.png){kind=icon}
.png){kind=icon}
Return
Enhancing Text-to-SQL generation with language sequential consistency
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131721.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Text-to-SQL is an important task that aims to convert unstructured natural language questions into Structured Query Language (SQL). The significant structural gap between questions and SQL increases the difficulty of conversion for the model. Sequential consistency of language refers to the fact that, despite differences in grammar and expression, the order of information conveyed tends to be similar. We argue that this consistency can help build the connection between questions and SQL, and alleviate the challenges posed by the structural gap to the model. This research proposes a Text-to-SQL framework (CT2S) that integrates sequential consistency prompts to generate SQL. Specifically, CT2S first uses a selector to select the database schema items that are most relevant to the question, and then utilizes specially constructed semi-structured prompt data to enhance the performance of SQL generation. The prompt data is constructed based on the sequential consistency of language, considering the order of schema items in both the questions and SQL. It links the questions, database schema, and SQL to alleviate the challenges posed by the structural gap. The relevant experimental results demonstrate the effectiveness of the proposed framework, which achieves over 2 % performance improvement on the Spider dataset and exhibits notable scalability.