (dual sentences)
3	Rule_3_Syllable	`ABလည်း Cတယ်။`	1
4	Rule_4_Syllable	`ABCလည်း Dတယ်။`	1
5+	Rule_{N}_Syllable	`ABCD...လည်း Zတယ်။`	1 per item

📁 Dataset Format

Each row in the CSV contains:

Column	Description
`my_sentence`	The full generated Myanmar sentence
`my_word`	The original verb the sentence is based on
`my_subword`	List of syllable-level tokens (as a string list)
`subword_number`	Number of syllables in `my_word`

Example:

my_sentence: ကလည်း ကတယ်။
my_word: က
my_subword: ["က"]
subword_number: 1

🤯 Why It’s Special

    •	✅ Only one pattern → yet over 11,000 real Myanmar sentences
    •	✅ Rule logic scales across syllable complexity
    •	✅ Cleaned, structured, and easy to extend
    •	✅ Represents real grammar, not artificial templates

😅 Problems We Faced

We started with a large list of Myanmar verbs and applied a syllable-level tokenizer to break each verb into structured chunks. Based on syllable count, we applied one of six rules to generate sentences.

Challenges included:

    •	Unicode inconsistencies (e.g., ဥ် vs ဉ်, န့် vs န့်)
    •	Visually similar characters causing mis-splitting
    •	Manual review needed for edge cases
    •	Some grammatically valid outputs lacked semantic sense

🔮 Future Plans

This is just Pattern 1.

Coming soon: • မ V နဲ့။

We aim to build a full-scale, pattern-rich Myanmar corpus — one rule at a time.

⸻

🎯 Use Cases

•	Fine-tune sentence generation models
•	Train grammar correction systems
•	Build linguistic datasets for Myanmar NLP
•	Teach Myanmar grammar through concrete patterns
•	Benchmark syllable tokenizers

⸻

🧪 Quality & Manual Review

Even though all sentences were generated using grammatical rules, not all combinations may sound natural or meaningful in everyday Myanmar.

📝 This dataset should be manually reviewed by native speakers to ensure each sentence:

    •	Sounds natural
    •	Makes semantic sense
    •	Feels appropriate for real-world use

That said — even if you:

    •	Remove awkward or illogical samples
    •	Filter or adjust by context
    •	Expand with more rules and patterns

➡️ You’ll still retain 10K+ high-quality, structured Myanmar sentences.

⚠️ Please don’t use this dataset blindly for production training without native review.

⸻

📜 License

MIT — free to use, adapt, remix, or improve. But give credit where it’s due 💛

⸻

🔗 Citation

@dataset{myanmar_verb_pattern_2025,
  title={1 Pattern 10K+ Myanmar Sentences},
  author={freococo},
  year={2025},
  url={https://huggingface.co/datasets/freococo/1_pattern_10Kplus_myanmar_sentences}
}