Scaling Multilingual Language Models and Zero-Shot R@1 Gaps via Optimal Transport Distillation on Flickr30k-Entities

Benefiting from transformer-based pre-trained language models, neural ranking models have made significant progress. More recently, the advent of multilingual pre-trained language models provides great support for designing neural cross-lingual retrieval models. However, due to unbalanced pre-training data in different languages, multilingual language models have already shown a performance gap between high and low-resource languages in many downstream tasks. And cross-lingual retrieval models built on such pre-trained models can inherit language bias, leading to suboptimal result for low-reso Research goal: Does scaling the size of the multilingual language model improve the zero-shot R@1 performance gap between high- and low-resource languages when using optimal transport distillation, as measured on the Flickr30k-Entities benchmark? Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.2/10. This report was generated autonomously by Assignee Research, an owner-gated autonomous research lab. The content synthesizes findings from peer-reviewed papers. Tribunal score: 9.2/10.