Module cross_context_eval 

Cross-Context Coreference Evaluation Harness.

Evaluation framework for xCoRe-style cross-context coreference resolution, supporting both long-document and cross-document benchmarks.

Supported Benchmarks

This harness is intended for ECB+, SciCo, LitBank, BookCoref, and similar long-document / cross-document benchmarks.

Usage

use anno_eval::eval::cross_context_eval::{CrossContextBenchmark, evaluate_benchmark};
use anno::metrics::cluster_encoder::{HeuristicClusterEncoder, CosineMergeScorer};

let encoder = HeuristicClusterEncoder::new(64);
let scorer = CosineMergeScorer::new();

let results = evaluate_benchmark(
    CrossContextBenchmark::ECBPlus,
    &encoder,
    &scorer,
    None, // Use default config
)?;

println!("CoNLL F1: {:.1}", results.conll_f1 * 100.0);

References

• Martinelli et al. (2025): “xCoRe: Cross-context Coreference Resolution”
• Cybulska & Vossen (2014): “ECB+ Event Coreference Bank”
• Cattan et al. (2021): “SciCo Hierarchical Cross-Document Coreference”
• Bamman et al. (2020): “LitBank”
• Martinelli et al. (2025): “BOOKCOREF: Coreference Resolution at Book Scale”
• Guo et al. (2023): “Animal Farm annotation”

Structs

CrossContextEvalConfig

Configuration for cross-context evaluation.

CrossContextEvalResults

Results from cross-context evaluation.

DocumentResults

Per-document results for long-document evaluation.

LongDocument

A long document with gold annotations (for long-document evaluation).

Topic

A topic containing multiple documents (for cross-document evaluation).

TopicResults

Per-topic results for cross-document evaluation.

Enums

CrossContextBenchmark

Cross-context coreference benchmarks.

StepwiseAnalysis

Stepwise error analysis configuration.

Functions

evaluate_cross_document

Evaluate cross-document coreference on a set of topics.

evaluate_long_document

Evaluate long-document coreference.

stepwise_error_analysis

Run stepwise error analysis as in xCoRe Table 5.