SG-UniBuc-NLP at SemEval-2026 Task 6: Multi-Head RoBERTa with Chunking for Long-Context Evasion Detection

TL;DR

The paper presents a multi-head RoBERTa-based system with chunking and ensembling for detecting political question evasions in long interview responses, achieving competitive results.

Contribution

It introduces a chunking strategy with Max-Pooling and multi-task learning for long-context classification in political interview analysis.

Findings

Achieved Macro-F1 of 0.80 on coarse-grained clarity detection.

Achieved Macro-F1 of 0.51 on fine-grained evasion strategy detection.

Ranked 11th in both subtasks at SemEval-2026.

Abstract

We describe our system for SemEval-2026 Task 6 (CLARITY: Unmasking Political Question Evasions), which classifies English political interview responses by coarse-grained clarity (3-way) and fine-grained evasion strategy (9-way). Since responses frequently exceed the 512-token limit of standard Transformer encoders, we apply an overlapping sliding-window chunking strategy with element-wise Max-Pooling aggregation over chunk representations. A shared RoBERTa-large encoder supplies two task-specific heads trained jointly via a multi-task objective, with inference-time ensembling over 7-fold stratified cross-validation. Our system achieves a Macro-F1 of 0.80 on Subtask 1 and 0.51 on Subtask 2, ranking 11th in both subtasks.