All projects
AI & Computing
Adaptive Language Models for Low-Resource Tongues
Building parameter-efficient transformer architectures that learn from fewer than ten thousand sentences, extending modern NLP to languages historically excluded from the digital record.
Objectives
- →Curate aligned corpora for eight under-resourced languages
- →Benchmark adapter-based fine-tuning against full fine-tuning
- →Release an open evaluation suite for morphological richness
Methodology
A mixed pipeline of community-sourced annotation and self-supervised pretraining, evaluated through held-out translation and cloze tasks with native-speaker review panels.
Current status
64%