ImaginE: An Imagination-Based Automatic Evaluation Metric for Natural Language Generation
Wanrong Zhu, Xin Eric Wang, An Yan, Miguel Patricio Eckstein, William Yang Wang · 2023
Automatic evaluations for natural language generation (NLG) conventionally rely on tokenlevel or embedding-level comparisons with the text references.This is different from human language processing, for which visual imagination often improves comprehension.In this work, we propose IMAGINE, an imaginationbased automatic evaluation metric for natural language generation.With the help of Sta-bleDiffusion (Rombach et al., 2022), a stateof-the-art text-to-image generator, we automatically generate an image as the embodied imagination for the text snippet and compute the imagination similarity using contextual embeddings.Experiments spanning several text generation tasks demonstrate that adding machinegenerated images with our IMAGINE displays great potential in introducing multi-modal information into NLG evaluation, and improves existing automatic metrics' correlations with human similarity judgments in both referencebased and reference-free evaluation scenarios.