CORDIAL: Can Multimodal Large Language Models Effectively Understand Coherence Relationships?

Multimodal Large Language Models (MLLMs) are renowned for their superior instructionfollowing and reasoning capabilities across diverse problem domains. However, existing benchmarks primarily focus on assessing factual and logical correctness in downstream tasks, with limited emphasis on evaluating MLLMs' ability to interpret pragmatic cues and intermodal relationships. To address this gap, we assess the competency of MLLMs in performing Multimodal Discourse Analysis (MDA) using Coherence Relations. Our benchmark, CORDIAL, encompasses a broad spectrum of Coherence Relations across 3 different discourse domains at varying levels of granularity. Through our experiments on 10+ MLLMs employing different prompting strategies, we show that even top models like Gemini 1.5 Pro and GPT-4o fail to match the performance of simple classifier-based baselines. This study emphasizes the need to move beyond similaritybased metrics and adopt a discourse-driven framework for evaluating MLLMs, providing a more nuanced assessment of their capabilities. The benchmark and code are available at: ht tps://aashish2000.github.io/CORDIAL/. Dataset Examples DisREL Part of my pile of branches after #HurricaneIrma -still no power in #Orlando Floridians rescue stranded manatees as Irma sucks water from shores Coherence Relation: Similar Coherence Relation: Complementary Tweet Subtitles Fresh never frozen jumbo wings tossed in a housemade buffalo sauce. Yum! Freshly picked off my allotment today, well chuffed. (strawberry) Cartel leader whose arrest sparked killings is sentenced to prison in Dallas court Amazon Prime delivers anything these days! (delivering a cat) Eiffel Tower shuts down as snow, freezing rain pummel France