When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents

Virginie Mouilleron, Théo Lasnier, Anna Mosolova, Djamé Seddah


Abstract
Vision-language models (VLMs) perform well on many document understanding tasks, yet their reliability in specialized, non-English domains remains underexplored. This gap is especially critical in finance, where documents mix dense regulatory text, numerical tables, and visual charts, and where extraction errors can have real-world consequences. We introduce SCRIBE FINANCE, the first multimodal benchmark for evaluating French financial document understanding. The dataset contains 1,204 expert-validated questions spanning text extraction, table comprehension, chart interpretation, and multi-turn conversational reasoning, drawn from real investment prospectuses, KIDs, and PRIIPs. We evaluate six open-weight VLMs (8B–124B parameters) using an LLM-as-judge protocol. While models achieve strong performance on text and table tasks (85–90% accuracy), they struggle with chart interpretation (34–62%). Most notably, multi-turn dialogue reveals a sharp failure mode: early mistakes propagate across turns, driving accuracy down to roughly 50% regardless of model size. These results show that current VLMs are effective for well-defined extraction tasks but remain brittle in interactive, multi-step financial analysis. SCRIBE FINANCE offers a challenging benchmark to measure and drive progress in this high-stakes setting.
Anthology ID:
2026.fnp-1.2
Volume:
The 7th Financial Narrative Processing Workshop
Month:
May
Year:
2026
Address:
Palma de Mallorca, Spain
Editors:
Mo El-Haj, Antonio Moreno Sandoval, Ana Garcia-Serrano, Chung-Chi Chen, Paul Rayson, Yanco Amor Torterolo Orta, Paloma Martinez, Jordi Porta
Venues:
FNP | WS
SIG:
Publisher:
European Language Resources Association (ELRA)
Note:
Pages:
12–27
Language:
External URL:
https://lrec.elra.info/lrec2026-ws-fnp-02
DOI:
10.63317/3bk8fk9sek7b
Bibkey:
Cite (ACL):
Virginie Mouilleron, Théo Lasnier, Anna Mosolova, and Djamé Seddah. 2026. When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents. In The 7th Financial Narrative Processing Workshop, pages 12–27, Palma de Mallorca, Spain. European Language Resources Association (ELRA).
Cite (Informal):
When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents (Mouilleron et al., FNP 2026)
Copy Citation: