chronolizer 0.3.0 → 0.4.1

package/dist/locales/tr.mjs

@@ -1,10 +1,9 @@
-import { isIsoDate } from "../ast/schemas.mjs";
 import { BaseLanguageContribution } from "../language/model.mjs";
 import { normalizeNaturalText } from "../natural/text.mjs";
 import { defineLanguagePlugin, languagePluginsLayer } from "../language/registry.mjs";
 import { correctWhitespaceSeparatedText } from "../natural/correction.mjs";
 import { completeNaturalPhrases, fixedCalendarPeriodPhrases, naturalCount, prefixNaturalPhrases } from "../natural/suggestion.mjs";
-import { absoluteDatePeriod, calendarPeriodOffset, candidate, currentYearDatePeriods, datedPeriods, datedQuarterPeriods, fixedDatePeriod, fixedMonthPeriod, fixedQuarterPeriod, fixedYearPeriod, fromNowRange, futurePeriod, futureRange, isoDate, joinedNowCandidate, joinedPeriodCandidate, monthOfRelativeYear, namedCurrentYearDatePeriod, namedDatePeriod, openBoundaryCandidate, parseTrailingCount, periodBoundaryCandidate, periodEndDay, periodRange, periodStartDay, periodToDateRange, periodsFromPhrases, quarterOfRelativeYear, relativePeriod, relativeWeekend, remainingPeriodRange, renderPeriodRange, textAt, trailingPeriod, trailingRange, untilNowRange, validYear } from "./shared.mjs";
+import { absoluteDatePeriod, calendarPeriodOffset, candidate, compileCountAliasNormalizer, compoundCountAliases, countAliasVocabulary, currentYearDatePeriods, datedPeriods, datedQuarterPeriods, decimalTens, decomposeShiftedPeriodRange, fixedMonthPeriod, fixedQuarterPeriod, fixedYearPeriod, fromNowRange, futurePeriod, futureRange, joinedNowCandidate, joinedPeriodCandidate, monthOfRelativeYear, namedCurrentYearDatePeriod, namedDatePeriod, openBoundaryCandidate, parseTrailingCount, periodBoundaryCandidate, periodEndDay, periodRange, periodStartDay, periodToDateRange, periodsFromPhrases, quarterOfRelativeYear, relativePeriod, relativeWeekday, relativeWeekend, remainingPeriodRange, renderPeriodRange, sequentialCountAliases, shiftPeriod, textAt, trailingPeriod, trailingRange, untilNowRange, validYear } from "./shared.mjs";
 import { Effect, Option, String as String$1 } from "effect";
 //#region src/locales/tr.ts
 const months = [
@@ -21,6 +20,61 @@ const months = [
 	"kasım",
 	"aralık"
 ];
+const weekdays = [
+	"pazartesi",
+	"salı",
+	"çarşamba",
+	"perşembe",
+	"cuma",
+	"cumartesi",
+	"pazar"
+];
+const nextWeekdayPhrases = weekdays.map((weekday) => `gelecek ${weekday}`);
+const turkishCountWords = [
+	["iki"],
+	["üç", "uc"],
+	["dört", "dort"],
+	["beş", "bes"],
+	["altı", "alti"],
+	["yedi"],
+	["sekiz"],
+	["dokuz"],
+	["on"],
+	["on bir"],
+	["on iki"],
+	["on üç"],
+	["on dört"],
+	["on beş"],
+	["on altı"],
+	["on yedi"],
+	["on sekiz"],
+	["on dokuz"],
+	["yirmi"]
+];
+const turkishCountOnes = [
+	[1, "bir"],
+	[2, "iki"],
+	[3, "üç"],
+	[4, "dört"],
+	[5, "beş"],
+	[6, "altı"],
+	[7, "yedi"],
+	[8, "sekiz"],
+	[9, "dokuz"]
+];
+const turkishCountAliases = [...sequentialCountAliases(turkishCountWords, 2), ...compoundCountAliases(decimalTens([
+	"yirmi",
+	"otuz",
+	"kırk",
+	"elli",
+	"altmış",
+	"yetmiş",
+	"seksen",
+	"doksan"
+]), turkishCountOnes, (ten, one) => [`${ten} ${one}`])];
+const normalizeTurkishCounts = compileCountAliasNormalizer(turkishCountAliases);
+const turkishCountVocabulary = new Set(countAliasVocabulary(turkishCountAliases));
+const correctTurkish = (input, vocabulary) => correctWhitespaceSeparatedText(input, vocabulary, turkishCountVocabulary);
 const monthAbbreviations = [
 	["oca"],
 	["şub", "sub"],
@@ -303,15 +357,7 @@ const parseNamedDate = (input) => {
 	const named = String$1.match(/^([0-3]?\d)(?:\.)? ([a-zçğıöşü]+\.?)(?:,)? (\d{4})$/u)(input);
 	if (Option.isSome(named)) return namedDatePeriod(textAt(named.value, 3), textAt(named.value, 2), textAt(named.value, 1), monthNumber, dateLabel);
 	const numeric = String$1.match(/^([0-3]?\d)[./-]([01]?\d)[./-](\d{4})$/u)(input);
-	if (Option.isSome(numeric)) {
-		const year = validYear(textAt(numeric.value, 3));
-		const month = Number(textAt(numeric.value, 2));
-		const day = Number(textAt(numeric.value, 1));
-		if (year !== void 0 && month >= 1 && month <= 12) {
-			const value = isoDate(year, month, day);
-			if (isIsoDate(value) && value !== "9999-12-31") return Option.some(fixedDatePeriod(value, dateLabel(day, month, year)));
-		}
-	}
+	if (Option.isSome(numeric)) return namedDatePeriod(textAt(numeric.value, 3), textAt(numeric.value, 2), textAt(numeric.value, 1), Number, dateLabel);
 	const current = String$1.match(/^([0-3]?\d)(?:\.)? ([a-zçğıöşü]+\.?)$/u)(input);
 	return Option.isSome(current) ? namedCurrentYearDatePeriod(textAt(current.value, 2), textAt(current.value, 1), monthNumber, currentDateLabel) : Option.none();
 };
@@ -340,13 +386,13 @@ const parseQuarter = (input) => {
 	const quarter = quarterNumber(textAt(standalone.value, 1));
 	return quarter === void 0 ? Option.none() : Option.some(quarterOfRelativeYear(quarter, 0, `Ç${quarter}`));
 };
-const parseBasePeriod = (input) => {
-	const absoluteDate = absoluteDatePeriod(input, "tr");
-	if (Option.isSome(absoluteDate)) return absoluteDate;
-	const namedDate = parseNamedDate(input);
-	if (Option.isSome(namedDate)) return namedDate;
-	const quarter = parseQuarter(input);
-	if (Option.isSome(quarter)) return quarter;
+const parseDatedPeriod = (input) => {
+	const knownPeriod = Option.firstSomeOf([
+		absoluteDatePeriod(input, "tr"),
+		parseNamedDate(input),
+		parseQuarter(input)
+	]);
+	if (Option.isSome(knownPeriod)) return knownPeriod;
 	const yearMatch = String$1.match(/^(?:yıl |yılı )?(\d{4})$/u)(input);
 	if (Option.isSome(yearMatch)) {
 		const year = validYear(textAt(yearMatch.value, 1));
@@ -358,6 +404,20 @@ const parseBasePeriod = (input) => {
 		const year = validYear(textAt(monthYear.value, 2));
 		if (month !== void 0 && year !== void 0) return Option.some(fixedMonthPeriod(year, month, `${title(textAt(months, month - 1))} ${year}`));
 	}
+	return Option.none();
+};
+const parseBasePeriod = (input) => {
+	const datedPeriod = parseDatedPeriod(input);
+	if (Option.isSome(datedPeriod)) return datedPeriod;
+	const prefixedRelativeMonth = String$1.match(/^(geçen|bu|gelecek) ([a-zçğıöşü]+\.?)$/u)(input);
+	if (Option.isSome(prefixedRelativeMonth)) {
+		const month = monthNumber(textAt(prefixedRelativeMonth.value, 2));
+		if (month !== void 0) {
+			const modifier = textAt(prefixedRelativeMonth.value, 1);
+			const direction = relativeYearDirection(modifier);
+			return Option.some(monthOfRelativeYear(month, direction, `${modifier} ${title(textAt(months, month - 1))}`));
+		}
+	}
 	const relativeMonth = String$1.match(/^([a-zçğıöşü]+\.?) (geçen yıl|gelecek yıl|bu yıl)$/u)(input);
 	const relativeMonthYearFirst = String$1.match(/^(geçen yıl|gelecek yıl|bu yıl) ([a-zçğıöşü]+\.?)$/u)(input);
 	const relativeMatch = Option.firstSomeOf([relativeMonth, relativeMonthYearFirst]);
@@ -378,16 +438,37 @@ const parseBasePeriod = (input) => {
 	if (["gelecek hafta sonu", "önümüzdeki hafta sonu"].includes(input)) return Option.some(relativeWeekend(1, "gelecek hafta sonu"));
 	if (input === "geçen hafta sonundan önceki hafta sonu") return Option.some(relativeWeekend(-2, input));
 	if (input === "gelecek hafta sonundan sonraki hafta sonu") return Option.some(relativeWeekend(2, input));
-	return Option.none();
+	const weekday = nextWeekdayPhrases.indexOf(input);
+	return weekday === -1 ? Option.none() : Option.some(relativeWeekday(weekday, 1, nextWeekdayPhrases[weekday] ?? input));
 };
-const parsePeriod = (input) => {
+const parsePeriodEdge = (input) => {
 	const edge = String$1.match(/^(.+?)(?:ın|in|un|ün) (başı|başlangıcı|sonu)$/u)(input);
-	if (Option.isSome(edge)) {
-		const period = parseBasePeriod(textAt(edge.value, 1));
-		if (Option.isSome(period)) {
-			const isEnd = textAt(edge.value, 2) === "sonu";
-			const canonical = `${period.value.canonical} ${isEnd ? "sonu" : "başı"}`;
-			return Option.some(isEnd ? periodEndDay(period.value, canonical) : periodStartDay(period.value, canonical));
+	if (Option.isNone(edge)) return Option.none();
+	const periodText = textAt(edge.value, 1);
+	const basePeriod = parseBasePeriod(periodText);
+	const implicitUnit = unitAliases.find((entry) => entry[0] === periodText)?.[1];
+	const implicit = units.find((entry) => entry.unit === implicitUnit);
+	const period = Option.isSome(basePeriod) || implicit === void 0 ? basePeriod : Option.some(relativePeriod(implicit.unit, 0, implicit.current));
+	if (Option.isNone(period)) return Option.none();
+	const isEnd = textAt(edge.value, 2) === "sonu";
+	return Option.some(isEnd ? periodEndDay(period.value, input) : periodStartDay(period.value, input));
+};
+const parsePeriod = (input) => {
+	const prefix = String$1.match(/^([1-9]\d*) (gün|gun|hafta|ay|çeyrek|ceyrek|yıl|yil|sene) (önce|sonra) (.+)$/u)(input);
+	const suffix = String$1.match(/^(.+) ([1-9]\d*) (gün|gun|hafta|ay|çeyrek|ceyrek|yıl|yil|sene) (önce|sonra)$/u)(input);
+	const shifted = Option.firstSomeOf([prefix, suffix]);
+	if (Option.isSome(shifted)) {
+		const prefixOrder = Option.isSome(prefix);
+		const amount = parseTrailingCount(textAt(shifted.value, prefixOrder ? 1 : 2));
+		const alias = unitAliases.find((unit) => unit[0] === textAt(shifted.value, prefixOrder ? 2 : 3));
+		const entry = alias === void 0 ? void 0 : units.find((unit) => unit.unit === alias[1]);
+		const period = parsePeriod(textAt(shifted.value, prefixOrder ? 4 : 1));
+		if (Option.isSome(amount) && entry !== void 0 && Option.isSome(period)) {
+			const past = textAt(shifted.value, prefixOrder ? 3 : 4) === "önce";
+			const direction = past ? -amount.value : amount.value;
+			const noun = amount.value === 1 ? entry.singular : entry.plural;
+			const canonical = `${amount.value} ${noun} ${past ? "önce" : "sonra"} ${period.value.canonical}`;
+			return Option.some(shiftPeriod(period.value, direction, entry.unit, canonical));
 		}
 	}
 	const wrapper = [
@@ -395,17 +476,19 @@ const parsePeriod = (input) => {
 		"içinde ",
 		"tüm "
 	].find((prefix) => input.startsWith(prefix));
-	return parseBasePeriod(wrapper === void 0 ? input : input.slice(wrapper.length));
+	return Option.firstSomeOf([
+		parsePeriodEdge(input),
+		parseBasePeriod(wrapper === void 0 ? input : input.slice(wrapper.length)),
+		parseCalendarOffset(input)
+	]);
 };
 const countedUnit = (value) => unitAliases.find((entry) => entry[0] === value)?.[1];
-const countedUnitPattern = "gün|gun|hafta|ay|çeyrek|ceyrek|yıl|yil|sene";
-const countedPattern = (source) => new RegExp(source.replace("UNIT", countedUnitPattern), "u");
-const calendarPastPattern = countedPattern("^([1-9]\\d*) (UNIT) önce$");
-const calendarFuturePatterns = [countedPattern("^([1-9]\\d*) (UNIT) sonra$"), countedPattern("^([1-9]\\d*) (UNIT) içinde$")];
-const rollingPastPatterns = [countedPattern("^(?:son|geçen|önceki|geçtiğimiz) ([1-9]\\d*) (UNIT)$"), countedPattern("^([1-9]\\d*) (?:son|geçen|önceki) (UNIT)$")];
-const rollingFuturePatterns = [countedPattern("^(?:gelecek|önümüzdeki|sonraki) ([1-9]\\d*) (UNIT)$"), countedPattern("^([1-9]\\d*) (?:gelecek|sonraki) (UNIT)$")];
-const rollingSincePattern = countedPattern("^([1-9]\\d*) (UNIT) boyunca$");
-const rollingBarePattern = countedPattern("^([1-9]\\d*) (UNIT)$");
+const calendarPastPattern = /^([1-9]\d*) ([^ ]+) önce$/u;
+const calendarFuturePatterns = [/^([1-9]\d*) ([^ ]+) sonra$/u, /^([1-9]\d*) ([^ ]+) içinde$/u];
+const rollingPastPatterns = [/^(?:son|geçen|önceki|geçtiğimiz) ([1-9]\d*) ([^ ]+)$/u, /^([1-9]\d*) (?:son|geçen|önceki) ([^ ]+)$/u];
+const rollingFuturePatterns = [/^(?:gelecek|önümüzdeki|sonraki) ([1-9]\d*) ([^ ]+)$/u, /^([1-9]\d*) (?:gelecek|sonraki) ([^ ]+)$/u];
+const rollingSincePattern = /^([1-9]\d*) ([^ ]+) boyunca$/u;
+const rollingBarePattern = /^([1-9]\d*) ([^ ]+)$/u;
 const firstPatternMatch = (input, patterns) => Option.firstSomeOf(patterns.map((pattern) => String$1.match(pattern)(input)));
 const singularRollingCanonical = (entry, future) => future ? `önümüzdeki bir ${entry.singular}` : `son bir ${entry.singular}`;
 const singularRollingPhrases = units.flatMap((entry) => [
@@ -446,7 +529,7 @@ const singularCalendarOffsets = units.flatMap((entry) => [{
 }]);
 const parseCalendarOffset = (input) => {
 	const singular = singularCalendarOffsets.find((entry) => entry.phrase === input);
-	if (singular !== void 0) return Option.some(candidate(periodRange(relativePeriod(singular.entry.unit, singular.direction, singular.phrase)), singular.phrase));
+	if (singular !== void 0) return Option.some(relativePeriod(singular.entry.unit, singular.direction, singular.phrase));
 	const past = String$1.match(calendarPastPattern)(input);
 	const future = firstPatternMatch(input, calendarFuturePatterns);
 	const match = Option.firstSomeOf([past, future]);
@@ -459,7 +542,7 @@ const parseCalendarOffset = (input) => {
 	const direction = Option.isSome(past) ? -amount.value : amount.value;
 	const noun = amount.value === 1 ? entry.singular : entry.plural;
 	const canonical = direction < 0 ? `${amount.value} ${noun} önce` : `${amount.value} ${noun} sonra`;
-	return Option.some(candidate(periodRange(relativePeriod(unit, direction, canonical)), canonical));
+	return Option.some(relativePeriod(unit, direction, canonical));
 };
 const parseRollingPeriod = (input) => {
 	const singular = singularRollingPhrases.find((entry) => entry.phrase === input);
@@ -543,8 +626,6 @@ const parseTurkish = (input) => {
 		"şu andan itibaren",
 		"bundan sonra"
 	].includes(input)) return Option.some(candidate(fromNowRange(), "bugünden itibaren"));
-	const offset = parseCalendarOffset(input);
-	if (Option.isSome(offset)) return offset;
 	const rolling = parseRollingPeriod(input);
 	if (Option.isSome(rolling)) return rolling;
 	const toDate = toDatePhrases.find((entry) => entry.phrase === input);
@@ -552,7 +633,12 @@ const parseTurkish = (input) => {
 	const elided = parseElidedDateRange(input);
 	if (Option.isSome(elided)) return elided;
 	const joinedInput = input.endsWith(" arası") ? input.slice(0, -6) : input;
-	const nowBounded = joinedNowCandidate(joinedInput, [["", " ile bugün"], ["", " ve bugün"]], ["bugün ile ", "bugün ve "], parsePeriod, (period) => `${period} ile bugün arası`, (period) => `bugün ile ${period} arası`);
+	const nowBounded = joinedNowCandidate(joinedInput, [["", " ile bugün"], ["", " ve bugün"]], [
+		"bugün ile ",
+		"bugün ve ",
+		"şimdi ile ",
+		"şimdi ve "
+	], parsePeriod, (period) => `${period} ile bugün arası`, (period) => `bugün ile ${period} arası`);
 	if (Option.isSome(nowBounded)) return nowBounded;
 	const bounded = joinedPeriodCandidate(joinedInput, [
 		["", " ile "],
@@ -591,6 +677,8 @@ const staticPeriodPhrases = [
 	"geçen yılın sonu",
 	"gelecek yılın başı",
 	"gelecek yılın sonu",
+	...nextWeekdayPhrases,
+	...months.flatMap((month) => [`geçen ${month}`, `gelecek ${month}`]),
 	...[
 		1,
 		2,
@@ -651,6 +739,17 @@ const suggestTurkish = (input, limit) => {
 	], limit);
 };
 const renderTurkish = (range) => {
+	const shifted = decomposeShiftedPeriodRange(range);
+	if (Option.isSome(shifted)) {
+		const base = renderTurkish(shifted.value.baseRange);
+		const entry = units.find((unit) => unit.unit === shifted.value.unit);
+		if (Option.isSome(base) && entry !== void 0) {
+			const amount = Math.abs(shifted.value.amount);
+			const noun = amount === 1 ? entry.singular : entry.plural;
+			const direction = shifted.value.amount < 0 ? "önce" : "sonra";
+			return Option.some(`${amount} ${noun} ${direction} ${base.value}`);
+		}
+	}
 	const offset = calendarPeriodOffset(range);
 	if (Option.isSome(offset) && Math.abs(offset.value.amount) > 1) {
 		const entry = units.find((unit) => unit.unit === offset.value.unit);
@@ -677,11 +776,12 @@ const renderTurkish = (range) => {
 	];
 	return renderPeriodRange(range, [...units.map((entry) => candidate(periodToDateRange(entry.unit), entry.toDate)), ...units.map((entry) => candidate(remainingPeriodRange(entry.unit), entry.remaining))], periods, (period) => `${period} itibarıyla`, (period) => `${period} öncesi`, (period) => `${period} sonuna kadar`, (period) => `${period} sonrası`, (lower, upper) => `${lower} ile ${upper} arası`, (period) => `${period} ile bugün arası`, (period) => `bugün ile ${period} arası`, () => "şimdiye kadar", () => "bugünden itibaren");
 };
-const normalizeTurkish = (input, locale) => normalizeNaturalText(input, locale).replaceAll("’", "'");
+const normalizeTurkish = (input, locale) => normalizeTurkishCounts(normalizeNaturalText(input, locale).replaceAll("’", "'"));
 const TurkishContribution = new BaseLanguageContribution({
 	locale: "tr",
 	vocabulary: [
 		...months,
+		...weekdays,
 		...monthAbbreviations.flatMap((aliases) => aliases),
 		...units.flatMap((entry) => [
 			entry.singular,
@@ -698,6 +798,7 @@ const TurkishContribution = new BaseLanguageContribution({
 		"gelecek",
 		"geçen",
 		"geri",
+		"ile",
 		"itibaren",
 		"itibarıyla",
 		"kadar",
@@ -708,10 +809,11 @@ const TurkishContribution = new BaseLanguageContribution({
 		"son",
 		"sonra",
 		"sonrası",
+		"şimdi",
 		"şimdiye"
 	],
 	normalize: normalizeTurkish,
-	correct: correctWhitespaceSeparatedText,
+	correct: correctTurkish,
 	parseExact: parseTurkish,
 	suggest: suggestTurkish,
 	render: renderTurkish

package/dist/natural/correction.d.mts

@@ -1,5 +1,5 @@
 //#region src/natural/correction.d.ts
-declare const correctWhitespaceSeparatedText: (input: string, vocabulary: ReadonlyArray<string>) => {
+declare const correctWhitespaceSeparatedText: (input: string, vocabulary: ReadonlyArray<string>, segmentationVocabulary?: ReadonlySet<string>) => {
   readonly text: string;
   readonly corrections: readonly {
     readonly original: string;

package/dist/natural/correction.mjs

@@ -1,50 +1,119 @@
 import { Correction, NaturalCorrectionCandidate } from "../language/model.mjs";
 import { naturalWords } from "./text.mjs";
-import { Option, String } from "effect";
 //#region src/natural/correction.ts
-const isProtectedValue = (word) => Option.isSome(String.match(/^\d+$/u)(word)) || Option.isSome(String.match(/^\d{4}-\d{2}-\d{2}$/u)(word));
-const damerauLevenshteinDistance = (left, right) => {
-	const fallback = left.length + right.length;
-	let previousPrevious;
-	let previous = Array.from({ length: right.length + 1 }, (_, index) => index);
+const isProtectedValue = (word) => /^(?:\d+|\d{4}-\d{2}-\d{2})$/u.test(word);
+const damerauLevenshteinDistance = (left, right, maximum = Math.max(left.length, right.length), workspace = [
+	[],
+	[],
+	[]
+]) => {
+	const outside = maximum + 1;
+	if (Math.abs(left.length - right.length) > maximum) return outside;
+	const width = right.length + 1;
+	let [previousPrevious, previous, current] = workspace;
+	for (const row of workspace) {
+		row.length = width;
+		row.fill(outside);
+	}
+	for (let column = 0; column <= Math.min(right.length, maximum); column += 1) previous[column] = column;
 	for (let row = 1; row <= left.length; row += 1) {
-		const current = [row];
-		for (let column = 1; column <= right.length; column += 1) {
-			const substitution = left.charAt(row - 1) === right.charAt(column - 1) ? 0 : 1;
-			const distance = Math.min((previous[column] ?? fallback) + 1, (current[column - 1] ?? fallback) + 1, (previous[column - 1] ?? fallback) + substitution);
-			const transposed = previousPrevious !== void 0 && row > 1 && column > 1 && left.charAt(row - 1) === right.charAt(column - 2) && left.charAt(row - 2) === right.charAt(column - 1) ? (previousPrevious[column - 2] ?? fallback) + 1 : fallback;
-			current.push(Math.min(distance, transposed));
+		current.fill(outside);
+		if (row <= maximum) current[0] = row;
+		const firstColumn = Math.max(1, row - maximum);
+		const lastColumn = Math.min(right.length, row + maximum);
+		let rowMinimum = row <= maximum ? row : outside;
+		for (let column = firstColumn; column <= lastColumn; column += 1) {
+			const substitution = left.charCodeAt(row - 1) === right.charCodeAt(column - 1) ? 0 : 1;
+			let distance = Math.min((previous[column] ?? outside) + 1, (current[column - 1] ?? outside) + 1, (previous[column - 1] ?? outside) + substitution);
+			if (row > 1 && column > 1 && left.charCodeAt(row - 1) === right.charCodeAt(column - 2) && left.charCodeAt(row - 2) === right.charCodeAt(column - 1)) distance = Math.min(distance, (previousPrevious[column - 2] ?? outside) + 1);
+			current[column] = distance;
+			rowMinimum = Math.min(rowMinimum, distance);
 		}
+		if (rowMinimum > maximum) return outside;
+		const reusable = previousPrevious;
 		previousPrevious = previous;
 		previous = current;
+		current = reusable;
+	}
+	return previous[right.length] ?? outside;
+};
+const segmentedReplacements = (word, vocabulary) => {
+	const segmentations = Array.from({ length: word.length + 1 }, () => []);
+	const initial = segmentations[0];
+	if (initial !== void 0) initial.push([]);
+	for (let start = 0; start < word.length; start += 1) {
+		const prefixes = segmentations[start] ?? [];
+		if (prefixes.length === 0) continue;
+		for (let end = start + 1; end <= word.length; end += 1) {
+			const part = word.slice(start, end);
+			if (!vocabulary.has(part)) continue;
+			const target = segmentations[end];
+			if (target === void 0) continue;
+			for (const prefix of prefixes) {
+				if (target.length >= 4) break;
+				target.push([...prefix, part]);
+			}
+		}
 	}
-	return previous[right.length] ?? fallback;
+	return (segmentations[word.length] ?? []).filter((parts) => parts.length > 1).map((parts) => ({
+		word: parts.join(" "),
+		distance: parts.length - 1
+	}));
+};
+const vocabularySets = /* @__PURE__ */ new WeakMap();
+const vocabularySet = (vocabulary) => {
+	const cached = vocabularySets.get(vocabulary);
+	if (cached !== void 0) return cached;
+	const words = new Set(vocabulary);
+	vocabularySets.set(vocabulary, words);
+	return words;
 };
-const replacementsFor = (word, vocabulary) => {
-	if (vocabulary.includes(word) || isProtectedValue(word)) return [{
+const replacementsFor = (word, vocabulary, segmentationVocabulary, workspace) => {
+	if (vocabulary.has(word) || isProtectedValue(word)) return [{
 		word,
 		distance: 0
 	}];
-	if (word.length <= 3) return [];
+	const segmented = segmentedReplacements(word, segmentationVocabulary);
+	if (word.length <= 3) return segmented;
 	const maximum = word.length >= 6 ? 2 : 1;
-	const matches = vocabulary.filter((candidate) => Math.abs(candidate.length - word.length) <= maximum).map((candidate) => ({
-		word: candidate,
-		distance: damerauLevenshteinDistance(word, candidate)
-	})).filter((candidate) => candidate.distance <= maximum);
-	if (matches.length === 0) return [];
-	const minimum = Math.min(...matches.map((candidate) => candidate.distance));
-	return matches.filter((candidate) => candidate.distance === minimum).slice(0, 4);
+	let minimum = Number.POSITIVE_INFINITY;
+	const matches = [];
+	const addMatch = (replacement, distance) => {
+		if (distance > minimum) return;
+		if (distance < minimum) {
+			minimum = distance;
+			matches.length = 0;
+		}
+		if (matches.length < 4) matches.push({
+			word: replacement,
+			distance
+		});
+	};
+	for (const replacement of segmented) addMatch(replacement.word, replacement.distance);
+	for (const candidate of vocabulary) {
+		if (Math.abs(candidate.length - word.length) > maximum) continue;
+		const distance = damerauLevenshteinDistance(word, candidate, maximum, workspace);
+		if (distance <= maximum) addMatch(candidate, distance);
+	}
+	return matches;
 };
-const correctWhitespaceSeparatedText = (input, vocabulary) => {
+const emptySegmentationVocabulary = /* @__PURE__ */ new Set();
+const correctWhitespaceSeparatedText = (input, vocabulary, segmentationVocabulary = emptySegmentationVocabulary) => {
 	const words = naturalWords(input);
+	const wordsInVocabulary = vocabularySet(vocabulary);
+	const workspace = [
+		[],
+		[],
+		[]
+	];
 	let partials = [{
-		words: [],
+		text: "",
 		corrections: [],
 		cost: 0,
 		offset: 0
 	}];
 	for (const word of words) {
-		const replacements = replacementsFor(word, vocabulary);
+		const replacements = replacementsFor(word, wordsInVocabulary, segmentationVocabulary, workspace);
 		if (replacements.length === 0) return [];
 		const next = [];
 		for (const partial of partials) for (const replacement of replacements) {
@@ -55,7 +124,7 @@ const correctWhitespaceSeparatedText = (input, vocabulary) => {
 				offset: partial.offset
 			})];
 			next.push({
-				words: [...partial.words, replacement.word],
+				text: partial.text.length === 0 ? replacement.word : `${partial.text} ${replacement.word}`,
 				corrections: correction,
 				cost: partial.cost + replacement.distance,
 				offset: partial.offset + word.length + 1
@@ -64,7 +133,7 @@ const correctWhitespaceSeparatedText = (input, vocabulary) => {
 		partials = next.slice(0, 32);
 	}
 	return partials.filter((partial) => partial.corrections.length > 0).map((partial) => NaturalCorrectionCandidate.make({
-		text: partial.words.join(" "),
+		text: partial.text,
 		corrections: partial.corrections,
 		cost: partial.cost
 	}));

package/dist/natural/format.mjs

@@ -7,7 +7,7 @@ const formatNatural = Effect.fn("chronolizer.formatNatural")(function* (range, o
 	if (Option.isSome(rendered)) return rendered.value;
 	return yield* new NaturalLanguageRenderError({
 		locale: options.locale,
-		message: "The range has no canonical natural-language form in this locale"
+		message: "The range has no standard natural-language form for this language"
 	});
 });
 //#endregion

package/dist/natural/parse.mjs

@@ -5,7 +5,7 @@ import { LanguageRegistry } from "../language/registry.mjs";
 import { applyFuturePolicy } from "./policy.mjs";
 import { Array, Effect, Order, Result } from "effect";
 //#region src/natural/parse.ts
-const distinctCandidates = (candidates) => Array.dedupeWith(candidates, (left, right) => rangeKey(left.range) === rangeKey(right.range));
+const distinctCandidates = (candidates) => candidates.length === 1 ? candidates : Array.dedupeWith(candidates, (left, right) => rangeKey(left.range) === rangeKey(right.range));
 const parseQuality = (hasAlternatives, hasCorrections) => {
 	if (hasAlternatives) return "ambiguous";
 	if (hasCorrections) return "corrected";
@@ -31,13 +31,13 @@ const parseNatural = Effect.fn("chronolizer.parseNatural")(function* (input, opt
 	if (normalized.length === 0) return yield* new NaturalLanguageParseError({
 		input,
 		locale: options.locale,
-		message: "The complete input must contain a date-range expression"
+		message: "The complete input must contain a date range"
 	});
 	const exact = language.parseExact(normalized);
 	const allowedExact = applyFuturePolicy(exact, options.allowFuture);
 	if (Array.isReadonlyArrayNonEmpty(allowedExact)) return resultFromCandidates(allowedExact, []);
 	if (options.typoMode !== "tolerant") {
-		const message = exact.length > 0 && options.allowFuture === false ? "The expression contains a positive relative shift, but future ranges are disabled" : "The complete input is not a supported date-range expression";
+		const message = exact.length > 0 && options.allowFuture === false ? "This date range is in the future, but future ranges are disabled" : "The complete input is not a supported date range";
 		return yield* new NaturalLanguageParseError({
 			input,
 			locale: options.locale,
@@ -46,7 +46,8 @@ const parseNatural = Effect.fn("chronolizer.parseNatural")(function* (input, opt
 	}
 	const corrected = language.correct?.(normalized, language.vocabulary) ?? [];
 	const parsedCorrections = Array.filterMap(corrected, (correction) => {
-		const candidates = applyFuturePolicy(language.parseExact(correction.text), options.allowFuture);
+		const correctedText = language.normalize(correction.text, language.locale);
+		const candidates = applyFuturePolicy(language.parseExact(correctedText), options.allowFuture);
 		return Array.isReadonlyArrayNonEmpty(candidates) ? Result.succeed({
 			correction,
 			candidates
@@ -56,7 +57,7 @@ const parseNatural = Effect.fn("chronolizer.parseNatural")(function* (input, opt
 	if (!Array.isReadonlyArrayNonEmpty(successful)) return yield* new NaturalLanguageParseError({
 		input,
 		locale: options.locale,
-		message: "No conservative typo correction produced a complete expression"
+		message: "No supported date range was found after spelling correction"
 	});
 	const first = Array.headNonEmpty(successful);
 	const best = Array.prepend(Array.takeWhile(Array.tailNonEmpty(successful), (entry) => entry.correction.cost === first.correction.cost), first);

package/dist/natural/suggest.mjs

@@ -1,24 +1,48 @@
 import { rangeKey } from "../filter/codec.mjs";
 import { NaturalSuggestion } from "../language/model.mjs";
+import { naturalWords } from "./text.mjs";
 import { LanguageRegistry } from "../language/registry.mjs";
 import { applyFuturePolicy } from "./policy.mjs";
-import { Array, Effect } from "effect";
+import { Array as Array$1, Effect } from "effect";
 //#region src/natural/suggest.ts
 const suggestionLimit = (limit) => {
 	if (limit === void 0) return 10;
 	if (!Number.isSafeInteger(limit) || limit <= 0) return 0;
 	return Math.min(limit, 100);
 };
+const singleWordCompletions = (input, vocabulary) => {
+	const words = naturalWords(input);
+	if (words.length === 0 || words.length > 12) return [];
+	const positions = Array.from({ length: words.length + 1 }, (_, index) => index);
+	const insertions = Array$1.flatMap(positions, (index) => Array$1.flatMap(vocabulary, (word) => word.length === 0 || word.includes(" ") ? [] : [[
+		...words.slice(0, index),
+		word,
+		...words.slice(index)
+	].join(" ")]));
+	return Array$1.dedupe(insertions);
+};
 const suggestNatural = Effect.fn("chronolizer.suggestNatural")(function* (input, options) {
 	const language = yield* (yield* LanguageRegistry).resolve(options.locale);
 	const normalized = language.normalize(input, language.locale);
 	const limit = suggestionLimit(options.limit);
 	if (limit === 0) return [];
-	const suggestions = Array.flatMap(Array.dedupe(language.suggest(normalized, limit * 2)), (text) => applyFuturePolicy(language.parseExact(language.normalize(text, language.locale)), options.allowFuture).map((candidate) => NaturalSuggestion.make({
+	const suggestionsFrom = (texts) => Array$1.flatMap(Array$1.dedupe(texts), (text) => applyFuturePolicy(language.parseExact(language.normalize(text, language.locale)), options.allowFuture).map((candidate) => NaturalSuggestion.make({
 		text: candidate.canonical,
 		range: candidate.range
 	})));
-	return Array.take(Array.dedupeWith(suggestions, (left, right) => rangeKey(left.range) === rangeKey(right.range)), limit);
+	const distinct = (suggestions) => {
+		const keys = /* @__PURE__ */ new Set();
+		return suggestions.filter((suggestion) => {
+			const key = rangeKey(suggestion.range);
+			if (keys.has(key)) return false;
+			keys.add(key);
+			return true;
+		});
+	};
+	const suggested = distinct(suggestionsFrom([...language.suggest(normalized, limit * 2), normalized]));
+	if (suggested.length > 0) return Array$1.take(suggested, limit);
+	const completionVocabulary = Array$1.dedupe([...Array$1.flatMap(language.suggest("", 100), naturalWords), ...language.vocabulary]);
+	return Array$1.take(distinct(suggestionsFrom(singleWordCompletions(normalized, completionVocabulary))), limit);
 });
 //#endregion
 export { suggestNatural };