Target-Augmented Shared Fusion-based Multimodal Sarcasm Explanation Generation

TL;DR

This paper introduces TURBO, a novel multimodal sarcasm explanation model that effectively leverages target information and inter-modality relationships to generate more accurate and nuanced explanations, outperforming existing methods.

Contribution

The paper proposes TURBO, a shared fusion-based model that incorporates sarcasm targets into multimodal explanation generation, advancing the state-of-the-art in sarcasm understanding.

Findings

TURBO outperforms baselines with a +3.3% accuracy margin.

LLMs generate explanations but often miss sarcasm nuances.

Human evaluation favors TURBO's explanations over others.

Abstract

Sarcasm is a linguistic phenomenon that intends to ridicule a target (e.g., entity, event, or person) in an inherent way. Multimodal Sarcasm Explanation (MuSE) aims at revealing the intended irony in a sarcastic post using a natural language explanation. Though important, existing systems overlooked the significance of the target of sarcasm in generating explanations. In this paper, we propose a Target-aUgmented shaRed fusion-Based sarcasm explanatiOn model, aka. TURBO. We design a novel shared-fusion mechanism to leverage the inter-modality relationships between an image and its caption. TURBO assumes the target of the sarcasm and guides the multimodal shared fusion mechanism in learning intricacies of the intended irony for explanations. We evaluate our proposed TURBO model on the MORE+ dataset. Comparison against multiple baselines and state-of-the-art models signifies the performance improvement of TURBO by an average margin of $+3.3\%$. Moreover, we explore LLMs in zero and one-shot settings for our task and observe that LLM-generated explanation, though remarkable, often fails to capture the critical nuances of the sarcasm. Furthermore, we supplement our study with extensive human evaluation on TURBO's generated explanations and find them out to be comparatively better than other systems.