Struct CorefChainStats 

pub struct CorefChainStats {
    pub long_chain_count: usize,
    pub short_chain_count: usize,
    pub singleton_count: usize,
    pub long_chain_f1: f64,
    pub short_chain_f1: f64,
    pub singleton_f1: f64,
}

Statistics for stratified coreference evaluation.

Why Chain Length Matters: A Narrative

Imagine analyzing “Pride and Prejudice”:

┌─────────────────────────────────────────────────────────────────────────┐
│                    COREFERENCE IN A NOVEL                               │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                         │
│  LONG CHAINS (>10 mentions) - THE PROTAGONISTS                          │
│  ─────────────────────────────────────────────                          │
│                                                                         │
│  "Elizabeth" ─── "she" ─── "Lizzy" ─── "her" ─── "Miss Bennet" ───...  │
│       │            │          │          │            │                 │
│       └────────────┴──────────┴──────────┴────────────┘                 │
│                         800+ mentions                                   │
│                                                                         │
│  Getting these right = understanding the PLOT.                          │
│  Who did what to whom? What's Elizabeth's arc?                          │
│                                                                         │
│  SHORT CHAINS (2-10 mentions) - SECONDARY CHARACTERS                    │
│  ───────────────────────────────────────────────────                    │
│                                                                         │
│  "Mr. Collins" ─── "he" ─── "the clergyman"                             │
│       │              │             │                                    │
│       └──────────────┴─────────────┘                                    │
│                  15 mentions                                            │
│                                                                         │
│  Important for context, but errors here are less catastrophic.          │
│                                                                         │
│  SINGLETONS (1 mention) - BACKGROUND                                    │
│  ───────────────────────────────────────                                │
│                                                                         │
│  "a tall man" ─── (no other mentions)                                   │
│  "the servant" ─── (no other mentions)                                  │
│                                                                         │
│  These are closer to entity detection than coreference.                  │
│  Including them in CoNLL F1 can change the interpretation of the score.  │
└─────────────────────────────────────────────────────────────────────────┘

The Problem with Averaged Metrics

Model Performance:

A single averaged metric can hide systematic differences across chain sizes.
Prefer reporting stratified metrics by chain length (and be explicit about
whether singletons are included).

Research Context (arXiv:2401.00238)

“How to Evaluate Coreference in Literary Texts?”

• A single CoNLL F1 score is “uninformative, or even misleading.”
• Stratify by chain length for interpretable results.

Example

use anno_eval::eval::CorefChainStats;

let stats = CorefChainStats {
    long_chain_count: 3,      // Main characters
    short_chain_count: 15,    // Secondary
    singleton_count: 42,      // Isolated
    long_chain_f1: 0.92,      // Good on main characters
    short_chain_f1: 0.71,     // Weaker on secondary
    singleton_f1: 0.45,       // Poor on singletons
};

// Report metrics separately, not averaged
println!("Main characters: {:.1}% F1", stats.long_chain_f1 * 100.0);

Chain-length stratified statistics for coreference evaluation.

Fields

long_chain_count: usize

Number of long chains (>10 mentions).

short_chain_count: usize

Number of short chains (2-10 mentions).

singleton_count: usize

Number of singletons (1 mention).

long_chain_f1: f64

F1 score on long chains only.

short_chain_f1: f64

F1 score on short chains only.

singleton_f1: f64

F1 score on singletons (if evaluated).

Implementations

impl CorefChainStats

pub fn total_chains(&self) -> usize

Total chain count.

pub fn weighted_f1(&self) -> f64

Weighted F1 (by chain count).

Note: this is not CoNLL F1; it is a diagnostic aggregation over chain strata.

Trait Implementations

impl Clone for CorefChainStats

fn clone(&self) -> CorefChainStats

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for CorefChainStats

impl Debug for CorefChainStats

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Default for CorefChainStats

fn default() -> CorefChainStats

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for CorefChainStats

fn deserialize<__D>( __deserializer: __D, ) -> Result<CorefChainStats, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for CorefChainStats

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.