Multi-judge system detects errors in vision-language text across languages
Vroom-Vroom at SHROOM-Visions: A Multi-Judge Committee for Detecting Hallucinated Spans in Vision-Language Outputs
Computation and LanguageArtificial Intelligence
Summary
Sometimes computers that describe images make mistakes by adding details that aren’t really there, called hallucinations. The authors created a system where several different AI models each point out suspicious parts in the descriptions, then combine their votes to decide which parts are likely errors. Their method worked very well in several languages, even beating other submissions in a shared challenge. They also found that when these models disagreed, humans often disagreed too.
What this means in practice
- •For natural language processing teams: Improve quality control by detecting false or misleading text spans in outputs combining image and text data across multiple languages.
- •For multilingual content moderation teams: Identify potentially inaccurate or hallucinated descriptions in automated image captioning to assist in moderation of user-generated content.
Authors
Toqeer Ehsan, Nico Penttilä, Richard Schmidt, Arash Hajikhani, Victoria Palacin
Abstract
This paper describes our submission to the SHROOM-Visions shared task on detecting and classifying hallucinated character spans in vision-language model outputs across four languages. We employ several fine-tuned vision-language models as independent annotators and combine their span predictions through character-level majority voting, and additionally explore activation probes. The approach ranks first in three of four languages and places on the podium in every language and metric. Our analysis indicates that disagreement among diverse models tracks disagreement among human annotators.