《Sidama AI: Low-Resource Language Translation》
An English-to-Sidama Afoo translation engine built for accessibility and linguistic preservation.
●Project Overview
Sidama AI was developed to address the digital scarcity of the Sidama language. This project provides a resilient, web-based interface that bridges the gap between English and one of Ethiopia's key local languages. The core focus is on creating a functional tool that operates efficiently in low-bandwidth environments.
●Technical Architecture
Unlike standard translation tools that rely on constant API calls, Sidama AI utilizes a client-side search architecture to ensure zero latency and offline-first potential.
1. Development Environment (Constraint-Driven)
This project was engineered entirely within a mobile environment. The full development lifecycle—from data architecture to cloud deployment—was executed on a Samsung Galaxy A13 via Termux. This required a self-taught mastery of Linux environments, CLI-based version control, and resource-efficient coding practices outside of a traditional academic setting.
2. Search Logic and Algorithmic Resilience
To handle phonetic variations and common orthographic errors in Sidama language input, the system employs Approximate String Matching:
•Algorithm: Bitap search logic via Fuse.js.
•Optimization: Weighted keys were implemented to prioritize English phrase matches while maintaining a strict threshold to minimize false positives.
3. Data Engineering Pipeline
The backend "knowledge base" is a structured JSON dataset curated through:
Manual Digitalization: Converting physical and academic linguistic records into structured data.
Automation: Python-based merging scripts used to deduplicate and normalize raw text into a production-ready dictionary.
●Current Stack
•Languages: JavaScript (ES6), Python 3.x, HTML5, CSS3
•Deployment: Vercel (CI/CD Pipeline)
•Libraries: Fuse.js for fuzzy matching
●Research & Future Roadmap
•Speech-to-Speech: …