Module coref_loader 

Coreference dataset loading and parsing.

Provides specialized loaders for coreference datasets that return CorefDocument structures rather than NER annotations.

Supported Datasets

Dataset	Format	Size	Features
GAP	TSV	8,908 pairs	Gender-balanced pronoun resolution
PreCo	JSON	~38k docs	Large-scale, includes singletons
Synthetic	Generated	Configurable	For testing metrics

Example

use anno_eval::eval::coref_loader::{CorefLoader, synthetic_coref_dataset};

// Load GAP development set (requires eval feature for download)
let loader = CorefLoader::new().unwrap();
let docs = loader.load_gap().unwrap();

// Or generate synthetic data for testing
let synthetic = synthetic_coref_dataset(10);

Structs

CorefLoader

Loader for coreference datasets.

GapExample

A single GAP example (pronoun-name pair).

PreCoDocument

A PreCo document with coreference annotations.

Functions

adversarial_coref_examples

Generate adversarial coreference examples.

domain_specific_coref_dataset

Generate domain-specific synthetic coreference documents.

parse_bookcoref_json

Parse BookCoref JSON/JSONL format.

parse_corefud_conllu

Parse CorefUD CoNLL-U format into coreference documents.

parse_ecb_plus_coref

Parse ECB+ coreference from any supported format.

parse_ecb_plus_zip

Parse ECB+ from the real XML zip archive (ECB+.zip).

parse_gap_tsv

Parse GAP TSV format.

parse_preco_json

Parse PreCo JSON format. Parse PreCo JSON format (public for use by DatasetLoader).

synthetic_coref_dataset

Generate synthetic coreference documents for testing.