We present our system for SemEval-2026 Task 12 on abductive event reasoning. Initial experiments with direct fine-tuning of large language models suffered from severe overfitting due to limited training data, while smaller models failed under context-length constraints, leading to random guessing under the strict Exact Match evaluation metric. To address these challenges, we propose a two-stage offline Retrieval-Augmented Generation (RAG) pipeline that separates semantic evidence retrieval from multi-label classification. We employ a dense retriever (all-MiniLM-L6-v2) to extract the single most relevant sentence (top-k=1) and feed it into a partially frozen DeBERTa-v3-Large classifier trained with BCEWithLogitsLoss. Freezing the lower 12 layers effectively mitigates overfitting while preserving pre-trained semantic knowledge. Our approach eliminates long-context truncation issues, reduces hallucination, and achieves a final Exact Match accuracy of 0.72 on the official test set.

Annotating entity mentions and linking them to a knowledge resource are essential tasks in many domains. It disambiguates mentions, introduces cross-document coreferences, and the resources contribute extra information, e.g. taxonomic relations. Such tasks benefit from text annotation tools that integrate a search which covers the text, the annotations, as well as the knowledge resource. However, to the best of our knowledge, no current tools integrate knowledge-supported search as well as entity linking support. We address this gap by introducing knowledge-supported search functionality into the INCEpTION text annotation platform. In our approach, cross-document references are created by linking entity mentions to a knowledge base in the form of a structured hierarchical vocabulary. The resulting annotations are then indexed to enable fast and yet complex queries taking into account the text, the annotations, and the vocabulary structure.