Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data
This work addresses the reliability issue of multimodal models for users by providing a method to localize hallucinations, though it is incremental as it builds on existing grounding data techniques.
The paper tackles the problem of detecting and localizing hallucinations in multimodal language models by framing it as a sequence labeling task and proposes pre-training with corrupted grounding data to improve sample efficiency, showing that this approach enhances performance when fine-tuning.
Multimodal language models can exhibit hallucinations in their outputs, which limits their reliability. The ability to automatically detect these errors is important for mitigating them, but has been less explored and existing efforts do not localize hallucinations, instead framing this as a classification task. In this work, we first pose multimodal hallucination detection as a sequence labeling task where models must localize hallucinated text spans and present a strong baseline model. Given the high cost of human annotations for this task, we propose an approach to improve the sample efficiency of these models by creating corrupted grounding data, which we use for pre-training. Leveraging phrase grounding data, we generate hallucinations to replace grounded spans and create hallucinated text. Experiments show that pre-training on this data improves sample efficiency when fine-tuning, and that the learning signal from the grounding data plays an important role in these improvements.