MULTIFINGROUND benchmarks multilingual financial evidence extraction
Original title: Recognition Is Not Grounding: Benchmarking Multilingual Financial Evidence Extraction
Eight reference models reach 97.3-99.4% concept recognition but miss a target fact in 26.6-41.9% of those groups.
The paper introduces MULTIFINGROUND to test whether models can both recognize shared accounting concepts across languages and recover all related numerical facts. It is built from FY2025 disclosures of 90 issuers in English, French, and Greek, aligned through a shared accounting taxonomy with fact-level annotations. Across 15 models, strong concept recognition coexists with incomplete fact recovery, and finance-specialized models show limited additional gains.
- The corpus covers 90 issuers and three languages.
- Tasks include fact identification, recognition, selection, and discrimination.
- Evaluations use eight API and open-weight reference models among 15.
- Concept discrimination varies substantially across models.
- AI models
- MULTIFINGROUND
From the original article
1 min read at huggingface.co
Comparing financial disclosures across languages requires identifying shared accounting concepts and recovering all numerical facts that express them. Existing benchmarks evaluate aspects of these capabilities, but provide limited insight into whether both succeed on the same multilingual evidence.
By Lingfei Qian, Yan Wang, Xueqing Peng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.