NYK-MS: A Well-annotated Multi-modal Metaphor and Sarcasm Understanding Benchmark on Cartoon-Caption Dataset

TL;DR

This paper introduces NYK-MS, a comprehensive multi-modal benchmark dataset for understanding metaphor and sarcasm in cartoons and captions, highlighting the challenges faced by current models and the importance of multimodal understanding.

Contribution

The creation of a well-annotated, multi-modal benchmark dataset for metaphor and sarcasm understanding, with extensive experiments revealing limitations of current models and the need for multimodal comprehension.

Findings

LLMs and LMMs perform poorly on classification tasks in zero-shot settings.

Model performance improves with scale on five tasks.

Augmentation and alignment enhance traditional model performance.

Abstract

Metaphor and sarcasm are common figurative expressions in people's communication, especially on the Internet or the memes popular among teenagers. We create a new benchmark named NYK-MS (NewYorKer for Metaphor and Sarcasm), which contains 1,583 samples for metaphor understanding tasks and 1,578 samples for sarcasm understanding tasks. These tasks include whether it contains metaphor/sarcasm, which word or object contains metaphor/sarcasm, what does it satirize and why does it contains metaphor/sarcasm, all of the 7 tasks are well-annotated by at least 3 annotators. We annotate the dataset for several rounds to improve the consistency and quality, and use GUI and GPT-4V to raise our efficiency. Based on the benchmark, we conduct plenty of experiments. In the zero-shot experiments, we show that Large Language Models (LLM) and Large Multi-modal Models (LMM) can't do classification task well, and as the scale increases, the performance on other 5 tasks improves. In the experiments on traditional pre-train models, we show the enhancement with augment and alignment methods, which prove our benchmark is consistent with previous dataset and requires the model to understand both of the two modalities.