SemEval-winning code-authorship work published at ACL 2026

In January, our young residents Agzam Shamsadinov and Yeraly Gainulla won SemEval Task 13C, part of the international series of NLP workshops and competitions. The task classifies code authorship into four classes: Human, AI, Hybrid, and Adversarial AI.

Agzam and Yeraly built a multimodal ensemble that analyzes code from three perspectives: code semantics through UniXcoder with Multiple Instance Learning for long files; textual patterns through MazgaBERT, which identifies characteristic templates; and statistical anomalies through XGBoost using hand-crafted style and code-structure features.

Their solution paper was published in the proceedings of the 20th International Workshop on Semantic Evaluation at ACL 2026, the 64th Annual Meeting of the Association for Computational Linguistics. The conference begins tomorrow in San Diego and is one of NLP's leading annual events. Congratulations to our students on this major achievement!

Paper: https://aclanthology.org/2026.semeval-1.137/