NeuralMind-UNICAMP at 2022 TREC NeuCLIR: Large Boring Rerankers for Cross-lingual Retrieval

TL;DR

This study demonstrates that a large fine-tuned mT5-XXL reranker, trained on same-language pairs, effectively improves cross-lingual information retrieval performance across multiple languages, even with suboptimal initial retrievals.

Contribution

It shows that a model fine-tuned only on monolingual pairs can be successfully applied to cross-lingual retrieval tasks, achieving outstanding results.

Findings

mT5-XXL reranker outperforms baselines in CLIR tasks

Model achieves high accuracy across multiple languages

Study provides insights into cross-lingual reranking effectiveness

Abstract

This paper reports on a study of cross-lingual information retrieval (CLIR) using the mT5-XXL reranker on the NeuCLIR track of TREC 2022. Perhaps the biggest contribution of this study is the finding that despite the mT5 model being fine-tuned only on query-document pairs of the same language it proved to be viable for CLIR tasks, where query-document pairs are in different languages, even in the presence of suboptimal first-stage retrieval performance. The results of the study show outstanding performance across all tasks and languages, leading to a high number of winning positions. Finally, this study provides valuable insights into the use of mT5 in CLIR tasks and highlights its potential as a viable solution. For reproduction refer to https://github.com/unicamp-dl/NeuCLIR22-mT5