Applying Flow Matching only to translation pairs, with contrastive and teacher-preserving losses for other tasks, yields higher Indic MTEB scores across five embedding backbones.
Transactions of the Association for Computational Linguistics , volume=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Task-Conditional Flow Matching for Balanced Multilingual Text Embedding Adaptation
Applying Flow Matching only to translation pairs, with contrastive and teacher-preserving losses for other tasks, yields higher Indic MTEB scores across five embedding backbones.