Multilingual and cross-lingual effectiveness of TEMPS

Determine the effectiveness of TEMPS, the temporal embedding branch for dense retrieval, in multilingual and cross-lingual retrieval settings beyond the English-language benchmarks evaluated in the paper.

Background

TEMPS is evaluated exclusively on English-language data, although its model-agnostic architecture is intended to augment different semantic retrievers. The paper explicitly identifies the absence of evaluation in other languages as an unresolved issue and notes that multilingual and cross-lingual effectiveness cannot be inferred from the reported English experiments.