FMI_SU_Yotkova_Kastreva at SemEval-2026 Task 13: Lightweight Detection of LLM-Generated Code via Stylometric Signals

TL;DR

This paper presents a lightweight, efficient method for detecting machine-generated code across languages using stylometric features, parsing, and simple classifiers, achieving fast inference with minimal resources.

Contribution

It introduces a novel, resource-efficient approach combining ratio-based features, parsing, and heuristic rules for cross-language code detection, outperforming large models in speed.

Findings

Achieves near-instant inference time with CPU-only training.

Uses ratio-based features less sensitive to code snippet length.

Employs a combination of parsing, classifiers, and heuristics for detection.

Abstract

SemEval-2026 Task 13 investigates machine-generated code detection across multiple programming languages and application scenarios, asking participating systems to generalize to unseen languages and domains. This paper describes our participation in Subtask A (binary classification) and explores both pretrained code encoders and lightweight feature-based methods. We design ratio-based features that are less sensitive to snippet length. To support the extraction of descriptiveness-related signals, we use parsing engines and a programming-language classifier. Additionally, we train a separate code-vs-text line classifier to identify raw natural language segments embedded within samples. We combine a shallow decision tree with heuristic rules derived from data analysis to produce the final predictions. Our approach is computationally efficient, requires only CPU resources for training, and achieves near-instant inference time, offering a lightweight alternative to large pretrained models.