#3

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.