Crate english 

English inflection backed by Wiktionary-derived lookup tables with a rule-engine fallback.

Architecture

Every query follows the same two-tier shape: consult the generated PHF tables (generated/*_phf.rs, compiled in via include!) for an attested exception first, and fall back to EnglishCore’s regular rules otherwise. The tables and the rules are NOT independent: the extractor drops any candidate the rules already predict, so the tables hold exactly the attested exceptions and nothing else. Changing a rule in english-core therefore requires regenerating the tables (cargo xtask refresh-data). Two dump-free tests partially guard this: the rule_table_sync test below catches a rule change that makes an existing bare table row redundant, and english-core’s regular_rules_golden test catches one that breaks the fallback for un-tabled regular words. Neither is exhaustive — cargo xtask accuracy (with the dump) is the authoritative check after a rule change.

Sense-numbered keys

Homograph senses and attested variants are published as _<n>-suffixed keys (die_2 -> dice). The underscore is unambiguous: the extractor never admits a lemma containing _, so a trailing _<digits> can only be a sense suffix — digit-bearing words (mp3, F16, F2) can never collide with one. Adjacent-digit keys (die2) are deliberately not parsed: they are indistinguishable from ordinary digit-bearing input.

Key numbers are DETERMINISTIC but NOT immutable. They are assigned by a pure sort of each lemma’s emitted forms (see extractor::assign): the bare key goes to the standard sense with the lexicographically-smallest signature (or is reserved for the rule engine when a regular form is attested), and the rest number upward. Regenerating from a newer Wiktionary dump can therefore renumber a lemma’s _<n> keys if its attested forms change — there is no lockfile or frozen identity. What is stable is the set of forms a lemma exposes and the rule/table layering below.

Lookup semantics (invariants shared by English::noun/English::verb/English::adj/English::adverb)

1. Case-insensitive convenience: Title-case and ALL-CAPS input hit the all-lowercase tables with the casing restored on the value; regular-rule fallbacks get the identical treatment via the internal rule_with_case helper, so HOT -> HOTTER exactly as GOOD -> BEST. Mixed case is never guessed at (McDonald falls through as McDonalds). Known acronym homographs of irregular rows are excluded from the case-folded fallback (US inflects by rule, never through the us row — see ACRONYM_SHADOWED). This is an ergonomic default, not semantic proper-noun detection; callers that care about acronyms, surnames, or house style should normalize casing before/after calling this crate.
2. Base-lemma agreement: when a _<n> suffix strips (the word or its base is a table key), EVERY code path treats the input as that base lemma — lookups fall back get_*(word) then get_*(base) so child_2 inflects exactly like child instead of mixing table singulars with rule plurals.
3. Opaqueness: input that resolves to no key inflects by rule on the whole string, unchanged (cat_2 -> cat_2s).

Structs

English

Entry point for English inflection and morphology.

Enums

Animacy

Animacy marker reserved for higher-level phrase APIs.

Case

Pronoun case.

Degree

Degree for adjectives and gradable adverbs.

Form

Verb form requested from the conjugator.

Gender

Pronoun gender for third-person singular forms.

Number

Grammatical number.

Person

Grammatical person.

Tense

Morphological tense realized by finite/simple verb forms.

Functions

count

Inflect a noun according to a numeric count.

count_with_number

Inflect a noun according to a numeric count and keep the number in the output.