Struct CorefConfig 

pub struct CorefConfig {
    pub max_pronoun_lookback: usize,
    pub fuzzy_matching: bool,
    pub include_singletons: bool,
    pub use_name_gazetteer: bool,
    pub acronym_matching: bool,
    pub relaxed_head_match: bool,
    pub proper_containment: bool,
    pub precise_constructs: bool,
    pub strict_head_match: bool,
    pub proper_head_word_match: bool,
}

Configuration for SimpleCorefResolver.

All boolean sieve toggles default to true. Disable individual sieves to measure their contribution or to reduce false merges on a particular dataset.

Fields

max_pronoun_lookback: usize

Maximum number of preceding entities to search when resolving a pronoun (sieve 1).

The actual lookback window is max_pronoun_lookback * 10 entities, a rough approximation of sentence distance by entity count.

Default: 3 (i.e., 30 entities).

fuzzy_matching: bool

Enable substring/fuzzy name matching (sieve 6).

When enabled, two mentions match if one is a substring of the other (>= 3 chars) or if they share a last word and entity type.

Default: true.

include_singletons: bool

Include singletons (entities with no coreferent) in output chains.

Default: true.

use_name_gazetteer: bool

Use a built-in name-to-gender gazetteer for pronoun resolution (sieve 1).

When enabled, common English names (e.g., “Alice” -> Feminine, “Bob” -> Masculine) are used to infer gender for proper nouns, improving pronoun resolution accuracy. Disable for non-English text or when names don’t follow English gender conventions.

Default: true.

acronym_matching: bool

Enable acronym matching (sieve 3).

Checks if one mention is a single-token acronym whose letters match the initial letters of the other mention’s words. Requires same entity type. Example: “IBM” matches “International Business Machines”.

Default: true.

relaxed_head_match: bool

Enable relaxed head match (sieve 4).

Two mentions corefer if their head words (last word) match and they share the same entity type. Only applies when both mentions have 2+ words. Example: “President Obama” ~ “Barack Obama” via shared head “Obama”.

Default: true.

proper_containment: bool

Enable proper noun containment (sieve 8).

One mention’s original text must be a proper subsequence of the other, matching at word boundaries, and both must share the same entity type. Unlike sieve 9 (fuzzy/substring), this operates on original text (not canonical lowercased form) and requires complete word-boundary alignment. Example: “Obama” is contained in “Barack Obama”.

Default: true.

precise_constructs: bool

Enable precise constructs sieve (sieve 3).

Detects appositive constructions where two entity mentions are immediately adjacent (gap <= 2 characters, allowing for “, “) with compatible entity types. E.g., “Barack Obama, the president” – merges the two NPs.

Default: true.

strict_head_match: bool

Enable strict head matching (sieve 5).

Two mentions match if they share the same head word (last word), have compatible entity types and gender, and do not violate the i-within-i constraint (one mention’s span must not contain the other’s span).

Default: true.

proper_head_word_match: bool

Enable proper head word matching (sieve 6).

For named entity mentions: the head word (last word) of one mention appears as any word in the other mention, with same entity type required. E.g., “President Obama” matches “Barack Hussein Obama” (head “Obama” found in both).

Default: true.

Trait Implementations

impl Clone for CorefConfig

fn clone(&self) -> CorefConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CorefConfig

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

Formats the value using the given formatter.

impl Default for CorefConfig

fn default() -> CorefConfig

Returns the “default value” for a type.