Setting up a machine translation service for Timor-Leste
summary
A volunteer project builds a machine translation service for Tetun from near-zero resources, scraping parallel articles, aligning sentences, standardizing spelling and training a neural model on ~120k sentence pairs with Fairseq on a P100 GPU (half‑day training, ~15–20 iterations). The model scores BLEU ~38–40, serves via a Django API, powers web and mobile apps and sustains 10,000 monthly active users for about $250/year while work continues on conversational data and synthetic backtranslation.
AI-written summary · about
chapters (5)
related talks
45mTranscend Language Barriers with Custom Machine Learning APIs: um...
44mSome tentative first steps towards a Star Trek universal communicator
43mBreaking the language barrier: how do we quickly add multilanguage support in our AI application?
26mMachine Translation into Learning
23mMultilingual Natural Language Processing
33mTranslation Augmented Generation - Breaking Language Barriers in RAG...
26mUsing deep learning in natural language processing
29mBuilding Machine Learning demos with Python