Data-QuestEval: A Referenceless Metric for Data-to-Text Semantic Evaluation
This work addresses the problem of evaluating data-to-text generation systems for researchers and practitioners, providing a reference-less and multimodal metric, though it is incremental as it builds on an existing metric.
The paper tackled the challenge of adapting QuestEval, a reference-less metric for text-to-text tasks, to data-to-text tasks by developing a method to create synthetic multimodal corpora for training multimodal components, resulting in a metric that achieves state-of-the-art correlations with human judgment on WebNLG and WikiBio benchmarks.
QuestEval is a reference-less metric used in text-to-text tasks, that compares the generated summaries directly to the source text, by automatically asking and answering questions. Its adaptation to Data-to-Text tasks is not straightforward, as it requires multimodal Question Generation and Answering systems on the considered tasks, which are seldom available. To this purpose, we propose a method to build synthetic multimodal corpora enabling to train multimodal components for a data-QuestEval metric. The resulting metric is reference-less and multimodal; it obtains state-of-the-art correlations with human judgment on the WebNLG and WikiBio benchmarks. We make data-QuestEval's code and models available for reproducibility purpose, as part of the QuestEval project.