Results and model outputs from the paper: MLLM-as-a-Judge for Financial Document Image Machine Translation

The paper explores an end-to-end (E2E) Document Image Machine Translation (DIMT) approach using Gemma 4 on financial reports from IBEX 35 companies. Since traditional Machine Translation (MT) metrics ignore the source image during evaluation, we propose a Multimodal LLM-as-a-judge (MLLMJ) to provide a grounded and explainable evaluation that considers both visual and textual information.

It follows a rubric to assign scores and provides both a justification and an internal thinking process leading to the assigned score. Additionally, we estimate the system's potential upper bound through Oracle-Guided Self-Refinement (OGSR), using the judge’s feedback to rectify initial predictions. Our study evaluates various configurations, including zero-shot and fine-tuning scenarios across different model sizes.