Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Whitehead, Spencer, Phillips, Jacob, Hendryx, Sean
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916377053364224
author Whitehead, Spencer
Phillips, Jacob
Hendryx, Sean
author_facet Whitehead, Spencer
Phillips, Jacob
Hendryx, Sean
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2409_00238
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data
Whitehead, Spencer
Phillips, Jacob
Hendryx, Sean
Computation and Language
Computer Vision and Pattern Recognition
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.
title Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data
topic Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2409.00238