Struct WordNetPos 

pub struct WordNetPos { /* private fields */ }

Lemma-to-POS membership lookup.

Why: #5399 rejected a process-wide OnceLock<HashMap> — CLAUDE.md permits global state only for the tracing subscriber. Binary-searching the sorted table directly removes the reason the global existed: there is nothing to build, so there is nothing to share. The type is Copy and 16 bytes, so threading it through crate::kg_extract::KgExtractConfig costs a pointer pair rather than an Arc. What: holds the table text and the offset its data starts at. Every lookup is an O(log n) probe over &'static str; no allocation, no interior mutability, no teardown. Test: shipped_table_answers_the_four_pos_classes, mask_is_case_insensitive.

Implementations

impl WordNetPos

pub const fn shipped() -> Self

The vendored WordNet 3.1 table.

Why: const so a caller that wants the shipped data pays nothing — this is what lets KgExtractConfig::default() stay free. What: pairs [TABLE] with its precomputed [TABLE_DATA_START]. Test: shipped_table_answers_the_four_pos_classes.

pub fn from_table(table: &'static str) -> Self

Build a lookup over a caller-supplied table in the shipped format.

Why: the binary search’s edge cases (first record, last record, absent key either side of the range) are invisible against 83k real lemmas but obvious against six synthetic ones. What: same contract as Self::shipped; the caller owes byte-sorted <lemma>\t<mask> lines and an optional # header. Test: lookup_finds_the_first_and_last_records.

pub fn mask(&self, word: &str) -> u8

POS bitmask for word, or 0 when WordNet has never heard of it.

Why: #5399 requires unknown words to FAIL OPEN. Returning 0 rather than an error or a default makes every caller’s “unknown” branch explicit at the call site instead of hidden here. What: probes as given first — the extractor lower-cases its content up front, so that path allocates nothing — and retries lower-cased only when the input actually contains an upper-case character. WordNet index lemmas are all lower-case. A final retry strips a regular inflection ([base_form_candidates]), which is what lets containing read as the verb it is instead of as an unknown word eligible to head a phrase. Test: mask_returns_zero_for_unknown_words, mask_is_case_insensitive, mask_resolves_regular_inflections_to_their_base_form.

pub fn is_known(&self, word: &str) -> bool

Whether WordNet lists word under any part of speech.

pub fn is_noun(&self, word: &str) -> bool

Whether word can be a noun.

pub fn is_adjective_only(&self, word: &str) -> bool

Whether word is an adjective and nothing else.

Why: this is the head-eligibility test the noun-phrase walk uses. A word that can ONLY be an adjective names a property, so it cannot be the head of the phrase — hard in a hard requirement modifies, it does not name. #5399 uses that to SKIP such a token when picking the head, not to reject the triple: the re-walk lands on requirement, which is what the sentence actually asserts. What: true when the ADJ bit is set and the NOUN bit is not. An unknown word has mask 0 and is therefore never adjective-only — the fail-open direction, which is what keeps unknown crate names eligible as heads. Test: adjective_only_catches_hard_and_spares_fast.

pub fn lemma_count(&self) -> usize

Number of records in the table.

Why: the measurement harness and the table’s own sanity floor need it. What: counts data lines — an O(n) scan, so it is not a hot-path call. Test: shipped_table_answers_the_four_pos_classes.

Trait Implementations

impl Clone for WordNetPos

fn clone(&self) -> WordNetPos

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for WordNetPos

impl Debug for WordNetPos

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

Formats the value using the given formatter.

impl Default for WordNetPos

fn default() -> Self

Returns the “default value” for a type.