SG-UniBuc-NLP at SemEval-2026 Task 6: Multi-Head RoBERTa with Chunking for Long-Context Evasion Detection
This work addresses the practical challenge of long-context classification in political discourse analysis, but the results are incremental given the mid-tier ranking.
The authors developed a multi-head RoBERTa model with overlapping sliding-window chunking to classify political interview responses by evasion clarity and strategy, achieving Macro-F1 scores of 0.80 (3-way) and 0.51 (9-way), ranking 11th in both subtasks at SemEval-2026 Task 6.
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.