BaFCo: A Document Understanding Benchmark for Complex Bangla Form Comprehension

2026-07-06

AI Summary

This study addresses the challenge of applying multimodal large language models to complex form understanding in low-resource languages like Bengali, which suffer from a scarcity of high-quality annotated data. To bridge this gap, the authors introduce BaFCo, the first benchmark dataset comprising 200 multi-page Bengali government forms spanning domains such as agriculture, education, banking, and land administration. They propose a dual-level entity annotation scheme with both fine-grained (26 categories) and coarse-grained (5 categories) labels. A systematic evaluation using state-of-the-art models—including ChatGPT, Gemini, Claude, Qwen, and Kimi—combined with zero-shot and chain-of-thought prompting reveals significant limitations in current models’ ability to accurately localize and comprehend fine-grained entities. This work provides a foundational resource and clear directions for advancing document intelligence in low-resource linguistic settings.

Abstract

Document comprehension is a challenging yet impactful task for Multimodal Large Language Models, especially as these systems see growing adoption in real-world, human-centric applications. However, this adoption is limited for low-resource languages such as Bangla due to the scarcity of high-quality annotated data. To address this gap, we introduce BaFCo, a benchmark dataset for Bangla form comprehension with a focus on Document Layout Analysis (DLA) and Key Information Extraction (KIE). BaFCo curates 200 multi-page complex Bangladeshi government forms, sourced from across diverse sectors including agriculture, education, banking, and land management. To accurately capture the structural and contextual complexity of these forms, we define a fine-grained annotation schema comprising 26 types of form entities, along with a separate coarse form entity set consisting of 5 types. We evaluate the latest MLLMs from the ChatGPT, Gemini, Claude, Qwen, and Kimi series using zero-shot and chain-of-thought prompts under both low and high reasoning setups. Our results reveal limitations in current MLLMs' ability in comprehending Bangla forms, particularly in accurately localizing highly granular form entities. Our dataset and code is available at: https://huggingface.co/datasets/Mausul/bafco

Problem

• Document Comprehension
• Bangla Form Understanding
• Low-resource Languages
• Key Information Extraction
• Document Layout Analysis

Innovation

• Bangla document understanding
• form comprehension benchmark
• fine-grained annotation
• multimodal LLM evaluation
• low-resource language