MindFlayer at SemEval-2026 Task 8:DUALRAG:Answerability-Aware Generation for Multi-Turn RAG Conversations
Year: 2026
Type: conference-paper
Abstract: We present DUALRAG, the Mind Flayer team's system for SemEval-2026 Task 8 (MTRAGEval), Subtask B (Generation with Reference Passages).Our system addresses generation of faithful responses in multi-turn retrieval augmented generation (RAG) conversations through an answerability sensitive prompting strategy based on meta-llama/ llama-4-scout-17b-16e-instruct using the Groq inference API.DUALRAG introduces a dual-prompt architecture that makes stern refusal questions impossible to answer, and domain-guided generated questions become answerable ones with a post-processing safety net to rectify misclassifications.The harmonic mean of our system achieves 0.7492 (RB agg =0.6024, RL F =0.8782, RB llm =0.8297) and it ranked 8th out of 26 teams in the competition.It also surpassed the top baseline (GPT-OSS-120B, 0.639).We also achieved 100% refusal accuracy on unanswerable questions.
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: Rajshahi University
OpenAlex ID: W7166881052