<i>A Four-Language Corpus of 11,809 Novice Comment-Code Pairs (Python, Java, C, C++)</i>

A Four-Language Corpus of 11,809 Comment-Code Pairs" submitted to SIGCSE '27.</p><h3 dir="ltr"><b>Overview</b></h3><p dir="ltr">The corpus contains 11,809 cleaned and associate-mapped novice comment-code pairs extracted across four programming languages: Java (7,194), C (2,493), Python (2,092), and C++ (30). The dataset was constructed by parsing source files with Tree-sitter from eight public introductory programming repositories (Menagerie, Rubric Is All You Need, CodeStream, Prutor, IntroClass, DCU, Refactory, and Python Code Plagiarism).<br>All pairs underwent Zhang-style cleaning rules followed by novice-specific filtering (deduplication, instructor boilerplate removal, and code-only fragment removal).</p><p><br></p><h3 dir="ltr"><b>File Format & Schema</b></h3><p dir="ltr">The primary file `AbkadyrDostiyar_final_dataset_11809.jsonl` is formatted as JSON Lines.

Each entry contains:</p><ul><li>`comment_text`: Cleaned text of the inline or block comment</li><li>`code_text`: Associated code block mapped via AST association rules</li><li>`language`: Programming language (`java`, `c`, `python`, `cpp`)</li></ul><p dir="ltr"><i>License: Creative Commons Attribution 4.0 International (CC BY 4.0)</i></p>