@stll/anonymize-wasm 1.4.7 → 1.4.9

package/dist/wasm.mjs

@@ -33,6 +33,10 @@ const isLegalFormsEnabled = (config) => config.enableLegalForms !== false;
 * shallow copies (spread). Uses position + label so the
 * key is identical for the original object and any
 * `{ ...entity }` copy produced by mergeAndDedup.
+*
+* @deprecated No longer used internally: coref alias
+*   links travel on the entities themselves
+*   (`corefSourceText`). Kept for API compatibility.
 */
 const corefKey = (e) => `${e.start}:${e.end}:${e.label}`;
 /** Create a fresh, empty pipeline context. */
@@ -62,8 +66,7 @@ const createPipelineContext = () => ({
 	roleStopSetPromise: null,
 	zoneHeadingPatterns: null,
 	zoneSigningPatterns: null,
-	zoneInitPromise: null,
-	corefSourceMap: /* @__PURE__ */ new Map()
+	zoneInitPromise: null
 });
 /**
 * Module-level default context. Used when callers
@@ -1761,11 +1764,14 @@ const isWordChar = (ch) => {
 * character before the start and after the end are NOT
 * word characters (letter/digit).
 *
-* Populates `ctx.corefSourceMap` with entries linking
-* each coref entity to its source entity text, for
-* consistent placeholder numbering.
+* Each returned alias carries `corefSourceText` linking
+* it to its source entity text, for consistent
+* placeholder numbering.
+*
+* @param _ctx Unused. Kept for signature compatibility;
+*   alias links now travel on the entities themselves.
 */
-const findCoreferenceSpans = (fullText, terms, ctx = defaultContext) => {
+const findCoreferenceSpans = (fullText, terms, _ctx = defaultContext) => {
 	const results = [];
 	for (const term of terms) {
 		let searchFrom = 0;
@@ -1779,16 +1785,15 @@ const findCoreferenceSpans = (fullText, terms, ctx = defaultContext) => {
 				searchFrom = idx + 1;
 				continue;
 			}
-			const entity = {
+			results.push({
 				start: idx,
 				end: matchEnd,
 				label: term.label,
 				text: term.alias,
 				score: .95,
-				source: DETECTION_SOURCES.COREFERENCE
-			};
-			ctx.corefSourceMap.set(corefKey(entity), term.sourceText);
-			results.push(entity);
+				source: DETECTION_SOURCES.COREFERENCE,
+				corefSourceText: term.sourceText
+			});
 			searchFrom = matchEnd;
 		}
 	}
@@ -8055,32 +8060,216 @@ const initNameCorpus = (ctx = defaultContext, dictionaries, languages) => {
 	ctx.nameCorpusPromise = promise;
 	return promise;
 };
-const INFLECTION_SUFFIXES = [
-	"ovi",
-	"em",
-	"om",
-	"ou",
-	"é",
-	"a",
-	"u"
+const DECLENSION_RULES = [
+	{
+		ending: "",
+		gate: /[bdfghklmnpqrstvwxzł]$/u,
+		forms: [
+			"a",
+			"u",
+			"ovi",
+			"em",
+			"om"
+		]
+	},
+	{
+		ending: "",
+		gate: /[bdflmnpstvwz]$/u,
+		forms: ["e"]
+	},
+	{
+		ending: "",
+		gate: /[cčďťňřšžjľ]$/u,
+		forms: [
+			"e",
+			"i",
+			"a",
+			"ovi",
+			"em",
+			"om"
+		]
+	},
+	{
+		ending: "ek",
+		gate: /[^aeiouyáéěíóôúůý]ek$/u,
+		forms: [
+			"ka",
+			"ku",
+			"kovi",
+			"kem",
+			"kom"
+		]
+	},
+	{
+		ending: "el",
+		gate: /[^aeiouyáéěíóôúůý]el$/u,
+		forms: [
+			"la",
+			"lu",
+			"le",
+			"lovi",
+			"lem",
+			"lom"
+		]
+	},
+	{
+		ending: "ec",
+		gate: /[^aeiouyáéěíóôúůý]ec$/u,
+		forms: [
+			"ce",
+			"ci",
+			"covi",
+			"cem",
+			"com"
+		]
+	},
+	{
+		ending: "a",
+		gate: /[^cčďťňřšžji]a$/u,
+		forms: [
+			"y",
+			"u",
+			"o",
+			"ou",
+			"ovi"
+		]
+	},
+	{
+		ending: "a",
+		gate: /[cčďťňřšžj]a$/u,
+		forms: [
+			"i",
+			"u",
+			"o",
+			"ou",
+			"ovi"
+		]
+	},
+	{
+		ending: "a",
+		gate: /ia$/u,
+		forms: [
+			"e",
+			"i",
+			"u",
+			"ou"
+		]
+	},
+	{
+		ending: "ka",
+		gate: /ka$/u,
+		forms: ["ce"]
+	},
+	{
+		ending: "ra",
+		gate: /ra$/u,
+		forms: ["ře", "re"]
+	},
+	{
+		ending: "ha",
+		gate: /ha$/u,
+		forms: ["ze"]
+	},
+	{
+		ending: "ga",
+		gate: /ga$/u,
+		forms: ["ze"]
+	},
+	{
+		ending: "cha",
+		gate: /cha$/u,
+		forms: ["še"]
+	},
+	{
+		ending: "a",
+		gate: /[bdfmnptv]a$/u,
+		forms: ["ě", "e"]
+	},
+	{
+		ending: "a",
+		gate: /[szl]a$/u,
+		forms: ["e"]
+	},
+	{
+		ending: "á",
+		gate: /á$/u,
+		forms: [
+			"é",
+			"ou",
+			"ej",
+			"ú"
+		]
+	},
+	{
+		ending: "ý",
+		gate: /ý$/u,
+		forms: [
+			"ého",
+			"ému",
+			"ém",
+			"ým"
+		]
+	},
+	{
+		ending: "",
+		gate: /[íiy]$/u,
+		forms: [
+			"ho",
+			"mu",
+			"m"
+		]
+	},
+	{
+		ending: "e",
+		gate: /[^aeouyáéěíóôúůý]e$/u,
+		forms: ["i", "í"]
+	},
+	{
+		ending: "o",
+		gate: /o$/u,
+		forms: [
+			"a",
+			"ovi",
+			"em",
+			"om"
+		]
+	}
 ];
 /**
-* Strip common Czech/Slovak case suffixes from a token.
-* Returns candidate base forms if stripping produces a
-* plausible name (capitalised, length >= 3).
-*
-* For the "-ou" instrumental feminine suffix, also yields
-* base + "a" (e.g., "Editou" → "Edit" and "Edita")
-* because Czech feminine names decline -a → -ou.
+* Generate declined Czech/Slovak variants of a
+* nominative name. Only variants licensed by the
+* ending-shape rules are produced, which bounds the
+* deny-list AC pattern growth. Returns an empty array
+* for names too short to decline safely.
+*/
+const expandNameDeclensions = (name) => {
+	if (name.length < 3) return [];
+	const lower = name.toLowerCase();
+	const variants = [];
+	for (const rule of DECLENSION_RULES) {
+		if (!rule.gate.test(lower)) continue;
+		const stem = name.slice(0, name.length - rule.ending.length);
+		if (stem.length < 2) continue;
+		for (const form of rule.forms) variants.push(stem + form);
+	}
+	return variants;
+};
+/**
+* Strip Czech/Slovak case endings from a token using
+* the inverse of DECLENSION_RULES. Returns candidate
+* nominative forms; each candidate is validated against
+* the rule's shape gate, so only bases the paradigm
+* could actually have declined are proposed. Callers
+* check candidates against the name corpus.
 */
 const stripInflection = (token) => {
 	const candidates = [];
-	for (const suffix of INFLECTION_SUFFIXES) if (token.length > suffix.length + 2 && token.endsWith(suffix)) {
-		const base = token.slice(0, -suffix.length);
-		if (/^\p{Lu}/u.test(base)) {
-			candidates.push(base);
-			if (suffix === "ou" || suffix === "é" || suffix === "u") candidates.push(`${base}a`);
-		}
+	for (const rule of DECLENSION_RULES) for (const form of rule.forms) {
+		if (token.length <= form.length + 2 || !token.endsWith(form)) continue;
+		const base = token.slice(0, token.length - form.length) + rule.ending;
+		if (!/^\p{Lu}/u.test(base)) continue;
+		if (!rule.gate.test(base.toLowerCase())) continue;
+		candidates.push(base);
 	}
 	return candidates;
 };
@@ -8340,7 +8529,7 @@ const detectNameCorpus = (fullText, ctx = defaultContext) => {
 const SLASH_S_RE = /\/s\/[ \t]*/g;
 const LABELLED_NAME_RE = /\b(?:By|Name)[ \t]*:[ \t]*(?:\/s\/[ \t]+)?([^\n]*?)(?=\s{3,}|[\t]|\n|$)/gi;
 const WITNESS_ANCHOR_RE = /\bIN WITNESS WHEREOF\b[^\n]*\n/gi;
-const NAME_PARTICLE = "(?:de|del|della|der|den|di|du|el|la|le|van|von|y|zu|af|ben|bin|al|d'|d’)";
+const NAME_PARTICLE = "(?:de|del|della|der|den|di|du|da|das|do|dos|el|la|le|van|von|y|zu|af|ben|bin|al|d'|d’)";
 const CAP_TOKEN = "\\p{Lu}[\\p{L}\\p{M}.'\\-]{0,30}";
 const NAME_SHAPE_RE = new RegExp(`^${CAP_TOKEN}(?:[ \\t]+(?:${NAME_PARTICLE}|${CAP_TOKEN})){1,4}$`, "u");
 const MAX_NAME_LEN = 60;
@@ -8592,6 +8781,35 @@ const HONORIFIC_BOUNDARY = new Set([
 	"Señora"
 ]);
 /**
+* Honorifics that are abbreviations: a dot after them is an
+* abbreviation dot (same sentence), so the detector keeps an
+* optional `.` between the title and the name ("Mr. Smith").
+* Every other honorific is a full word — a trailing dot ends a
+* sentence, so the detector must NOT consume it (otherwise a span
+* like "President. The Employee" crosses the sentence boundary).
+*
+* Maintained explicitly, NOT derived from "ends in a dot": "Mr" is
+* an abbreviation written without a dot, and "Lord" is a full word.
+*/
+const HONORIFIC_ABBREVIATION = new Set([
+	"Avv.",
+	"Dott.",
+	"M.",
+	"Me",
+	"Messrs",
+	"Mlle",
+	"Mme",
+	"Mr",
+	"Mrs",
+	"Ms",
+	"Pr",
+	"Pr.",
+	"Sig.",
+	"Sig.ra",
+	"Sr.",
+	"Sra."
+]);
+/**
 * Post-nominal degrees (comma or space separated after name).
 * Plain text; the detector auto-escapes for regex.
 */
@@ -9118,11 +9336,14 @@ const POST_NOMINAL = POST_NOMINALS.toSorted((a, b) => b.length - a.length).map(e
 const NAME_WORD = `\\p{Lu}\\p{Ll}+`;
 const PARTICLE = "(?:van der|van den|de la|della|von|van|dos|ibn|ben|bin|del|zum|zur|ten|ter|da|de|di|al|el|le|la|zu|af|av)";
 const SP = "[^\\S\\n]";
-/** Honorific alternation built from titles.ts config. */
-const HONORIFIC_ALT = [...HONORIFICS].toSorted((a, b) => b.length - a.length).map((h) => {
+/** Honorific alternation built from titles.ts config. Sorted
+* longest-first so e.g. "Sig.ra" wins over "Sig.". */
+const buildHonorificAlt = (entries) => [...entries].toSorted((a, b) => b.length - a.length).map((h) => {
 	const escaped = escapeRegex$2(h);
 	return HONORIFIC_BOUNDARY.has(h) ? `\\b${escaped}` : escaped;
 }).join("|");
+const HONORIFIC_ABBREV_ALT = buildHonorificAlt(HONORIFICS.filter((h) => HONORIFIC_ABBREVIATION.has(h)));
+const HONORIFIC_FULLWORD_ALT = buildHonorificAlt(HONORIFICS.filter((h) => !HONORIFIC_ABBREVIATION.has(h)));
 const AMOUNT_WORDS = amount_words_default;
 const toEntry = (validator, label, score) => {
 	const pattern = toRegex(validator).source;
@@ -9214,7 +9435,7 @@ const TITLED_PERSON = {
 	score: .95
 };
 const HONORIFIC_PERSON = {
-	pattern: `(?:${HONORIFIC_ALT})\\.?${SP}+${NAME_WORD}(?:(?:${SP}|-){1,2}(?:${PARTICLE}${SP}+)?${NAME_WORD}){0,3}(?:${SP}+(?:QC|KC|SC|LJ|AG))?`,
+	pattern: `(?:(?:${HONORIFIC_ABBREV_ALT})\\.?|(?:${HONORIFIC_FULLWORD_ALT}))${SP}+${NAME_WORD}(?:(?:${SP}|-){1,2}(?:${PARTICLE}${SP}+)?${NAME_WORD}){0,3}(?:${SP}+(?:QC|KC|SC|LJ|AG))?`,
 	label: "person",
 	score: .95
 };
@@ -10089,7 +10310,8 @@ const dropOverlapping = (candidates) => {
 //#region src/detectors/triggers.ts
 const VALID_ID_VALIDATORS = {
 	"br.cpf": br.cpf,
-	"br.cnpj": br.cnpj
+	"br.cnpj": br.cnpj,
+	"us.rtn": us.rtn
 };
 const TRIGGER_SCORE = .95;
 const WHITESPACE_RE$1 = /\s+/;
@@ -10105,6 +10327,8 @@ const DECIMAL_COMMA_RE = new RegExp(`^,(?:\\d|${DASH}{1,2})`);
 * etc.), skip the comma and degree, then continue.
 */
 const POST_NOMINAL_RE = new RegExp(`^,\\s*(?:${POST_NOMINALS.toSorted((a, b) => b.length - a.length).map((d) => d.replace(/[.*+?^${}()|[\]\\]/g, "\\$&").replace(/\\\./g, "\\.\\s*")).join("|")})\\.?`, "i");
+const PERSON_NAME_TOKEN = "\\p{Lu}[\\p{L}\\p{M}.'’\\-]*";
+const PERSON_NAME_RUN_RE = new RegExp(`^${PERSON_NAME_TOKEN}(?:,?[ \\t]+(?:${`(?:${NAME_PARTICLE}[ \\t]+|d['’])`}){0,2}${PERSON_NAME_TOKEN})*`, "u");
 let cachedLegalFormCheckSource = null;
 let cachedLegalFormCheckForms = [];
 const LEGAL_FORM_ALNUM_RE = /[\p{L}\p{N}]/u;
@@ -10772,9 +10996,16 @@ const processTriggerMatches = (allMatches, sliceStart, sliceEnd, fullText, rules
 			if (!applyValidations(value.text, rule.validations)) continue;
 			if (rule.label === "phone number" && !isPlausiblePhoneTriggerValue(value.text)) continue;
 			const entityStart = rule.includeTrigger ? match.start : value.start;
-			const entityEnd = value.end;
-			const entityText = fullText.slice(entityStart, entityEnd);
+			let entityEnd = value.end;
+			let entityText = fullText.slice(entityStart, entityEnd);
 			const effectiveLabel = rule.label === "person" && hasKnownLegalFormSuffix(entityText) ? "organization" : rule.label;
+			if (effectiveLabel === "person") {
+				const run = PERSON_NAME_RUN_RE.exec(value.text);
+				if (run !== null && run[0].length < value.text.length) {
+					entityEnd = value.start + run[0].length;
+					entityText = fullText.slice(entityStart, entityEnd);
+				}
+			}
 			results.push({
 				start: entityStart,
 				end: entityEnd,
@@ -11398,6 +11629,9 @@ const EMPTY_ALLOW_LIST = /* @__PURE__ */ new Set();
 const getAllowList = (ctx) => ctx.allowList ?? EMPTY_ALLOW_LIST;
 let commonWordsPromise = null;
 let commonWordsCache = null;
+const EMPTY_COMMON_WORDS = /* @__PURE__ */ new Set();
+/** Sync accessor — returns empty set before init. */
+const getCommonWords = () => commonWordsCache ?? EMPTY_COMMON_WORDS;
 const loadCommonWords = () => {
 	if (commonWordsCache) return Promise.resolve(commonWordsCache);
 	if (commonWordsPromise) return commonWordsPromise;
@@ -11845,8 +12079,21 @@ const appendNameCorpusEntries = (config, ctx, patternList, labelList, sourceList
 			sourceList.push(source);
 		}
 	};
-	for (const name of getNameCorpusFirstNames(ctx)) addNameEntry(name, "first-name");
-	for (const name of getNameCorpusSurnames(ctx)) addNameEntry(name, "surname");
+	const commonWords = getCommonWords();
+	const addDeclinedVariants = (name, source) => {
+		for (const variant of expandNameDeclensions(name)) {
+			if (commonWords.has(variant.toLowerCase())) continue;
+			addNameEntry(variant, source);
+		}
+	};
+	for (const name of getNameCorpusFirstNames(ctx)) {
+		addNameEntry(name, "first-name");
+		addDeclinedVariants(name, "first-name");
+	}
+	for (const name of getNameCorpusSurnames(ctx)) {
+		addNameEntry(name, "surname");
+		addDeclinedVariants(name, "surname");
+	}
 	for (const title of getNameCorpusTitles(ctx)) {
 		const norm = stripCuratedPatternSyntax(normalizeForSearch(title));
 		if (norm.length === 0) continue;
@@ -12623,25 +12870,33 @@ const isCallerOwnedEntity$2 = (entity) => entity.sourceDetail === "custom-deny-l
 * to get the base name, and re-scan the full text for
 * bare mentions of that base name. Returns new entities
 * for occurrences not already covered.
+*
+* Propagated mentions are coref aliases: each carries
+* `corefSourceText` linking it to the full seed entity
+* text, so placeholder numbering assigns the bare
+* mention the same placeholder as its source ("Acme"
+* and "Acme Corp." both become [ORGANIZATION_1]).
 */
 const propagateOrgNames = (entities, fullText) => {
-	const seeds = [];
-	const seenBases = /* @__PURE__ */ new Set();
+	const seedByBase = /* @__PURE__ */ new Map();
 	for (const e of entities) {
 		if (e.label !== "organization") continue;
 		if (isCallerOwnedEntity$2(e)) continue;
 		for (const suffix of LEGAL_SUFFIXES) if (e.text.endsWith(suffix)) {
 			const base = e.text.slice(0, -suffix.length).replace(TRAILING_SEP, "").trim();
-			if (base.length >= 3 && !seenBases.has(base)) {
-				seenBases.add(base);
-				seeds.push({
+			if (base.length >= 3) {
+				const existing = seedByBase.get(base);
+				if (existing === void 0) seedByBase.set(base, {
 					baseName: base,
-					label: e.label
+					label: e.label,
+					sourceText: e.text
 				});
+				else if (existing.sourceText !== e.text) existing.sourceText = base;
 			}
 			break;
 		}
 	}
+	const seeds = [...seedByBase.values()];
 	if (seeds.length === 0) return [];
 	const covered = entities.map((e) => [e.start, e.end]);
 	const isOverlapping = (start, end) => covered.some(([cs, ce]) => start < ce && end > cs);
@@ -12675,7 +12930,8 @@ const propagateOrgNames = (entities, fullText) => {
 					label,
 					text: fullText.slice(spanStart, matchEnd),
 					score: ORG_PROPAGATION_SCORE,
-					source: DETECTION_SOURCES.COREFERENCE
+					source: DETECTION_SOURCES.COREFERENCE,
+					corefSourceText: seed.sourceText
 				});
 				covered.push([spanStart, matchEnd]);
 			}
@@ -14764,7 +15020,6 @@ const runPipeline = async (options) => {
 	const merged = filterFalsePositives(postOrgEntities, ctx, fullText);
 	if (merged.length < postOrgEntities.length) log("filter", `removed ${postOrgEntities.length - merged.length} FPs`);
 	checkAbort(signal);
-	ctx.corefSourceMap.clear();
 	if (config.enableCoreference) {
 		if (!legalFormsEnabled) await warmLegalRoleHeads();
 		const terms = await extractDefinedTerms(fullText, merged.filter((entity) => !isCallerOwnedEntity(entity)), ctx);
@@ -14830,12 +15085,11 @@ const normalizeEntityText = (label, text) => {
 * Placeholder format: [LABEL_N] where LABEL is uppercase
 * and N is a 1-based counter per label.
 *
-* @param ctx Pipeline context. Must be the same instance
-*   passed to `runPipeline` (or `findCoreferenceSpans`)
-*   so coreference placeholder links are preserved.
-*   Defaults to `defaultContext` for single-tenant usage.
+* @param _ctx Unused. Kept for signature compatibility;
+*   coref alias links now travel on the entities
+*   themselves (`corefSourceText`).
 */
-const buildPlaceholderMap = (entities, ctx = defaultContext) => {
+const buildPlaceholderMap = (entities, _ctx = defaultContext) => {
 	const counters = /* @__PURE__ */ new Map();
 	const textLabelToPlaceholder = /* @__PURE__ */ new Map();
 	const normalizedToPlaceholder = /* @__PURE__ */ new Map();
@@ -14844,9 +15098,9 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const compositeKey = `${entity.label}\0${entity.text}`;
 		if (textLabelToPlaceholder.has(compositeKey)) continue;
 		const labelKey = entity.label.toUpperCase().replace(WHITESPACE_RE, "_");
-		const sourceText = ctx.corefSourceMap.get(corefKey(entity));
-		if (sourceText !== void 0) {
-			const sourceNormalizedKey = `${labelKey}\0${normalizeEntityText(entity.label, sourceText)}`;
+		const sourceText = entity.source === "coreference" ? entity.corefSourceText : void 0;
+		const sourceNormalizedKey = sourceText === void 0 ? void 0 : `${labelKey}\0${normalizeEntityText(entity.label, sourceText)}`;
+		if (sourceNormalizedKey !== void 0) {
 			const sourceExisting = normalizedToPlaceholder.get(sourceNormalizedKey);
 			if (sourceExisting) {
 				textLabelToPlaceholder.set(compositeKey, sourceExisting);
@@ -14857,6 +15111,7 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const existing = normalizedToPlaceholder.get(normalizedKey);
 		if (existing) {
 			textLabelToPlaceholder.set(compositeKey, existing);
+			if (sourceNormalizedKey !== void 0) normalizedToPlaceholder.set(sourceNormalizedKey, existing);
 			continue;
 		}
 		const count = (counters.get(labelKey) ?? 0) + 1;
@@ -14864,6 +15119,7 @@ const buildPlaceholderMap = (entities, ctx = defaultContext) => {
 		const placeholder = `[${labelKey}_${count}]`;
 		textLabelToPlaceholder.set(compositeKey, placeholder);
 		normalizedToPlaceholder.set(normalizedKey, placeholder);
+		if (sourceNormalizedKey !== void 0) normalizedToPlaceholder.set(sourceNormalizedKey, placeholder);
 	}
 	return textLabelToPlaceholder;
 };
@@ -14906,7 +15162,7 @@ const redactText = (fullText, entities, config = DEFAULT_OPERATOR_CONFIG, ctx =
 		const replacement = operator.apply(entity.text, entity.label, placeholder, config.redactString);
 		parts.push(replacement);
 		operatorMap.set(placeholder, opType);
-		if (operator.reversibility === "reversible" && !redactionMap.has(placeholder)) redactionMap.set(placeholder, entity.text);
+		if (operator.reversibility === "reversible" && !redactionMap.has(placeholder)) redactionMap.set(placeholder, entity.source === "coreference" ? entity.corefSourceText : entity.text);
 		cursor = entity.end;
 	}
 	if (cursor < fullText.length) parts.push(fullText.slice(cursor));