New Benchmark IRC-Bench Targets Implicit Entity Recognition in Reminiscence Transcripts
Researchers have introduced IRC-Bench, the Implicit Reminiscence Context Benchmark, designed to evaluate implicit entity recognition in first-person reminiscence narratives. The benchmark addresses the computational challenge of identifying people, places, and events that are referenced indirectly through contextual cues rather than explicit names. IRC-Bench comprises 25,136 samples constructed from 12,337 Wikidata-linked entities across 1,994 transcripts spanning 11 thematic domains. The benchmark targets non-locality, where entity-identifying cues are distributed across multiple, non-contiguous clauses, distinguishing it from named entity recognition, entity linking, or coreference resolution. This work is published on arXiv with identifier 2605.06142.
Key facts
- IRC-Bench is a new benchmark for implicit entity recognition in reminiscence transcripts.
- It comprises 25,136 samples from 12,337 Wikidata-linked entities.
- The benchmark uses 1,994 transcripts across 11 thematic domains.
- It targets non-locality: cues are distributed across non-contiguous clauses.
- The work is published on arXiv with identifier 2605.06142.
- Implicit references are common in therapeutic, archival, and social settings.
- The benchmark differs from NER, entity linking, and coreference resolution.
- The problem involves inferring entities from dispersed narrative evidence.
Entities
Institutions
- arXiv
- Wikidata