Symbal: Detecting Systematic Misalignments in Model-Generated Captions

2026-07-16

AI Summary

This work addresses the tendency of multimodal large language models (MLLMs) to systematically misalign generated image captions with specific visual features. It formalizes, for the first time, the task of systematic misalignment detection and introduces Symbal, a two-stage method that requires no access to the MLLM’s internal architecture. Symbal leverages off-the-shelf vision-language foundation models to automatically identify and summarize recurring misalignment patterns. The authors also construct SymbalBench, a benchmark comprising 1.7 million image-text pairs, to evaluate this task. Experiments demonstrate that Symbal achieves 63.8% accuracy on SymbalBench—nearly four times higher than the strongest baseline—and successfully audits four prominent MLLMs as well as real-world datasets.

Abstract

Multimodal large language models (MLLMs) often introduce errors when generating image captions, resulting in misaligned image-text pairs. Our work focuses on a class of captioning errors that we refer to as systematic misalignments, where a recurring error in MLLM-generated captions is closely associated with the presence of a specific visual feature in the paired image. Given a vision-language dataset with MLLM-generated captions, our aim in this work is to detect such errors, a task we refer to as systematic misalignment detection. As our first key contribution, we present Symbal, which utilizes a structured, dual-stage setup with off-the-shelf foundation models to identify systematic misalignments and summarize results in natural language. As our second key contribution, we introduce SymbalBench, a benchmark designed to evaluate automated methods on our proposed task. SymbalBench consists of 1.7 million image-text pairs from two domains (natural and medical images), organized into 420 vision-language datasets with annotated systematic misalignments. Symbal exhibits strong performance on this benchmark, correctly identifying systematic misalignments in 63.8% of datasets, a nearly 4x improvement over the closest baseline. We supplement our evaluations on SymbalBench with real-world evaluations, showing that (1) Symbal can accurately surface systematic misalignments in captions generated by four MLLMs and (2) Symbal is a powerful tool for auditing off-the-shelf image-caption datasets. Ultimately, our novel task, method, and benchmark can aid users with auditing MLLM-generated captions and identifying critical errors, without requiring access to the underlying MLLM. Code is available at https://github.com/Stanford-AIMI/Symbal.

Problem

• systematic misalignment
• multimodal large language models
• image captioning
• vision-language alignment
• caption error detection

Innovation

• systematic misalignment detection
• multimodal large language models
• image caption auditing
• Symbal
• SymbalBench