ByteBreaker@DravidianLangTech 2026: XLM-RoBERTa Large with Sliding-Window Chunking and Top-K Mean Pooling for Writing Style Classification
Year: 2026
Type: conference-paper
Abstract: Identifying writing styles in long documents is challenging because stylistic signals are distributed unevenly and differ in subtle ways.We describe ByteBreaker, our system for the Prompt Recovery for LLM Shared Task at DravidianLangTech@ACL-2026, which classifies the writing style of LLM-rewritten Telugu documents into nine categories.To handle documents exceeding the 512-token transformer limit, we apply sliding-window chunking (stride = 256) and fine-tune XLM-RoBERTa Large on the style-rewritten text.At inference, Top-K Mean Pooling emphasises the most confident chunks.We train with five random seeds and submit three runs: a weighted ensemble (Run 1), a mean-guided single model (Run 2), and a Top-K-guided single model (Run 3).Run 3 achieves the best macro F1 of 0.3306; Run 1 achieves the best accuracy of 0.3256 with macro F1 of 0.3290.Our code is available on GitHub.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166797707