SAM-NER: Semantic Archetype Mediation for Zero-Shot NER
A new framework called SAM-NER (Semantic Archetype Mediation for Zero-Shot Named Entity Recognition) has been proposed to address the brittleness of zero-shot NER under domain and schema shifts. The approach introduces a three-stage process: Entity Discovery via cooperative extraction and consensus-based denoising, Abstract Mediation that projects entities into a compact set of universal semantic archetypes, and Semantic Calibration. The method aims to stabilize cross-domain transfer by using an intermediate, domain-invariant archetype space, reducing systematic semantic drift when target schemas are novel or overlapping. The research is published on arXiv under identifier 2605.03706.
Key facts
- SAM-NER is a three-stage framework for zero-shot NER.
- It uses Semantic Archetype Mediation to stabilize cross-domain transfer.
- Entity Discovery uses cooperative extraction and consensus-based denoising.
- Abstract Mediation projects entities into universal semantic archetypes.
- Semantic Calibration is the third stage.
- The approach addresses domain and schema shifts.
- It reduces semantic drift from direct mapping to fine-grained labels.
- The paper is available on arXiv with ID 2605.03706.
Entities
Institutions
- arXiv