MUStReason: A Benchmark for Diagnosing Pragmatic Reasoning in Video-LMs for Multimodal Sarcasm Detection

TL;DR

This paper introduces MUStReason, a benchmark for diagnosing pragmatic reasoning in Video-LMs for multimodal sarcasm detection, highlighting current model limitations and proposing a new framework to improve understanding of implied intentions.

Contribution

It presents MUStReason, a diagnostic benchmark with annotations for multimodal sarcasm detection, and proposes PragCoT, a framework to enhance VideoLMs' focus on implied intentions.

Findings

VideoLMs struggle with sarcasm detection across modalities.

MUStReason effectively evaluates reasoning capabilities of models.

PragCoT improves focus on implied intentions in VideoLMs.

Abstract

Sarcasm is a specific type of irony which involves discerning what is said from what is meant. Detecting sarcasm depends not only on the literal content of an utterance but also on non-verbal cues such as speaker's tonality, facial expressions and conversational context. However, current multimodal models struggle with complex tasks like sarcasm detection, which require identifying relevant cues across modalities and pragmatically reasoning over them to infer the speaker's intention. To explore these limitations in VideoLMs, we introduce MUStReason, a diagnostic benchmark enriched with annotations of modality-specific relevant cues and underlying reasoning steps to identify sarcastic intent. In addition to benchmarking sarcasm classification performance in VideoLMs, using MUStReason we quantitatively and qualitatively evaluate the generated reasoning by disentangling the problem into perception and reasoning, we propose PragCoT, a framework that steers VideoLMs to focus on implied intentions over literal meaning, a property core to detecting sarcasm.