AI-Assisted Generation of SQL Comprehension Questions - Tool Demonstration

Sep 17, 2026

Uploaded by Eliana

Students often struggle to fully comprehend the SQL queries they write, with gaps in understanding persisting until later stages when misconceptions about relational concepts, semantics, formulation, or execution are hard to remediate. GenAI tools can mask conceptual weaknesses further. The paper demonstrates an open extension that automatically generates multiple-choice SQL comprehension questions from queries and integrates with CodeRunner for rapid, scalable, personalised assessment, enabling timely feedback and reduced instructor workload.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Abstract

1. Introduction

2. Related Work

AI-Assisted Generation of SQL Comprehension Questions⋆ Martin Goodfellow1 , Alasdair Lambert1 , Andrew Fagan1 and Robbie Booth 1 1 University of Strathclyde, Glasgow, Scotland, UK Abstract Students often struggle to fully comprehend the SQL queries they write. These gaps in understanding may remain hidden until later in their studies. At that stage, misconceptions about relational concepts, query semantics, formulation, or execution behaviour are difficult to remediate, particularly when students encounter increasingly complex queries. The increasing use of GenAI tools (e.g., GitHub Copilot) for query construction can further mask conceptual weaknesses. A common instructional strategy for addressing this issue is the use of SQL comprehension questions that require students to predict query results, explain query logic, or identify semantic errors. Such questions help surface misconceptions and may also support academic integrity by requiring students to demonstrate conceptual understanding beyond producing a working query. However, creating scalable sets of high-quality questions is labour-intensive. This tool demonstration presents an openly available extension of a prior GenAI-based system that automatically generates Java code comprehension questions, adapted to the domain of SQL. The tool automatically derives multiple-choice SQL comprehension questions directly from queries and integrates with the CodeRunner automated assessment platform to support rapid generation and deployment of scalable, personalised assessments. It is designed to support timely feedback, reduce instructor workload, and support conceptual understanding of SQL. Keywords SQL, comprehension, database, data systems, GPT, GPT-4o mini, generative artificial intelligence, GenAI, automated marking, CodeRunner 1. Introduction Students in introductory database and data management courses frequently produce SQL queries that return the expected output, yet may not fully understand how those queries operate [1] . This pattern aligns with findings from introductory programming contexts, where students can write code that appears correct despite holding incomplete or inaccurate mental models of program behaviour [2] . Recent work in SQL education similarly reports that novices often struggle with core relational conceptsand query semantics, leading to persistent misconceptions about how SQL evaluates queries [1, 3, 4] . SQL’s declarative nature can make these misconceptions difficult to detect, as intermediate results are not visible and incorrect reasoning can produce apparently correct outputs on simple tasks. Comprehension-oriented tasks have been shown to provide insight into how students interpret and reason about program behaviour [5] . When applied to SQL, comprehension questions offer instructorsa way to probe students’ understanding of query intent and expected behaviour, helping to surface conceptual gaps that may not be evident from the final output alone. However, generating tailored comprehension questions for each student submission is time-consuming and difficult to scale for large cohorts. Pre-prepared question banks may be perceived as less relevant to students’ own work and are susceptible to reuse across cohorts [6, 7] . Automated feedback tools, including those used for large-scale practice and assessment, have shown promise for supporting scalable formative feedback [8, 9, 10], suggesting potential for similar approaches in SQL. DataEd’26: 5th International Workshop on Data Systems Education ⋆ Published in the Proceedings of the Workshops ofthe EDBT/ICDT 2026 Joint Conference (March 24-27, 2026), Tampere, Finland $ [martin.h.goodfellow@strath.ac.uk](martin.h.goodfellow@strath.ac.uk) (M. Goodfellow); [alasdair.lambert@strath.ac.uk](alasdair.lambert@strath.ac.uk) (A. Lambert); [andrew.fagan@strath.ac.uk](andrew.fagan@strath.ac.uk) (A. Fagan); [robbie@tech.scot.ac.uk](robbie@tech.scot.ac.uk) (R. Booth) 􀀚 0000-0003

FAQs

Why do students often misunderstand SQL queries even when outputs are correct?

Students may execute queries successfully while holding incomplete or inaccurate mental models of relational concepts and query semantics. Because SQL is declarative and intermediate results are not visible, incorrect reasoning can still produce apparently correct outputs on simpler tasks.

How do SQL comprehension questions help instructors address misconceptions?

Comprehension questions ask students to predict results, explain query logic, or identify semantic errors. They surface conceptual gaps and can also support academic integrity by requiring evidence of understanding beyond producing a working query.

What does the presented tool do and how is it assessed at scale?

The tool automatically derives multiple-choice SQL comprehension questions directly from submitted queries, based on SQL-specific analysis. It integrates with the CodeRunner automated assessment platform to generate and deploy scalable, personalised assessments that provide timely feedback while reducing instructor workload.

Sep 17, 2026

Uploaded by Eliana

AI-Assisted Generation of SQL Comprehension Questions - Tool Demonstration_Page 1

Open in WPS to view all 10 pages