Mitacs Globalink Research Intern - Collaborating Researcher

Description

• Co-authored a study on online adaptive language sampling for better distributed cross-lingual transfer, accepted to Findings of EMNLP 2026.
• Lead its direct continuation, studying when to stop cross-lingual realignment: more realignment is not always better for transfer, and low-resource languages offer no labeled data to tell when to stop; paper under review at ACL Rolling Review (ARR).
• Co-develop M2-ALIGN, studying where multilingual alignment should live in vision-language models: how to extend visual grounding to 11 languages, including low-resource ones, without eroding the multilingual reasoning the base models already have; ongoing work on culturally grounded evaluation (CVQA).
• Contributed to a PRISMA systematic literature review of multilingual, multimodal, and multicultural conversational agents.