HiligaynonEngine: Building an Open-Source NLP Ecosystem for Low-Resource Languages

HiligaynonEngine is a community-driven machine learning and NLP platform focused on the processing, translation, and preservation of the Philippine Hiligaynon language, covering a complete technical roadmap from corpus construction to neural machine translation. The project aims to fill the gap in NLP for the Hiligaynon language, enabling it to have equal technical expression opportunities in the digital age, and it has the dual significance of being both a technical project and a language preservation initiative.

In the global development of AI, high-resource languages dominate the NLP field, and thousands of low-resource languages face the risk of technical marginalization. Hiligaynon (Ilonggo) is a major regional language in the Philippines with approximately 7 million speakers, but it has almost no presence in the NLP field. The HiligaynonEngine project was born to fill this gap; it is not only a technical project but also a language preservation initiative—by building open-source NLP infrastructure, it enables Hiligaynon to have equal technical expression opportunities in the digital age.

The project adopts a modular architecture, decomposed into independently developable subsystems:

1. Corpus Construction Layer: Community contribution system (sentence submission, voting verification), JSON structured storage, initial target of 1k-5k parallel sentence pairs;
2. Preprocessing Layer: Tokenizer tailored to Hiligaynon's morphological features, text normalization (handling non-standard spelling), sentence splitter;
3. Morphological Analysis Layer: Supports prefix analysis (e.g., naga-, gin-), root extraction, basic POS tagging;
4. Translation Engine Layer: Three-stage strategy—rule-based baseline translator (dictionary mapping + grammar reordering), neural machine translation (pre-trained model fine-tuning + BLEU evaluation), hybrid optimization (rules + ML error correction + confidence scoring).

Complete Processing Flow: Input text → Tokenizer → Normalizer → Morphological Analyzer → Translation Engine → Post-processor → Output translation. Technology Stack Selection:

Layer	Technology Selection	Description
Backend	ASP.NET Core / Node.js	Flexible API services
Frontend	React / Next.js	Contributor interface and dashboard
Database	PostgreSQL	Structured corpus storage
Machine Learning	Python (PyTorch / Hugging Face)	Model training and inference
NLP Tools	Custom tokenizer + Transformers	Domain-specific processing

The project adopts an open-source community-driven model and welcomes various contributions:

• Add English-Hiligaynon parallel sentence pairs;
• Improve the quality of existing translations;
• Refine morphological analysis rules;
• Develop tokenization logic;
• Participate in neural translation model optimization. The contribution process follows the standard GitHub workflow: Fork → Create branch → Add data/features → Submit PR → Review and merge.

Future Outlook: Expand to speech recognition/synthesis, grammar correction AI, multilingual transfer (Cebuano, Tagalog, etc.), mobile translation applications. Core Insights: HiligaynonEngine provides a replicable path for NLP construction for low-resource languages—community collaboration + progressive technical strategy, enabling the building of complete digital infrastructure even without large-scale labeled data; technical inclusivity requires that every language has its rightful voice in the digital world.