Image Caption Concreteness

Image caption concreteness research focuses on quantifying the level of tangibility and specificity in textual descriptions of images, aiming to improve the quality and reliability of multimodal datasets and large language models (LLMs). Current research utilizes LLMs, such as ChatGPT, to automatically assess concreteness, often correlating strongly with human judgments, and investigates how linguistic features like readability and formality in prompts influence the concreteness of generated captions and the likelihood of model hallucinations. This work is crucial for improving the accuracy and efficiency of multimodal learning, enabling better data curation and ultimately leading to more robust and reliable AI systems.