Building Myanmar's AI Future
MyMind AI Company Limited is committed to developing impactful AI technologies for Myanmar. It is developing Burmese language models, public data, reliable AI performance evaluation standards, and practical AI solutions for the community and business sectors.
Understanding the Technologies Shaping Our Future
Explore the latest developments in artificial intelligence and understand how emerging technologies can create new opportunities for Myanmar.

Generative AI
What Is Generative AI?
Generative AI (Gen AI) is a type of artificial intelligence that learns patterns from data to create content such as text, images, audio, video, and computer code. Guided by instructions known as prompts, these systems generate responses based on learned patterns and the context provided by users.
Its applications range from drafting documents and summarizing reports to translating languages, developing visual concepts, and assisting with software development. For example, a Burmese-language assistant can help prepare an announcement, explain a complex topic, or produce a translation draft for review.
Generative AI can support creativity and productivity, but its outputs may contain errors, biases, or misleading information. Effective use combines clear instructions, appropriate data protection, and human review to ensure that results are accurate, relevant, and suitable for their intended purpose.
Common Applications of Generative AI
Generative AI supports a wide range of creative and practical tasks, helping individuals and organizations turn ideas and information into useful content.
Text and Content Creation
Generative AI can draft articles, announcements, emails, and social media posts. It also helps refine writing style and adapt content for different audiences.
Translation and Summarization
AI tools can produce translation drafts and condense lengthy documents into concise summaries. Human review helps preserve meaning, terminology, and cultural context.
Image and Design Generation
Text prompts can guide the creation of illustrations, marketing visuals, and design concepts. These tools help designers explore ideas and develop initial drafts.
Audio and Video Generation
Generative AI can create narration, synthetic speech, music, and video content. Applications include educational materials, presentations, and public information campaigns.
Software Development
Coding assistants can suggest code, explain functions, generate documentation, and help identify potential errors. Developers remain responsible for testing, security, and the quality of the final software.
Conversational Assistance
AI assistants can answer questions, explain topics, and help users navigate information. When connected to approved knowledge sources, they can also support customer enquiries and internal knowledge access.
Across these applications, effective use requires appropriate data protection and human review to ensure that outputs meet their intended purpose.
Source A Survey of Generative AI Applications (2023) https://arxiv.org/abs/2306.02781?
Read More â
Large Language Models
Follow advances in multilingual and low-resource language models, including technologies designed to understand and generate high-quality Burmese content.
Read More â
Responsible AI
Learn about AI safety, transparency, privacy, fairness, governance, and the responsible deployment of intelligent systems.
Read More âBuilding the Foundation for Myanmar's AI Ecosystem
Burmese LLM Evaluation Dataset
Most AI language models are trained mainly on English and a few other big languages. Very few tests exist to check how well they work in Burmese.
We built a Burmese evaluation dataset, had it checked by human reviewers, and used it to test four small, open AI models on everyday language tasks.
The dataset follows the format and task list of SEA-HELM, the Southeast Asian language benchmark from AI Singapore, so our results can be compared with other Southeast Asian languages.
How the Evaluation Data Was Created
We built the data in four steps:
Step 1 â Start from SEA-HELM (source data). We downloaded the SEA-HELM datasets. Some tasks already have Burmese items, for example sentiment, toxicity and instruction following. We kept all of them.
Step 2 â Add more Burmese data (extended data). To make the tests bigger and more varied, we added public Burmese data: FLORES+ translation sentences, a Burmese sentiment dataset and two Burmese hate-speech datasets from Hugging Face.
Step 3 â Translate what is missing. Some SEA-HELM tasks have no Burmese items. We translated them from Thai, Indonesian, Tamil, Vietnamese, Chinese, Tagalog or English using Gemini. Items that only make sense in another country's culture were not translated.
Step 4 â Check by hand. Human reviewers checked the items, gave each one a score and corrected mistakes (see âHuman Reviewâ below).
Examples from the Dataset
Here are some real items from the dataset. The English line is a translation to help readers.
Task: Sentiment analysis
example:áááŻá¸áĄáąá¸áĄáąá¸áá˛áˇ áĄáááşáá˛áážáŹ ááąáŹáşááŽáá°áá°ááąá¸ááąáŹááşáááşá¸ á áŹáááşááᏠáĄáááşá¸ááááşáážáááŹáá˛
âReading at home with a hot coffee on a cool rainy day is so relaxing.â
Correct Answer: Positive
Task: Machine translation
Example: English: 28-year-old Vidal had joined Barça three seasons ago, from Sevilla.
Burmese reference: áĄáááş áááážá áşáĄáá˝ááş ááŽáááşááşáááş áá˝ááşáá˛áˇááąáŹ ááŻáśá¸ááŹááŽáá˝ááş ááŽááŽá¸ááŹá¸ááž ááŹááŹáááŻáˇ ááźáąáŹááşá¸ááŹáá˛áˇáááşá
Model's translation is compared with the reference
Task: Causal reasoning
Example: áááŻáá ášá ááşá¸ááᯠááąááąáŹááşá¸ááááşá áá áşááźááˇáş ááŻááşáááŻá¸ááŹá¸áááşá (áĄááźáąáŹááşá¸áĄáááşá¸)
âThe item was packed in bubble wrap.â What is the reason? (A) The item is fragile. (B) The item is small.
Correct Answer: A
Human Review (Manual Verification)
Machine translation and automatic tools can make mistakes, so human reviewers from our QC team checked the data in Excel review sheets. They read each item and gave it a score from 1 to 5 for each of these points:
- Completeness: nothing is missing from the text, question or answer.
- Fluency: the Burmese reads naturally and smoothly.
- Sensibility: the item makes sense and is logical.
Text clearance: correct spelling and font, and no Zawgyi (old Burmese encoding) problems.
Some tasks had extra points. Translation was also scored for grammar and faithfulness to the original meaning. Summaries were scored for relevance and coherence, and instructions for clarity. Reviewers also marked whether the given answer or label was correct (yes or no). When an item had a problem, they wrote a corrected version and a comment.
On the 1 to 5 scale, 5 means fully correct and natural, 4 means mostly good, 3 means partly good, and 1 or 2 means serious problems.
How We Tested the Models
We tested four small open models that can run on a single GPU:
Gemma-4-E2B-it (Google)
Gemma-SEA-LION-v4.5-E2B-IT (AI Singapore, built on Gemma)
Qwen3.5-4B (Alibaba)
Llama-3.2-3B-Instruct (Meta)
Each model was run with three different generation settings (fixed, shared, and each maker's recommended setting). This shows whether a score comes from the model itself or only from the setting. The scoring methods were adjusted for Burmese, which has no spaces between words, so we count characters or syllables instead of words. For multi-turn chat, an AI judge (Gemini) rated each answer from 1 to 10.
Results
The following list shows each model's average score over the three settings. Higher is better.
Sentiment Analysis (Accuracy) Gemma - 76%, SEA-LION-76%, QWEN- 25%, LLAMA-1%
Machine Translation (chrF) Gemma - 49.6%, SEA-LION-49.8%, QWEN- 29.6%, LLAMA- "-"
Question Answering (Char-F1) Gemma - 45.7%, SEA-LION-44.1%, QWEN- 32.9%, LLAMA- "-"
Metaphor understanding (Accuracy) Gemma - 74%, SEA-LION-73%, QWEN- 51%, LLAMA-47%
Instruction following (Pass rate) Gemma - 70%, SEA-LION-67%, QWEN- 44%, LLAMA-25%
Multi-turn chat (Judge score /10) Gemma - 5.5%, SEA-LION-5.5%, QWEN- 2.1%, LLAMA-1.1%
Summarization (ROUGE-L) Gemma - 17.6%, SEA-LION-17.6%, QWEN- 13.3%, LLAMA-"-"
âââ means the model was not tested on that task. Results for reading, toxicity, inference and causal reasoning, and for the four new datasets, will be added later.
Key findings
- Gemma and SEA-LION are clearly the best for Burmese and score almost the same on every task. Qwen is weaker, and Llama is the weakest.
- The ranking is real. It stayed the same under all three generation settings. The settings only made a small difference in open-ended chat.
- Even the best models have gaps. They rarely detect neutral sentiment, they write summaries that are too long, they struggle with exact rules such as âend with this phraseâ, and they reach only about 5.5 out of 10 in conversation.
- Smaller is faster, not better. Llama and Qwen use less memory (2.5 GB and 3.4 GB) and run faster, but Gemma and SEA-LION (about 6.9 GB) give much better Burmese answers.
Next Steps
- Apply the reviewers' corrections, starting with the question-answering and reading datasets.
- Review the Burmese grammar dataset with a Burmese linguist.
- Test the models on the four new Burmese datasets.
- Add the results for the remaining four tasks.
References
These are the published research papers behind the benchmark, data and scoring methods used in this work.
Benchmarks and datasets
1.Susanto, Y. et al. (2025). SEA-HELM: Southeast Asian Holistic Evaluation of Language Models. arXiv. https://arxiv.org/abs/2502.14301
The benchmark our dataset follows.
2.Bandarkar, L. et al. (2024). The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants. ACL 2024. https://arxiv.org/abs/2308.16884
Source of the multiple-choice reading task.
3.NLLB Team (2022). No Language Left Behind: Scaling Human-Centered Machine Translation. arXiv. https://arxiv.org/abs/2207.04672
Introduced the FLORES-200 translation benchmark, which FLORES+ is based on.
Evaluation methods
4.PopoviÄ, M. (2015). chrF: character n-gram F-score for automatic MT evaluation. WMT 2015. https://aclanthology.org/W15-3049
Translation score (chrF).
5.Rei, R. et al. (2020). COMET: A Neural Framework for MT Evaluation. EMNLP 2020. https://aclanthology.org/2020.emnlp-main.213
Translation score (COMET).
6.Juraska, J. et al. (2024). MetricX-24: The Google Submission to the WMT 2024 Metrics Shared Task. WMT 2024. https://aclanthology.org/2024.wmt-1.35
Translation score (MetricX-24).
7.Lin, C.-Y. (2004). ROUGE: A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out. https://aclanthology.org/W04-1013
Summary score (ROUGE-L).
8.Rajpurkar, P. et al. (2016). SQuAD: 100,000+ Questions for Machine Comprehension of Text. EMNLP 2016. https://aclanthology.org/D16-1264
Question-answering scores (Exact Match and F1).
9.Zhou, J. et al. (2023). Instruction-Following Evaluation for Large Language Models (IFEval). arXiv. https://arxiv.org/abs/2311.07911
Instruction-following test.
10.Zheng, L. et al. (2023). Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. NeurIPS 2023 Datasets and Benchmarks. https://arxiv.org/abs/2306.05685
Multi-turn chat test and AI-judge scoring.
Read More âááźááşááŹ-áĄááşášáááááş AI ááŹááŹááźááşá áá áşáá áşáᯠáááşááąáŹááşááźááşá¸ - Gemma-4 ááᯠFine-tuning ááźáŻááŻááşááźááşá¸
Large Language Model (LLM) áá˝áąáᏠááŹááŹá ááŹá¸ááŹá¸áááşááźááşá¸á á áŹááŹá¸ááŻááşááŻááşááźááşá¸áá˛áˇ ááŹááŹááźááşáááŻááźááşá¸áááŻáˇáážáŹ ááááááŹáᏠáááŻá¸áááşááŹááŤáááşá
ááŤááąáá˛áˇ ááŹááŹá ááŹá¸áĄáŹá¸ááŻáśá¸áĄáá˝ááş á á˝ááşá¸ááąáŹááşáááşáá°ááŽááąááźááşá¸ááąáŹáˇ áááŻááşááŤáá°á¸á áĄáá°á¸áááźááˇáş ááźááşááŹááŹááŹá ááŹá¸ááᯠLanguage Resources áááşá¸ááŤá¸áá˛áˇ ááŹááŹá ááŹá¸áá˝áąáĄáá˝ááş áĄáááşáĄáá˝áąá¸ááąáŹááşá¸áá˝ááşáá˛áˇ AI Translation System áá˝áą áááşááąáŹááşáááŻáˇ áĄáá˝ááˇáşáĄáááşá¸áá˛áˇ á áááşááąáŤáşáážáŻáá˝áą áážáááąáá˛ááźá áşááŤáááşá áĄááźáąáĄááąááᯠáĄááźáąááśááźáŽá¸ Google áá˛áˇ Gemma-4 ááᯠáĄááŻáśá¸ááźáŻááźáŽá¸ ááźááşááŹ-áĄááşášáááááş áážá áşááŹááŹáĄááźááşáĄáážááşááŹááŹááźááşáááŻááşáá˛áˇ Model áá áşááŻááᯠFine-tuning ááźáŻááŻááşáá˛áˇááŤáááşá ááŽá ááşá¸áááşáážáŻáážáŹ Hugging Face áá˛áˇ FineTranslations Datasetá Unsloth Framework áá˛áˇ QLoRA áááşá¸áááŹáááŻáˇááᯠáĄááŻáśá¸ááźáŻáá˛áˇááźáŽá¸ Model áá˛áˇ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸ááᯠCOMET áá˛áˇ chrF++ Metrics áá˝áąááźááˇáş áĄáá˛ááźááşáá˛áˇááŤáááşá ááźáŻááŻááşáá˛áˇááąáŹ research ááž Model áá˝áąá¸ááťááşáážáŻáááą Training áá˛áˇ Evaluation ááááşáá˝áąáĄáá ááŻááşááąáŹááşáá˛áˇáá˛áˇ ááŻááşáááşá¸á ááşááᯠááťážááąáá˝áŹá¸áážáŹ ááźá áşááŤáááşá
áá Gemma-4 ááᯠááŹááźáąáŹááˇáş áá˝áąá¸ááťááşáá˛áˇááá˛á
áᎠresearch áá˛áˇ áĄááááĄááźáąááśá Gemma-4 ááźá áşááŤáááşá Gemma-4 áᏠGoole DeepMind ááž ááŻááşááąááŹá¸áá˛áˇ Open-weight Model Family áá áşááŻááźá áşááźáŽá¸ Language Understanding, Reasoning áá˛áˇ Multimodal Processing áááŻáˇáĄáá˝ááş áááşáá˝ááşáááşááąáŹááşááŹá¸ááŤáááşá Open-weight Model ááźá áşáá˛áˇáĄáá˝ááş Model Weights áá˝áąááᯠááá°ááźáŽá¸ áááŻááşáááŻááş Hardware ááąáŤáşáážáŹ áĄááŻáśá¸ááźáŻááźááşá¸á Fine-tuning ááźáŻááŻááşááźááşá¸áááŻáˇ ááŻááşááąáŹááşáááŻááşááŤáááşá
Architecture áá˛áˇ Model Variants
Gemma-4 áážáŹ Model Size áá˛áˇ Architecture ááá°ááŽáá˛áˇ Variants áá˝áą ááŤáááşááŤáááşá
E2B áážááˇáş E4B: Memory áá˛áˇ Compute Efficiency ááᯠáĄááąá¸ááŹá¸áá˛áˇ Compact Model ááťáŹá¸á
12B: áááŻáááŻááźáŽá¸ááŹá¸áá˛áˇ Dense Architecture ááᯠáĄááźáąááśááŹá¸áá˛áˇ Modelá
26B A4B: Mixture-of-Experts (MoE) Architecture ááᯠáĄááŻáśá¸ááźáŻááŹá¸áá˛áˇ Modelá
31B: Dense Architecture ááᯠáĄááŻáśá¸ááźáŻááŹá¸áá˛áˇ Modelá
áᎠresearch áážáŹááąáŹáˇ unsloth/gemma-4-E2B-it ááᯠáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
E2B áá˛áˇ á áááşáááşá áŹá¸á ááŹááąáŹááşá¸áá˛áˇ áĄááťááşáá áşááŻá Per-Layer Embeddings (PLE) áá˛áˇ Hybrid Attention Architecture áááŻáˇ ááŤáááşááźááşá¸ááźá áşááŤáááşá
Hybrid Attention áááŻáᏠá áŹááŹá¸áĄááŽá¸áĄááŹá¸á áĄááťááşáĄáááşáá˝áąááᯠáĄáŹááŻáśá ááŻááşáá˛áˇ Local Attention áá˛áˇ áááŻáááŻááťááşááźááˇáşáá˛áˇ Context ááᯠááźááˇáşáááŻááşáá˛áˇ Global Attention áááŻáˇááᯠááąáŤááşá¸á ááşáĄááŻáśá¸ááźáŻááźááşá¸ ááźá áşááŤáááşá
ááŤáˇáĄááźááş Gemma-4 áᏠText áááŻááŹáá Image áá˛áˇ Audio Input áá˝áąáááŻá፠áááŻááşáá˝ááşáááŻááşáá˛áˇ Multimodal Model Family áá áşááŻááźá áşááŤáááşá
áááşááŹááŹá ááŹá¸áá˝áąááᯠSupport ááŻááşááá˛á
Gemma-4 ááᯠááŹááŹá ááŹá¸ ááá ááťáąáŹáşáá˛áˇ ááąáˇááťááˇáşááŹá¸ááŤáááşá ááŤááźáąáŹááˇáş English ááᯠáĄááŻáśá¸ááťáŹá¸áá˛áˇ ááŹááŹá ááŹá¸áá˝áąáĄááźááş áĄááźáŹá¸ááŹááŹá ááŹá¸ááťáŹá¸áĄáá˝ááşá፠áĄááźáąááś Language Understanding á á˝ááşá¸áááşáá˝áą ááážáááŹá¸ááŤáááşá
ááŤááąáá˛áˇ ááŹááŹá ááŹá¸áĄáŹá¸ááŻáśá¸áážáŹ á á˝ááşá¸ááąáŹááşáááşáá°ááŽáááşáááŻáˇ ááááŻáááŻááşááŤáá°á¸á Training Data áááŹááá˛áˇ áĄáááşáĄáá˝áąá¸ááąáŤáşáá°áááşááźáŽá¸ ááááşáá˝áą áá˝áŹááźáŹá¸áááŻááşááŤáááşá
ááŤááźáąáŹááˇáş ááźááşááŹááŹááŹá ááŹá¸áĄáá˝ááş ááŽá¸ááźáŹá¸ Fine-tuning ááźáŻááŻááşááźáŽá¸ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸ áááşááąáŹááşáááŻá¸áááşááŹáááŻááşááá˛áááŻááŹááᯠá ááşá¸áááşáááŻáˇ áááşáá˝ááşáá˛áˇááŤáááşá
Model: Gemma-4 E2B - Unsloth
áá Training Data - FineTranslations
Machine Translation Model áá áşááŻáá˛áˇ áĄáááşáĄáá˝áąá¸áᏠTraining Data áá˛áˇ áĄáá˝ááşááŽá¸á ááşáá˛áˇ áááşáá˝ááşáážáŻáážáááŤáááşá
áᎠProject áážáŹ FineTranslations Dataset ááᯠáĄááźáąááśáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
FineTranslations áᏠHugging Face á ááŻááşááąááŹá¸áá˛áˇ Multilingual Translation Dataset áá áşááŻááźá áşááźáŽá¸ ááŹááŹá ááŹá¸ á áá ááťáąáŹáşáá˛áˇ Parallel Text Data áá˝áą ááŤáááşááŤáááşá
áᎠDataset ááᯠFineWeb2 ááž ááážááá˛áˇ áá°áááşá¸ááŹááŹá ááŹá¸á áŹááŹá¸áá˝áąááᯠGemma 3 27B áĄááŻáśá¸ááźáŻááźáŽá¸ English ááŹááŹáááŻáˇ ááŹááŹááźááşáᏠáááşááŽá¸ááŹá¸ááźááşá¸ ááźá áşááŤáááşá
áᎠProject áážáŹ Dataset áá˛á ááźááşááŹááŹááŹá ááŹá¸áá˛áˇ áááşáááŻááşáá˛áˇ Data áá˝áąááᯠáá˝áąá¸ááťááşáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
áá°ááááşáá˝ááşááťááşáĄá Document ááąáŤááşá¸ á.á á áááşá¸áááˇáşááᯠáĄááŻáśá¸ááźáŻáááŻáˇ ááźááşáááşáá˛áˇááźáŽá¸ áááşáážá Training Log áĄá Training Examples 1,162,346 áᯠááŤáááşááŤáááşá
Data áá˝áąááᯠáĄáąáŹááşááŤáĄááťááşáĄáááşáá˝áąáĄááąáŤáş áĄááźáąááśááźáŽá¸ ááźááşáááşáá˛áˇááŤáááşá
Source Text - áá°áááşá¸á áŹááŹá¸á
Target Text - ááŹááŹááźááşááŹá¸áá˛áˇ á áŹááŹá¸á
Source Language - áá°áááşá¸ááŹááŹá ááŹá¸á
Target Language - áááşáá˝ááşááŹá¸áá˛áˇ ááŹááŹá ááŹá¸á
áᎠResearch áá˛áˇ áááşáá˝ááşááťááşá ááŹááŹááźááşááźááşá¸ áážá áşááťááŻá¸á ááŻáśá¸ááᯠáááşááźáŹá¸ááąá¸áááŻáˇ ááźá áşááŤáááşá
Burmese â English
English â Burmese
Dataset áá˛áˇ áá°áááŻáśá áśáážáŹ English áááŻáˇ ááŹááŹááźááşááŹá¸áá˛áˇ Data áá˝áąááᯠáĄáááááŹá¸áá˛áˇáĄáá˝ááş English â Burmese Data áá˝áąááᯠReverse Pair áĄááźá áş ááźáąáŹááşá¸áá˛áááˇáşáá˝ááşá¸ááŹá¸ááŤá Dataset Preparation áĄáááŻááşá¸áážáŹ áĄá˛ááŽááŻááşáááşá¸á ááşáááŻáááşá¸ áážááşáááşá¸áááşááŹá¸áááŻáˇ áááŻáĄááşááŤáááşá
Synthetic Translation Data áá˛áˇ áĄáŹá¸ááŹááťááşáá˛áˇ áĄáŹá¸áááşá¸ááťááş
FineTranslations áá˛áˇ áĄáŹá¸ááŹááťááşá ááŹááŹá ááŹá¸áĄááťáŹá¸áĄááźáŹá¸áĄáá˝ááş ááźáŽá¸ááŹá¸áá˛áˇ Parallel Dataset ááᯠááážááááŻááşááźááşá¸ ááźá áşááŤáááşá
ááŤááąáá˛áˇ ááŹááŹááźááşááŹá¸áá˛áˇ á áŹááŹá¸áá˝áąááᯠAI Model áá áşááŻá ááŻááşááŻááşááŹá¸ááŹááźá áşáá˛áˇáĄáá˝ááş áĄáážáŹá¸áĄáá˝ááşá¸áá˛áˇ áááŹááááťáá˛áˇ áĄááŻáśá¸áĄáážáŻááşá¸áĄááťááŻáˇ ááŤáááşáááŻááşááŤáááşá
áĄáá°á¸áááźááˇáş ááźááşááŹááŹááŹá ááŹá¸ááᯠáĄáááşáĄáá˝áąá¸ááźááˇáş Human Translation Data áááşá¸ááŤá¸áá˛áˇ ááŹááŹá ááŹá¸áá˝áąáĄáá˝ááş ááŽáĄááťááşááᯠáááˇáşáá˝ááşá¸á ááşá¸á áŹá¸áááŻáˇ áááŻáĄááşááŤáááşá
ááŤááźáąáŹááˇáş Dataset áĄáá˝ááşáĄá áŹá¸áá áşááŻáááşá¸áááŻáᏠáĄáŹá¸áááŻá¸áááąáᲠData Quality áááŻáááşá¸ áĄááąá¸ááźáŽá¸áá˛áˇ áĄááťááşáá áşááŻáĄááźá áş áááşáážááşááŹá¸ááŤáááşá
áá Fine-tuning áááşá¸ááᏠ- Unsloth áážááˇáş QLoRA
Gemma-4 ááᯠFine-tuning ááźáŻááŻááşááŹáážáŹ Unsloth Framework áá˛áˇ QLoRA áááşá¸áááŹááᯠáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
QLoRA ááᯠááŹá¸áááşáááŻáˇ Model áá áşááŻááŻáśá¸ááᯠááźááşáááşááąáˇááťááˇáşá ááŹááááŻáᲠáááŻáĄááşáá˛áˇ Parameters áĄáááşá¸áááşáááŻáᏠááźáŻááźááşáááŻááşáááşáááŻáˇ á ááşá¸á áŹá¸áááŻááşááŤáááşá
ááŽáááşá¸áááşá¸áážáŹ áá°á Model Weights áá˝áąááᯠ4-bit áĄááźá áş Quantization ááźáŻááŻááşááźáŽá¸ LoRA Adapter áá˝áąááᯠTraining ááŻááşááŤáááşá
LoRA Adapter áááŻáᏠáá°á Model áá˛áˇ Parameters áá˝áąááᯠáĄááźááˇáşáĄá ááźáąáŹááşá¸áá˛áááˇáşáĄá áŹá¸ áááşáá°áááŻááşáá˛áˇ Parameter áĄáááşá¸áááşááᯠáááşááąáŤááşá¸áááˇáşááŹá¸ááźááşá¸ ááźá áşááŤáááşá
ááŽáááŻááŻááşááąáŹááşááźááşá¸áĄáŹá¸ááźááˇáş GPU Memory áĄááŻáśá¸ááźáŻáážáŻááᯠááťážáąáŹáˇááťáááŻááşááźáŽá¸ Model áá áşááŻááŻáśá¸ááᯠFine-tuning ááźáŻááŻááşááŹáááş áááŻáááŻááááąáŹááşáá˛áˇ Training ááŻááşáááşá¸á ááşááᯠááážááááŻááşááŤáááşá
áĄááŻáśá¸ááźáŻáá˛áˇáá˛áˇ LoRA Configuration
Quantization : 4-bit
LoRA Rank: 16
LoRA Alpha: 16LoRA
Dropout: 0
Target Modules:all-linear
Trainable Parameters: 59.7M
Model Parameters á áŻá áŻááąáŤááşá¸ 5.18 Billion áááˇáşáážáááąáá˛áˇ Training ááźáŻááŻááşáá˛áˇáá˛áˇ Parameters á 59.7 Million áááˇáşáᏠáážáááŤáááşá
ááŤáᏠá áŻá áŻááąáŤááşá¸ Parameters áá˛áˇ 1.15% áááˇáş ááźá áşááŤáááşá
Training áážáŹ Unsloth áá˛áˇ Optimized Implementationá Gradient Checkpointing áá˛áˇ Hugging Face TRL áá˛áˇ SFTTrainer ááᯠáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
Hardware áĄááąáá˛áˇ NVIDIA H100 GPU ááᯠáĄááŻáśá¸ááźáŻááŹá¸ááŤáááşá
áá Model ááᯠáááşááᯠEvaluation ááŻááşáá˛áˇááá˛á
Training Loss ááťáááşá¸ááŹááŹáá˛áˇ Model áá˛áˇ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸ áááŻá¸áááşááŹáááşáááŻáˇ áááşá¸ááźáŽá¸ áááŻáśá¸ááźááşáááŻááşááŤáá°á¸á
ááŤááźáąáŹááˇáş Training ááźáŻááŻááşááąá ááş Model á ááŻááşááąá¸áá˛áˇ ááŹááŹááźááşá áŹááŹá¸áá˝áąááᯠááŽá¸ááźáŹá¸ Evaluation ááźáŻááŻááşáá˛áˇááŤáááşá
áᎠProject áážáŹ COMET áá˛áˇ chrF++ Metrics áážá áşááťááŻá¸ááᯠáĄááŻáśá¸ááźáŻáá˛áˇááŤáááşá
chrF++ - ááŹááŹááźááşá áŹááŹá¸ áááşááąáŹááşáá°ááá˛á
chrF++ áᏠModel áá˛áˇ ááŹááŹááźááşá áŹááŹá¸áá˛áˇ áážááşáááşáá˛áˇ Reference Translation ááᯠáážááŻááşá¸áážááşááąá¸áá˛áˇ Metric ááźá áşááŤáááşá
áĽááᏠáá°áá áşááąáŹááşá "ááťá˝ááşááąáŹáş áĄáááşá¸ááťáąáŹáşááŤáááş" áááŻáˇ ááŹááŹááźááşááŹá¸ááźáŽá¸ AI á "ááťá˝ááşááąáŹáş áĄáá˝ááşááťáąáŹáşáá˝ážááşááŤáááş" áááŻáˇ ááŻááşááąá¸áááşáááŻááŤá ááŻáˇá
á áŹááŻáśá¸ááťááşá¸ ááááááťááť ááá°ááąáá˛áˇ áááťááŻáˇ Character áá˛áˇ Word áĄáááŻááşá¸áĄá áá˝áą áá°ááŽááąááŤáááşá
chrF++ á ááŽáááŻáá°ááŽáážáŻáá˝áąááᯠCharacter N-grams áá˛áˇ Word N-grams áá˝áąáĄááąáŤáş áĄááźáąááśááźáŽá¸ áá˝ááşááťááşááąá¸ááŤáááşá
áááŻá¸áááŻá¸ááźáąáŹááááş AI áá˛áˇ ááŹááŹááźááşá áŹááŹá¸á Reference á áŹááŹá¸áá˛áˇ áááşááąáŹááşáááşáá°ááá˛áááŻááŹááᯠáááŻááşá¸ááŹááąá¸áᏠááźá áşááŤáááşá
Score ááźááˇáşááąááą á áŹááŹá¸áá°ááŽáážáŻ áááŻááťáŹá¸ááąááą ááźá áşááŤáááşá
COMET - ááŹááŹááźááşáĄáááášááŹááş áááşááąáŹááşáážááşááá˛á
COMET áááąáŹáˇ Neural Network áĄááźáąááśááŹá¸áá˛áˇ Translation Evaluation Metric ááźá áşááŤáááşá
chrF++ áá˛áˇ ááá°ááŹá COMET áᏠá áŹááŹá¸áá°ááŽáážáŻáááŻáᏠáĄááááááŹá¸ááŤáá°á¸á
áá°áááşá¸á áŹááŹá¸á Reference Translation áá˛áˇ AI áá˛áˇ ááŹááŹááźááşá áŹááŹá¸áááŻáˇááᯠáĄááŻáśá¸ááźáŻááźáŽá¸ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸áá˛áˇ áĄáááášááŹááşáááŻááşáᏠáá°ááŽáážáŻááᯠáĄáá˛ááźááşáááŻáˇ ááźááŻá¸á áŹá¸ááŤáááşá
áĽááᏠ-
Reference: "I am very happy."
AI Translation: "ááťá˝ááşááąáŹáş áĄáá˝ááşáááşá¸ááŹááŤáááşá"
Reference áá˛áˇ AI Translation áááŻáˇáᏠá áŹááŻáśá¸ááťááşá¸ ááá°ááąáá˛áˇ áĄáááášááŹááşáá°ááŽáááŻááşááŤáááşá
ááŽáááŻáĄááźáąáĄááąááťááŻá¸áážáŹ COMET á á áŹááŻáśá¸áá°ááŽáážáŻáá áşááŻáááşá¸áááş áááŻáááŻááťááşááźááˇáşáá˛áˇ áĄááťááşáĄáááşáá˝áąááᯠáĄááŻáśá¸ááźáŻááŤáááşá
áᎠProject áážáŹ áĄááŻáśá¸ááźáŻáá˛áˇáá˛áˇ Metric Model á Unbabel/wmt22-comet-da ááźá áşááŤáááşá
áááŻá¸áááŻá¸áážááşá¸áážááşá¸ááźáąáŹááááş chrF++ á á áŹááŹá¸áááşááąáŹááşáá°ááá˛áááŻááŹááᯠááźááˇáşááźáŽá¸ COMET á ááŹááŹááźááşáĄáááşáĄáá˝áąá¸áá˛áˇ áĄáááášááŹááşáááŻááşáᏠáážááşáááşáážáŻááᯠáááŻáááŻáĄáŹááŻáśá ááŻááşááŤáááşá
Metric áážá áşááťááŻá¸ááᯠáá˝á˛áááşáĄááŻáśá¸ááźáŻááźááşá¸áĄáŹá¸ááźááˇáş Model áá˛áˇ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸ááᯠááá°ááŽáá˛áˇ áážáŻááąáŹááˇáşáážá áşááŻáááą ááąáˇááŹáááŻááşááŤáááşá
á á áááŚá¸ Evaluation ááááşááťáŹá¸
Training ááźáŻááŻááşááąá ááş Step 50á 100 áá˛áˇ 150 áááŻáˇáážáŹ Evaluation ááźáŻááŻááşáá˛áˇááŤáááşá
áááşáážáááážáááŹá¸áá˛áˇ ááááşáá˝áąáááąáŹáˇ -
Training Step: 50 chrF++ : 38.03 COMET: 0.7878
Training Step: 100 chrF++ : 37.67 COMET: 0.7813
Training Step: 150 chrF++ : 35.90 COMET: 0.7213
Step 50 áážáŹ COMET 0.7878 áá˛áˇ chrF++ 38.03 ááážááá˛áˇááŤáááşá
ááŤááąáá˛áˇ Step 100 áá˛áˇ 150 áážáŹ Score áá˝áą áááşáááŻááşááťáááşá¸ááŹááŹááᯠáá˝áąáˇáááŤáááşá
ááŽááąááŹáážáŹ á áááşáááşá áŹá¸á ááŹááąáŹááşá¸áá˛áˇ áĄááťááşáá áşáᯠáážáááŤáááşá
Training Loss á ááťáááşá¸ááąááąáá˛áˇ Evaluation Score áá˝áąáááąáŹáˇ ááťáááşá¸ááŹááŤáááşá
ááŽáááŻááźá áşáááźááşá¸áá˛áˇ ááźá áşáááŻááşáá˛áˇ áĄááźáąáŹááşá¸áááşá¸áá áşááŻá Overfitting ááźá áşááŤáááşá Model áᏠTraining Data áá˝áąááᯠáááŻáááŻáážááşááááŹááąáá˛áˇ áááźááşáá°á¸ááąá¸áá˛áˇ á áŹááŹá¸áĄáá áşáá˝áąááᯠááŹááŹááźááşáá˛áˇáĄá፠áĄáááşáĄáá˝áąá¸ ááťáááşá¸ááŹáááŻááşááŤáááşá
ááŤááąáá˛áˇ áááşáážáááááşáá˝áąáá˛áˇ Overfitting ááźá áşááąááźáŽáááŻáˇ áĄáááşááźáŻáááŻáˇááąáŹáˇ áááŻáśááąáŹááşááąá¸ááŤáá°á¸á
Evaluation áážáŹ Sample 150 ááŻáááŻáᏠáĄááŻáśá¸ááźáŻááŹá¸ááŹááźáąáŹááˇáş áááŻáááŻááźáŽá¸ááŹá¸áá˛áˇ Test Set áá˛áˇ áááşááśá á áşááąá¸áááŻáˇ áááŻáĄááşááŤáááşá
áááşáážáááááşáá˝áąááᯠáĄááąáŹááşá¸ááŻáśá¸ Model áá˛áˇ ááąáŹááşááŻáśá¸á á˝ááşá¸ááąáŹááşáááşáĄááźá áş áááŻááşáᲠáááŚá¸áá˝áąáˇáážáááťááşáá˝áąáĄááźá áşáᏠáááşáážááşááŹá¸ááŤáááşá
áá ááąáŹááşáááşááŻááşááąáŹááşáá˝áŹá¸áááˇáşáĄááŹááťáŹá¸
áᎠProject áᏠáááşáážááážáŹáááşá¸ áááşáááşá ááşá¸áááşááźáŽá¸ áááŻáááŻááąáŹááşá¸áá˝ááşáĄáąáŹááş ááźáŻááźááşááąáᲠááźá áşááŤáááşá
ááąáŹááşáááşáĄáááˇáşáá˝áąáážáŹ áĄáááááŹá¸ ááŻááşááąáŹááşáá˝áŹá¸áááˇáşáĄááŹáá˝áąáááąáŹáˇ -
Dataset Quality ááᯠáááŻáááŻááąáŹááşá¸áá˝ááşáĄáąáŹááş á á áşááąá¸ááźááşá¸á
Learning Rate áá˛áˇ LoRA Rank áááŻáˇááᯠáááşááśá ááşá¸áááşááźááşá¸á
Burmese â English áá˛áˇ English â Burmese ááᯠááŽá¸ááźáŹá¸ Evaluation ááźáŻááŻááşááźááşá¸á
áááŻáááŻááźáŽá¸ááŹá¸áá˛áˇ Test Set áĄááŻáśá¸ááźáŻááźááşá¸á
áá°ááŹá¸ááŹááŹááźááşáá°áá˝áąáá˛áˇ áĄáá˛ááźááşáážáŻááᯠáááˇáşáá˝ááşá¸ááźááşá¸á
áá°áááşá¸ Gemma-4 Model áá˛áˇ Fine-tuned Model áááŻáˇááᯠáá°ááŽáá˛áˇ Test Set ááąáŤáşáážáŹ áážááŻááşá¸áážááşááźááşá¸á
áĄáá°á¸áááźááˇáş Human Evaluation áᏠáĄááąá¸ááźáŽá¸ááŤáááşá Automatic Metrics áá˝áąá ááŹááŹááźááşáĄáááşáĄáá˝áąá¸ááᯠáááˇáşáážááşá¸ááąá¸áááŻááşááąáá˛áˇ áá°ááŹá¸áá áşááąáŹááşáá˛áˇ áĄááźááşáá˛áˇ áááŹáááťáážáŻá áĄáááášááŹááşááááťáážáŻáááŻáˇááᯠáĄááźááˇáşáĄá áĄá áŹá¸áááŻá¸áááŻááşáᏠáááŻááşááŤáá°á¸á
ááááŻáśá¸ááťáŻááş
áᎠProject áᏠGemma-4 ááᯠáĄááźáąááśááźáŽá¸ ááźááşááŹ-áĄááşášáááááş ááŹááŹááźááşáááŻáážáŻáĄáá˝ááş áááŻááşáááŻááş AI Model áá áşáᯠáááşááąáŹááşáááŻáˇ ááźááŻá¸á áŹá¸ááŹá¸áá˛áˇ á ááşá¸áááşáážáŻáá áşáᯠááźá áşááŤáááşá
Unsloth áá˛áˇ QLoRA ááᯠáĄááŻáśá¸ááźáŻááźáŽá¸ Model Parameters áá˛áˇ 1.15% áááˇáşáááŻáᏠFine-tuning ááźáŻááŻááşáááŻááşáá˛áˇááŤáááşá
áááŚá¸ Evaluation ááááşáá˝áąá á áááşáááşá áŹá¸á ááŹááąáŹááşá¸ááąáá˛áˇ Training Step áááŻá¸ááŹááŹáá˛áˇáĄááťáž Validation Score ááťáááşá¸ááŹáá˛áˇáĄáá˝ááş áááşááśá ááşá¸áááşáááŻáˇ áááŻáĄááşááąááŤáááşá
ááŽá ááşá¸áááşáážáŻáááą áááážáááá˛áˇ áĄááąá¸ááźáŽá¸ááŻáśá¸áĄááťááşáá áşááŻá Model áá áşááŻáá˛áˇ ááŹááŹááźááşáĄáááşáĄáá˝áąá¸áᏠModel Size áá˛áˇ Training Data áááŹááá áşááŻáááşá¸áĄááąáŤáş ááá°áááşáᲠData Qualityá Training Strategy áá˛áˇ Evaluation Method áá˝áąáĄááąáŤáşáááşá¸ áá°áááşááąááźááşá¸ ááźá áşááŤáááşá
ááźááşááŹááŹááŹá ááŹá¸áĄáá˝ááş áááŻáááŻááąáŹááşá¸áá˝ááşáá˛áˇ AI áááşá¸áááŹáá˝áą áááşááąáŹááşáááŻááşáááŻáˇ áᎠReseachááᯠáááşáááşááąáˇááŹááźáŽá¸ áááŻá¸áááşáĄáąáŹááş ááŻááşááąáŹááşáá˝áŹá¸áááŻáˇ áááşáá˝ááşááŹá¸ááŤáááşá
References
1.Gemma-4 Technical Report (2026) â Gemma-4 áá˛áˇ Architectureá Model Design áá˛áˇ Multimodal Capabilities áĄááźáąáŹááşá¸á
https://arxiv.org/abs/2607.02770
2.LoRA: Low-Rank Adaptation of Large Language Models (2021) â LoRA áááşá¸áááŹáá˛áˇ áá°áááşá¸ Research Paperá
https://arxiv.org/abs/2106.09685
3.QLoRA: Efficient Finetuning of Quantized LLMs (2023) â 4-bit Quantization áá˛áˇ Parameter-efficient Fine-tuning áĄááźáąáŹááşá¸á
https://arxiv.org/abs/2305.14314
4.COMET: A Neural Framework for MT Evaluation (2020) â Neural Network áĄááźáąááś Machine Translation Evaluation áááşá¸áááŹá
https://aclanthology.org/2020.emnlp-main.213/
5,chrF++: Words Helping Character N-grams (2017) â Character áá˛áˇ Word N-grams ááᯠááąáŤááşá¸á ááşáĄááŻáśá¸ááźáŻáá˛áˇ Translation Evaluation Metricá
https://aclanthology.org/W17-4770/
6.FineTranslations Dataset (2026) â Multilingual Parallel Translation Dataset áá˛áˇ áá°áááşá¸áĄááźáąáŹááşá¸áĄááŹáá˛áˇ Dataset Constructioná
https://huggingface.co/datasets/HuggingFaceFW/finetranslations
7.Unsloth Documentation â Efficient Fine-tuning Implementation áá˛áˇ Training Configurationá
https://unsloth.ai/docs/
8.Hugging Face TRL Documentation â Supervised Fine-tuning Trainer áá˛áˇ Training Workflowá
https://huggingface.co/docs/trl/
Read More âMyanmar OCR - Tesseract OCR Fine-tuning
Many Myanmar newspapers exist only as scanned pages or PDF files.
A computer cannot easily read the Burmese text inside them, so this text cannot be searched or used to build Burmese AI models.
Our OCR (Optical Character Recognition) research builds a system that reads these newspaper pages and turns them into clean, accurate Burmese text.
Why this is hard
Burmese letters stack and join in complex ways. For example, the word ááźááşáᏠ("Myanmar") is stored as six characters. One of them (áź) wraps around the first letter, and another (áş) is a small mark at the top. An OCR tool must read every mark correctly, because one small mistake changes the word. Burmese also does not always put spaces between words.
Newspaper pages add more problems: many columns, headlines in large decorative fonts, photos, advertisements and boxes of text side by side. Very few OCR tools are trained on Burmese, and almost none on Myanmar newspapers.
How the system works
Two parts of this system are models that we fine-tuned ourselves: the layout model and the OCR engine. The sections below explain how we built them and how well they work.
Step 1: Comparing existing tools
We first tested seven existing page-layout models on real Myanmar newspaper pages, including DocLayout-YOLO [1], PP-DocLayout [4], RT-DETR [5] and DeepSeek-OCR [6]. On the same page, one model found 59 parts of the page while another found only 5. No single tool worked well enough on its own, so we decided to fine-tune our own models.
Step 2: Creating our own training data
To teach a model, we need examples with the correct answers. On nearly 500 newspaper pages, we drew a box around every part of the page (headline, body text, photo, caption, advertisement and so on), gave it a label, and marked the reading order. We always tested the models on pages from newspaper issues that they had never seen, so the scores show how they work on new pages.
Step 3: Fine-tuning the layout model
What is fine-tuning?
Fine-tuning means taking a model that has already learned from many general examples and training it a little more on our own examples. The model keeps what it already knows and learns the special features of our data. This needs much less data and time than training a new model from zero.
We started from DocLayout-YOLO [1], a model already trained on many kinds of documents such as reports and books, and trained it further on our labelled pages.
Result: the detection score rose from 49% to 91%, and the model now finds 89% of the parts of a page, up from 44%. (The detection score shows how many parts the model finds and how accurately it places the boxes. 100% is perfect.)
Example: the model finds most advertisements, photos and body text areas. Headlines are the hardest part, because their size and style change a lot from page to page.
Next version: a newer version that is better at small headlines is being tested. We will use it once it passes all our checks.
Step 4: Fine-tuning the OCR engine
Tesseract [2] is a free, open-source OCR engine. Its standard Burmese model made many mistakes on newspaper print, so we fine-tuned it with about 10,700 line images from our own pages, each paired with its correct text. We then tested it on a full newspaper issue that it had never seen.
Character errors (lower is better)
Old setup (standard model) 5.7%
Our fine-tuned model 3.4%
The character error rate is the share of characters that are wrong: 3.4% means about 3 wrong characters in every 100. Our model makes about 40% fewer mistakes. Here are three real lines from the test issue, read by the old setup and by our fine-tuned model:
Example 1: numbers read correctly
Text (mistakes in underline)
Old setup (before): áá˝á˛ááťáááş áá áááá áşáážááˇáş áá˝á˛ááťááᲠááŻá áááá áş
Fine-tuned model (after): áá˝á˛ááťáááş áá áááá áşáážááˇáş áá˝á˛ááťáááş áá áááá áş
Correct text: áá˝á˛ááťáááş áá áááá áşáážááˇáş áá˝á˛ááťáááş áá áááá áş
Meaning in English:: "in the 18th minute and the 74th minute of the match". The old setup read the number 7 (á) as letters.
Example 2: stacked letters read correctly
Text (mistakes in underline)
Old setup (before): á áá˛áˇááŽá¸áážáśáá˝áą á ááŻááşááá°á°á¸ááŤáááş á ááŻááşááťááŻá¸áá˛áˇ
Fine-tuned model (after): á áá˛áˇááŽá¸áážáśáá˝áą á ááŻááşááťááŻá¸ááŤáááş á ááŻááşááťááŻá¸áá˛áˇ
Correct text: á áá˛áˇááŽá¸áážáśáá˝áą á ááŻááşááťááŻá¸ááŤáááş á ááŻááşááťááŻá¸áá˛áˇ
Meaning in English:: "... we grow crops such as these. The crops we grow ...". The old setup misread the word for "grow".
Example 3: much better, but not perfect
Text (mistakes in underline)
Old setup (before) : áźááŻááşáá˝á˛áááşáá áŹá¸áááŹá¸ 99095) ááąáŹááşáááŻáˇ ááŤáááş
Fine-tuned model (after): ááźááŻááşáá˝á˛áááşáá áŹá¸áááŹá¸ ááááá ááąáŹááşáááŻáˇ ááŤáááş
Correct text: ááźááŻááşáá˝á˛áááşáá áŹá¸áááŹá¸ áááá ááąáŹááşáááŻáˇ ááŤáááş
Meaning in English: "1,375 players took part in the competition". The fine-tuned model read the words correctly but added one extra digit to the number.
Step 5: Testing AI text correction
We also asked Gemma 4 [3], a large language model, to look at each line image and its OCR text and fix the mistakes. We compared the results with text typed by a person. Here are two real lines from this test, using the OCR from before fine-tuning:
Example 1: all mistakes fixed
OCR output (before): ááŽááŹááśááźáŽá¸á፠ááŻáśá¸ááźááşááťááşááźááˇáş áážá
áşá
ááş ááŽá§áá˛áᏠá áááşááąáˇááᯠáĄááźááşááźááşáááŻááşáᏠááááá
ášáááşá¸áá°ááťáŹá¸ááąáˇ
After AI correction: ááŽááŹááśááźáŽá¸á ááŻáśá¸ááźááşááťááşááźááˇáş áážá áşá ááş ááŽáááşáᏠá áááşááąáˇááᯠáĄááźááşááźááşáááŻááşáᏠááááşá á˝ááşá¸áá°ááťáŹá¸ááąáˇ
Correct text (typed by a person) :ááŽááŹááśááźáŽá¸á ááŻáśá¸ááźááşááťááşááźááˇáş áážá áşá ááş ááŽáááşáᏠá áááşááąáˇááᯠáĄááźááşááźááşáááŻááşáᏠááááşá á˝ááşá¸áá°ááťáŹá¸ááąáˇ
Meaning in English: "By decision of the General Assembly, 3 December every year [was named] International Day of Persons with Disabilities"
Example 2: the number is fixed, but one mistake is missed
Text (mistakes in underline)
OCR output (before)
áááá ááŻáážá áş áĄáąáŹááşáááŻáᏠáááááşáá˝ááş ááťááşá¸ááá˛áˇááąáŹ (áááŻ) ááźáááşááźáąáááş ááŻááááášáááĄáá˝áąáá˝áą
After AI correction: áááá ááŻáážá áş áĄáąáŹááşáááŻáᏠáááááşáá˝ááş ááťááşá¸ááá˛áˇááąáŹ (áá) ááźáááşááźáąáŹááş ááŻááááášáááĄáá˝áąáá˝áą
Correct text (typed by a person): áááá ááŻáážá áş áĄáąáŹááşáááŻáᏠáááááşáá˝ááş ááťááşá¸ááá˛áˇááąáŹ (áá) ááźáááşááźáąáŹááş ááŻááááášááĄáá˝áąáá˝áą
Meaning in English: "... held on 14 October 1992, the 47th United Nations General [Assembly] ..."
Better overall: lines with no mistakes rose from 28% to 47%.
Not perfect: the AI sometimes added new mistakes, so people still need to review important text.
Faster with better OCR: with our fine-tuned OCR, only a few unclear lines need AI correction. On a test page, the correction time fell from tens of minutes to about 17 seconds.
Step 6: Building Version 4
Many newer newspaper PDFs already contain the real text inside the file. Reading that text directly is faster and more accurate than OCR. Our current system, Version 4, checks every page: if the PDF text is usable, it uses that text; if not, for example on a scanned page, it uses our fine-tuned OCR. It then puts headlines, author names and body text in reading order and saves them as a clean dataset.
Quality check: we compared 598 lines with text typed by a person. 572 lines matched exactly, and 99.9% of characters were correct (not counting spaces).
Key results
Finding the parts of a page (layout model)| Detection score rose from 49% to 91%. | Newspaper pages not used in training
Reading Burmese text (OCR) | Character errors fell from 5.7% to 3.4%, about 40% fewer mistakes.| A newspaper issue not used in training
AI text correction| Lines with no mistakes rose from 28% to 47%.| Checked against text typed by a person
Correction speed| Time per page fell from tens of minutes to about 17 seconds.| One test page
Final text quality (Version 4)| 99.9% of characters were correct (not counting spaces).| Checked against text typed by a person
Text extracted so far
Our system has extracted Burmese text from 787 newspaper pages across three Myanmar newspapers, about 2.8 million tokens in total. A token is a small piece of text that language models read.
Limits and what comes next
The Version 4 quality check covered pages with usable PDF text. Next, we will check scanned pages read by OCR in the same way.
The OCR engine was trained on one newspaper's print. We plan to add more newspapers and years and retrain it with more data.
Small headlines are still the hardest part for the layout model. A new version is being tested.
References
1.Zhao, Z., Kang, H., Wang, B. and He, C. (2024). DocLayout-YOLO: Enhancing Document Layout Analysis through Diverse Synthetic Data and Global-to-Local Adaptive Perception. arXiv:2410.12628. https://arxiv.org/abs/2410.12628
2.Smith, R. (2007). An Overview of the Tesseract OCR Engine. Proceedings of the Ninth International Conference on Document 3.Analysis and Recognition (ICDAR 2007), pp. 629â633. https://research.google/pubs/an-overview-of-the-tesseract-ocr-engine/
Gemma Team (2026). Gemma 4 Technical Report. arXiv:2607.02770. https://arxiv.org/abs/2607.02770
4.Sun, T., Cui, C., Du, Y. and Liu, Y. (2025). PP-DocLayout: A Unified Document Layout Detection Model to Accelerate Large-Scale Data Construction. arXiv:2503.17213. https://arxiv.org/abs/2503.17213
5.Zhao, Y., Lv, W., Xu, S., et al. (2023). DETRs Beat YOLOs on Real-time Object Detection. arXiv:2304.08069. https://arxiv.org/abs/2304.08069
6.Wei, H., Sun, Y. and Li, Y. (2025). DeepSeek-OCR: Contexts Optical Compression. arXiv:2510.18234. https://arxiv.org/abs/2510.18234
Read More âBurmese Language AI Research
Advancing Burmese language AI through model development, quality datasets, practical applications, reliable evaluation, and responsible AI practices.
Insights, Research and News
Myanmar Intelligence. Global Possibilities.
MyMind AI is building artificial intelligence that understands Myanmar, serves its people, supports its industries, and connects the nation to the future.
Let's Build Myanmar's AI Future Together
We welcome collaboration with researchers, universities, technology companies, government institutions, development organizations, and industry leaders committed to responsible and meaningful AI innovation.



