SCOPE and SCION: New Benchmark and Pipeline for Schema Induction from Text
A new benchmark called SCOPE (Schema Construction and Ontology-induction Pipeline Evaluation) has been developed by researchers to facilitate corpus-to-schema induction from unprocessed text. Additionally, they introduced SCION (Schema Construction and Induction with Ontology Normalization), which serves as an auditable reference pipeline. SCOPE comprises 24 publicly available information extraction sources, consisting of 15 for relation extraction and 9 for event extraction, all normalized into gold schema graphs. It focuses on event types and the roles of arguments within events, with inter-event connections documented separately. SCION generates candidate spaces from training text and implements naming, merging, filtering, validation, and conservative fusion while adhering to strict JSON guidelines. This research is detailed in arXiv preprint 2607.21610.
Key facts
- SCOPE is a benchmark for schema induction from raw text.
- SCION is an auditable reference pipeline for schema construction.
- The benchmark uses 24 public information extraction sources.
- Sources include 15 relation extraction and 9 event extraction datasets.
- Gold schema graphs are normalized for evaluation.
- The core target covers event types and within-event argument roles.
- Inter-event links are reported separately.
- SCION uses candidate spaces from train text and strict JSON constraints.
Entities
Institutions
- arXiv