Researchers propose a data transformation algorithm to automatically generate entity-relationship (ER) models from natural language utterances, a task known as NL2ERM. Traditional rule-based approaches have limitations in generalizing to varied linguistic descriptions of the same requirement. The proposed algorithm leverages the high similarity between NL2ERM and the text-to-SQL task, transforming data from the popular text-to-SQL Spider dataset into NL2ERM data. The algorithm was applied to two state-of-the-art information extraction models and showed high performance, outperforming existing baselines.
Publication date: 22 Dec 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2312.13694
