Information-Theoretic Text Hallucination Reduction for Video-grounded Dialogue

TL;DR

This paper introduces an information-theoretic framework to reduce text hallucination in video-grounded dialogue systems, improving answer accuracy and interpretability by discouraging indiscriminate copying from input texts.

Contribution

It proposes the Text Hallucination Mitigating (THAM) framework with a novel regularization loss based on information theory to address text hallucination in VGD systems.

Findings

Improved performance on AVSD@DSTC7 and AVSD@DSTC8 benchmarks.

Enhanced interpretability of dialogue systems.

Effective reduction of text hallucination in VGD tasks.

Abstract

Video-grounded Dialogue (VGD) aims to decode an answer sentence to a question regarding a given video and dialogue context. Despite the recent success of multi-modal reasoning to generate answer sentences, existing dialogue systems still suffer from a text hallucination problem, which denotes indiscriminate text-copying from input texts without an understanding of the question. This is due to learning spurious correlations from the fact that answer sentences in the dataset usually include the words of input texts, thus the VGD system excessively relies on copying words from input texts by hoping those words to overlap with ground-truth texts. Hence, we design Text Hallucination Mitigating (THAM) framework, which incorporates Text Hallucination Regularization (THR) loss derived from the proposed information-theoretic text hallucination measurement approach. Applying THAM with current dialogue systems validates the effectiveness on VGD benchmarks (i.e., AVSD@DSTC7 and AVSD@DSTC8) and shows enhanced interpretability.