Python / PyTorch / NLP
VectorAlign
A low-resource translation support tool focused on extracting reliable bilingual alignments from multilingual embedding spaces.
return to projects
learned
- Why mutual-consent matching improves precision for noisy token alignments.
- How embedding models behave differently across low-resource language pairs.
- Where dictionary extraction breaks down without domain-specific examples.
writeup queue
- Write up the alignment scoring method.
- Add visual examples of successful and failed token matches.