@kensio/pinyinjs 1.7.4 → 1.7.6

package/dist/decode/run-lattice.js

@@ -0,0 +1,91 @@
+import { buildLattice } from "./lattice.js";
+export { runLattice } from "./lattice-slice.js";
+/**
+ * Building the lattice one Han run is decoded over.
+ *
+ * A run is decoded on its own and the pieces are put back together, so this
+ * is where a lattice is cut down to a run and the words are read back out of
+ * it. `decode.ts` is what asks for one.
+ */
+import { decodeSpacing } from "./decode.js";
+import { applyEdgeRules } from "./rules.js";
+/**
+ * Gather the reading units into the words the boundaries make of them.
+ *
+ * Where the two decodes disagree the readings win, which is what "advisory"
+ * means for the spacing decode: a boundary falling inside a reading unit is
+ * dropped rather than allowed to break the unit up. 玩儿 read as `wánr` is one
+ * syllable over two characters, and no spacing can be put between them.
+ */
+export function groupUnits(units, boundaries) {
+    const starts = new Set(units.map((unit) => unit.from));
+    const kept = new Set(boundaries.filter((at) => starts.has(at)));
+    const groups = [];
+    let held = [];
+    for (const unit of units) {
+        if (kept.has(unit.from) && held.length > 0) {
+            groups.push(held);
+            held = [];
+        }
+        held.push(unit);
+    }
+    if (held.length > 0) {
+        groups.push(held);
+    }
+    return groups;
+}
+/**
+ * Assemble one word from the units the boundaries kept together.
+ */
+export function wordFrom(dictionary, lattice, group) {
+    const first = group[0];
+    const last = group.at(-1);
+    /* c8 ignore next 3 -- a group is only made by pushing a unit into it */
+    if (first === undefined || last === undefined) {
+        throw new Error("a word must cover at least one reading unit");
+    }
+    const text = lattice.characters.slice(first.from, last.to).join("");
+    const entry = dictionary.lookup(text);
+    return {
+        text,
+        reading: group.flatMap((unit) => [...unit.reading]),
+        isProperNoun: entry?.isProperNoun ?? false,
+        partOfSpeech: entry?.partOfSpeech ?? "",
+        isKnown: entry !== undefined,
+    };
+}
+/**
+ * The lattice a run decodes over: stage 1 built, then stage 4 applied.
+ *
+ * The rules run between building and decoding rather than over the output,
+ * which is the whole point of them — they change what the decode gets to choose
+ * between, so everything downstream sees one lattice rather than a decode and
+ * a list of corrections to it.
+ */
+export function ruledLattice(dictionary, run, rules, hints) {
+    return applyEdgeRules(buildLattice(dictionary, run, hints), dictionary, rules);
+}
+/**
+ * Cut a decode of the whole lattice down to the run at the end of it.
+ *
+ * Everything before `at` was context rather than run — 汉字 a caller put in
+ * front so that the decode could see them — and is dropped once it has done its
+ * work.
+ */
+export function runGroups({ lattice, at }, units) {
+    return groupUnits(units.filter((unit) => unit.from >= at), decodeSpacing(lattice).filter((from) => from >= at));
+}
+/**
+ * Decode a Han run with the lattice: the recommended path.
+ *
+ * Stages 1 to 3 of ALGORITHM.md end to end — build every candidate reading,
+ * lock the positions that cannot vary, score the rest — where the greedy
+ * baseline commits to the longest match at each position and never revisits it.
+ *
+ * `before` is 汉字 standing in front of the run that the decode should see and
+ * not report: the 汉字 a number written in digits would have been written with.
+ * A run is decoded on its own without it, and a run reached that way has no idea
+ * what preceded it — 个人 alone is the word `gèrén`, while the 两 of 2个人 makes
+ * 两个 the word and leaves 人 to itself.
+ */
+//# sourceMappingURL=run-lattice.js.map

package/dist/decode/run-lattice.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-lattice.js","sourceRoot":"","sources":["../../src/decode/run-lattice.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAgB,MAAM,cAAc,CAAC;AAI1D,OAAO,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAChD;;;;;;GAMG;AACH,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAG5C,OAAO,EAAE,cAAc,EAAiB,MAAM,YAAY,CAAC;AAG3D;;;;;;;GAOG;AACH,MAAM,UAAU,UAAU,CACxB,KAAsB,EACtB,UAA6B;IAE7B,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACvD,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAChE,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,IAAI,GAAW,EAAE,CAAC;IAEtB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3C,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClB,IAAI,GAAG,EAAE,CAAC;QACZ,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACpB,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,QAAQ,CACtB,UAAsB,EACtB,OAAgB,EAChB,KAA6B;IAE7B,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1B,wEAAwE;IACxE,IAAI,KAAK,KAAK,SAAS,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACpE,MAAM,KAAK,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAEtC,OAAO;QACL,IAAI;QACJ,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;QACnD,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,KAAK;QAC1C,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,EAAE;QACvC,OAAO,EAAE,KAAK,KAAK,SAAS;KAC7B,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,YAAY,CAC1B,UAAsB,EACtB,GAAW,EACX,KAA0B,EAC1B,KAAgC;IAEhC,OAAO,cAAc,CACnB,YAAY,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,CAAC,EACpC,UAAU,EACV,KAAK,CACN,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,SAAS,CACvB,EAAE,OAAO,EAAE,EAAE,EAAc,EAC3B,KAAsB;IAEtB,OAAO,UAAU,CACf,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC,EACvC,aAAa,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CACpD,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;GAYG"}

package/dist/decode/run-words.d.ts

@@ -0,0 +1,17 @@
+/**
+ * Recutting a run's words for 分词连写.
+ *
+ * Grouping rewrites word boundaries and never the readings behind them, so the
+ * syllables — and the confidence beside them — survive it in order, and are
+ * simply cut in different places.
+ */
+import type { Dictionary } from "../dictionary/dictionary.js";
+import type { ScoredWord } from "./word.js";
+/**
+ * The words a Han run decodes to, with 分词连写 applied.
+ *
+ * Grouping rewrites word boundaries and never the readings behind them, so the
+ * syllables — and the confidence beside them — survive it in order.
+ */
+export declare function wordsOf(decoded: readonly ScoredWord[], dictionary: Dictionary, isGrouped: boolean): readonly ScoredWord[];
+//# sourceMappingURL=run-words.d.ts.map

package/dist/decode/run-words.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-words.d.ts","sourceRoot":"","sources":["../../src/decode/run-words.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAE9D,OAAO,KAAK,EAAe,UAAU,EAAE,MAAM,WAAW,CAAC;AAEzD;;;;;GAKG;AACH,wBAAgB,OAAO,CACrB,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,UAAU,EAAE,UAAU,EACtB,SAAS,EAAE,OAAO,GACjB,SAAS,UAAU,EAAE,CAUvB"}

package/dist/decode/run-words.js

@@ -0,0 +1,29 @@
+import { applyGrouping } from "../orthography/grouping.js";
+/**
+ * The words a Han run decodes to, with 分词连写 applied.
+ *
+ * Grouping rewrites word boundaries and never the readings behind them, so the
+ * syllables — and the confidence beside them — survive it in order.
+ */
+export function wordsOf(decoded, dictionary, isGrouped) {
+    return isGrouped
+        ? regroup(decoded, applyGrouping(decoded.map((scored) => scored.word), dictionary))
+        : decoded;
+}
+/**
+ * Redistribute the decode's per-syllable confidence over regrouped words.
+ *
+ * 分词连写 moves word boundaries — 看 and 了 become 看了 — without touching a
+ * reading, so the run's syllables are the same syllables in the same order and
+ * are simply cut in different places.
+ */
+function regroup(decoded, grouped) {
+    const confidence = decoded.flatMap((scored) => [...scored.confidence]);
+    let at = 0;
+    return grouped.map((word) => {
+        const held = confidence.slice(at, at + word.reading.length);
+        at += word.reading.length;
+        return { word, confidence: held };
+    });
+}
+//# sourceMappingURL=run-words.js.map

package/dist/decode/run-words.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-words.js","sourceRoot":"","sources":["../../src/decode/run-words.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAG3D;;;;;GAKG;AACH,MAAM,UAAU,OAAO,CACrB,OAA8B,EAC9B,UAAsB,EACtB,SAAkB;IAElB,OAAO,SAAS;QACd,CAAC,CAAC,OAAO,CACL,OAAO,EACP,aAAa,CACX,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,EACpC,UAAU,CACX,CACF;QACH,CAAC,CAAC,OAAO,CAAC;AACd,CAAC;AAED;;;;;;GAMG;AACH,SAAS,OAAO,CACd,OAA8B,EAC9B,OAA+B;IAE/B,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC;IACvE,IAAI,EAAE,GAAG,CAAC,CAAC;IACX,OAAO,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC,EAAE,EAAE,EAAE,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5D,EAAE,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;QAC1B,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC;IACpC,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/decode/sandhi-input.d.ts

@@ -0,0 +1,30 @@
+/**
+ * What the sandhi pass needs from a decoded run, beyond the syllables.
+ *
+ * Two questions sandhi cannot answer from a reading alone. Third-tone sandhi
+ * applies inside a prosodic foot, so it has to be told where the words and
+ * their parts are. 一 sandhi asks what stands in front of the 一, and the answer
+ * is a character rather than a spelling. Both are things the decode knows and
+ * the syllable array does not.
+ */
+import type { Dictionary } from "../dictionary/dictionary.js";
+import type { Syllable } from "../syllable/syllable.js";
+import type { SandhiGrouping } from "./sandhi.js";
+import type { ScoredWord } from "./word.js";
+/**
+ * Where the words and their constituents fall, for third-tone sandhi.
+ *
+ * Undefined unless third-tone sandhi was asked for, since dividing a word costs
+ * dictionary lookups and nothing else reads the answer.
+ */
+export declare function groupingOf(dictionary: Dictionary, words: readonly ScoredWord[], readings: readonly (readonly Syllable[])[], isThirdTone: boolean): SandhiGrouping | undefined;
+/**
+ * One 汉字 per syllable, where a word offers one.
+ *
+ * A word whose reading is a different length from its text — 玩儿 as `wánr` —
+ * has no character to give any one of its syllables, so its syllables get none
+ * and the pass falls back to their spellings. See
+ * {@link import("./sandhi-tones.js").isCounting}.
+ */
+export declare function charactersPerSyllable(words: readonly ScoredWord[], readings: readonly (readonly Syllable[])[]): readonly (string | undefined)[];
+//# sourceMappingURL=sandhi-input.d.ts.map

package/dist/decode/sandhi-input.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandhi-input.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi-input.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAE9D,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AAExD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAE5C;;;;;GAKG;AACH,wBAAgB,UAAU,CACxB,UAAU,EAAE,UAAU,EACtB,KAAK,EAAE,SAAS,UAAU,EAAE,EAC5B,QAAQ,EAAE,SAAS,CAAC,SAAS,QAAQ,EAAE,CAAC,EAAE,EAC1C,WAAW,EAAE,OAAO,GACnB,cAAc,GAAG,SAAS,CAQ5B;AAED;;;;;;;GAOG;AACH,wBAAgB,qBAAqB,CACnC,KAAK,EAAE,SAAS,UAAU,EAAE,EAC5B,QAAQ,EAAE,SAAS,CAAC,SAAS,QAAQ,EAAE,CAAC,EAAE,GACzC,SAAS,CAAC,MAAM,GAAG,SAAS,CAAC,EAAE,CAQjC"}

package/dist/decode/sandhi-input.js

@@ -0,0 +1,35 @@
+import { toCharacters } from "../script/characters.js";
+import { divisionOf } from "./constituents.js";
+/**
+ * Where the words and their constituents fall, for third-tone sandhi.
+ *
+ * Undefined unless third-tone sandhi was asked for, since dividing a word costs
+ * dictionary lookups and nothing else reads the answer.
+ */
+export function groupingOf(dictionary, words, readings, isThirdTone) {
+    if (!isThirdTone) {
+        return undefined;
+    }
+    return words.map((scored, index) => {
+        const reading = readings[index] ?? [];
+        return divisionOf(dictionary, scored.word.text, reading) ?? reading.length;
+    });
+}
+/**
+ * One 汉字 per syllable, where a word offers one.
+ *
+ * A word whose reading is a different length from its text — 玩儿 as `wánr` —
+ * has no character to give any one of its syllables, so its syllables get none
+ * and the pass falls back to their spellings. See
+ * {@link import("./sandhi-tones.js").isCounting}.
+ */
+export function charactersPerSyllable(words, readings) {
+    return words.flatMap((scored, index) => {
+        const reading = readings[index] ?? [];
+        const held = toCharacters(scored.word.text);
+        return held.length === reading.length
+            ? held
+            : Array.from({ length: reading.length });
+    });
+}
+//# sourceMappingURL=sandhi-input.js.map

package/dist/decode/sandhi-input.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandhi-input.js","sourceRoot":"","sources":["../../src/decode/sandhi-input.ts"],"names":[],"mappings":"AAUA,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAEvD,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAI/C;;;;;GAKG;AACH,MAAM,UAAU,UAAU,CACxB,UAAsB,EACtB,KAA4B,EAC5B,QAA0C,EAC1C,WAAoB;IAEpB,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE;QACjC,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;QACtC,OAAO,UAAU,CAAC,UAAU,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,CAAC;IAC7E,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,qBAAqB,CACnC,KAA4B,EAC5B,QAA0C;IAE1C,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,KAAK,EAAmC,EAAE;QACtE,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;QACtC,MAAM,IAAI,GAAG,YAAY,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5C,OAAO,IAAI,CAAC,MAAM,KAAK,OAAO,CAAC,MAAM;YACnC,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,KAAK,CAAC,IAAI,CAAqB,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IACjE,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/decode/sandhi-tones.d.ts

@@ -0,0 +1,97 @@
+/**
+ * Which tone 一 and 不 take, and when a 一 is being counted rather than said.
+ *
+ * Both characters change tone by what follows them, and 一 additionally does
+ * not change at all when it is a number being counted off.
+ */
+import { type Syllable } from "../syllable/syllable.js";
+import { type Tone } from "../tone/tone.js";
+/**
+ * Whether a syllable is 一, which is the only thing spelled `yi` that sandhis.
+ */
+export declare function isYi(syllable: Syllable): boolean;
+/**
+ * Whether a syllable is 不.
+ */
+export declare function isBu(syllable: Syllable): boolean;
+/**
+ * The tone 一 takes before a given following tone.
+ *
+ * Fourth tone flattens it to second; anything else raises it to fourth. Before
+ * a neutral tone or at the end of a phrase it keeps its citation first tone,
+ * since there is nothing to assimilate to.
+ */
+export declare function yiToneBefore(following: Tone | undefined): Tone;
+/**
+ * The numeral words a 一 can be the last digit of.
+ *
+ * Toneless, because what is being asked is which word it is rather than what
+ * tone it is carrying, and 十 is `shí` in every context it appears in here.
+ *
+ * **亿 is deliberately absent**, and that is measured rather than assumed: no
+ * 一 in 88,866 lines of Tatoeba and zh.wikipedia ends a number in 亿, while 意,
+ * 议, 义 and 議 all read `yì`, all precede a 一 that really is counting, and
+ * would all lose their sandhi for it — 11 conversions against nothing gained.
+ * The other three stay, because a number can end 百一 or 千一 even though this
+ * corpus has none, and the one thing they cost is 擺一個.
+ */
+export declare const NUMERAL_WORDS: Set<string>;
+/**
+ * The words that carry a number on past a 一, so that it is still counting.
+ *
+ * 一百一十's second 一 counts the ten it is followed by and takes the sandhi;
+ * 二十一's does not, because nothing numeric comes after it.
+ */
+export declare const CONTINUES_NUMBER: Set<string>;
+/**
+ * The ordinal prefix 第.
+ */
+export declare const ORDINAL_PREFIX = "di";
+/**
+ * The 汉字 {@link NUMERAL_WORDS} names, for a caller that has them.
+ *
+ * The spellings are ambiguous and the characters are not, which is the whole
+ * point of carrying both. 是 is spelt `shi` and stands in front of a 一 in
+ * 那是一条狗, 这是一个好主意 and several hundred sentences like them, and every
+ * one of them lost its sandhi to a 十 that was never there. 时, 事, 使 and 试 do
+ * the same.
+ *
+ * The 繁體 forms are here because a 繁體 text reaches this pass with 繁體
+ * characters. 拾, 佰 and 仟 are the 大写 banking forms, which appear on cheques
+ * and in contracts and count exactly as their everyday spellings do.
+ */
+export declare const NUMERAL_CHARACTERS: Set<string>;
+/**
+ * The 汉字 {@link CONTINUES_NUMBER} names.
+ */
+export declare const CONTINUES_NUMBER_CHARACTERS: Set<string>;
+/**
+ * The ordinal prefix as it is written.
+ */
+export declare const ORDINAL_CHARACTER = "\u7B2C";
+/**
+ * Whether a 一 is counting something, which is the only time it takes sandhi.
+ *
+ * The rule 一 sandhi is usually stated with — fourth tone flattens it, anything
+ * else raises it — is about the 一 that *counts*. Two shapes of 一 are not
+ * counting and keep the citation tone, and both are standard:
+ *
+ * - **The last digit of a larger number.** 十一月 is `shíyīyuè` and 二十一岁 is
+ *   `èrshíyī suì`. The signal is a numeral word before it with nothing numeric
+ *   after, which is what leaves 一百一十's middle 一 alone: 十 follows it, so it
+ *   is counting the ten.
+ * - **An ordinal.** 第一个 is `dìyī gè`.
+ *
+ * **Answered from the characters wherever the caller has them**, and from the
+ * spellings where it does not. `pinyinjs sandhi shíyī gè` is handed nothing but
+ * pinyin, so there the pass still cannot tell 十 from 时 or 第 from 地 and
+ * `docs/sandhi/` carries what that costs. A conversion knows which 汉字 it read,
+ * and passing them along is what stops 那是一条狗 losing its sandhi to the 是 in
+ * front of the 一.
+ *
+ * The two sides are decided separately, because a run can know one and not the
+ * other: a word whose reading has a different syllable count from its character
+ * count has no character to offer for any one syllable.
+ */
+export declare function isCounting(syllables: readonly Syllable[], at: number, characters?: readonly (string | undefined)[]): boolean;
+//# sourceMappingURL=sandhi-tones.d.ts.map

package/dist/decode/sandhi-tones.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandhi-tones.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi-tones.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,EAAE,KAAK,QAAQ,EAAyB,MAAM,yBAAyB,CAAC;AAC/E,OAAO,EAAgB,KAAK,IAAI,EAAE,MAAM,iBAAiB,CAAC;AAE1D;;GAEG;AACH,wBAAgB,IAAI,CAAC,QAAQ,EAAE,QAAQ,GAAG,OAAO,CAEhD;AAED;;GAEG;AACH,wBAAgB,IAAI,CAAC,QAAQ,EAAE,QAAQ,GAAG,OAAO,CAEhD;AAED;;;;;;GAMG;AACH,wBAAgB,YAAY,CAAC,SAAS,EAAE,IAAI,GAAG,SAAS,GAAG,IAAI,CAK9D;AAED;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,aAAa,aAAyC,CAAC;AAEpE;;;;;GAKG;AACH,eAAO,MAAM,gBAAgB,aAAsC,CAAC;AAEpE;;GAEG;AACH,eAAO,MAAM,cAAc,OAAO,CAAC;AAEnC;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,kBAAkB,aAS7B,CAAC;AAEH;;GAEG;AACH,eAAO,MAAM,2BAA2B,aAItC,CAAC;AAEH;;GAEG;AACH,eAAO,MAAM,iBAAiB,WAAM,CAAC;AAErC;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAgB,UAAU,CACxB,SAAS,EAAE,SAAS,QAAQ,EAAE,EAC9B,EAAE,EAAE,MAAM,EACV,UAAU,GAAE,SAAS,CAAC,MAAM,GAAG,SAAS,CAAC,EAAO,GAC/C,OAAO,CAwBT"}

package/dist/decode/sandhi-tones.js

@@ -0,0 +1,139 @@
+/**
+ * Which tone 一 and 不 take, and when a 一 is being counted rather than said.
+ *
+ * Both characters change tone by what follows them, and 一 additionally does
+ * not change at all when it is a number being counted off.
+ */
+import { writeSyllableSpelling } from "../syllable/syllable.js";
+import { NEUTRAL_TONE } from "../tone/tone.js";
+/**
+ * Whether a syllable is 一, which is the only thing spelled `yi` that sandhis.
+ */
+export function isYi(syllable) {
+    return syllable.initial === "" && syllable.final === "i";
+}
+/**
+ * Whether a syllable is 不.
+ */
+export function isBu(syllable) {
+    return syllable.initial === "b" && syllable.final === "u";
+}
+/**
+ * The tone 一 takes before a given following tone.
+ *
+ * Fourth tone flattens it to second; anything else raises it to fourth. Before
+ * a neutral tone or at the end of a phrase it keeps its citation first tone,
+ * since there is nothing to assimilate to.
+ */
+export function yiToneBefore(following) {
+    if (following === undefined || following === NEUTRAL_TONE) {
+        return 1;
+    }
+    return following === 4 ? 2 : 4;
+}
+/**
+ * The numeral words a 一 can be the last digit of.
+ *
+ * Toneless, because what is being asked is which word it is rather than what
+ * tone it is carrying, and 十 is `shí` in every context it appears in here.
+ *
+ * **亿 is deliberately absent**, and that is measured rather than assumed: no
+ * 一 in 88,866 lines of Tatoeba and zh.wikipedia ends a number in 亿, while 意,
+ * 议, 义 and 議 all read `yì`, all precede a 一 that really is counting, and
+ * would all lose their sandhi for it — 11 conversions against nothing gained.
+ * The other three stay, because a number can end 百一 or 千一 even though this
+ * corpus has none, and the one thing they cost is 擺一個.
+ */
+export const NUMERAL_WORDS = new Set(["shi", "bai", "qian", "wan"]);
+/**
+ * The words that carry a number on past a 一, so that it is still counting.
+ *
+ * 一百一十's second 一 counts the ten it is followed by and takes the sandhi;
+ * 二十一's does not, because nothing numeric comes after it.
+ */
+export const CONTINUES_NUMBER = new Set([...NUMERAL_WORDS, "ling"]);
+/**
+ * The ordinal prefix 第.
+ */
+export const ORDINAL_PREFIX = "di";
+/**
+ * The 汉字 {@link NUMERAL_WORDS} names, for a caller that has them.
+ *
+ * The spellings are ambiguous and the characters are not, which is the whole
+ * point of carrying both. 是 is spelt `shi` and stands in front of a 一 in
+ * 那是一条狗, 这是一个好主意 and several hundred sentences like them, and every
+ * one of them lost its sandhi to a 十 that was never there. 时, 事, 使 and 试 do
+ * the same.
+ *
+ * The 繁體 forms are here because a 繁體 text reaches this pass with 繁體
+ * characters. 拾, 佰 and 仟 are the 大写 banking forms, which appear on cheques
+ * and in contracts and count exactly as their everyday spellings do.
+ */
+export const NUMERAL_CHARACTERS = new Set([
+    "十",
+    "百",
+    "千",
+    "万",
+    "萬",
+    "拾",
+    "佰",
+    "仟",
+]);
+/**
+ * The 汉字 {@link CONTINUES_NUMBER} names.
+ */
+export const CONTINUES_NUMBER_CHARACTERS = new Set([
+    ...NUMERAL_CHARACTERS,
+    "零",
+    "〇",
+]);
+/**
+ * The ordinal prefix as it is written.
+ */
+export const ORDINAL_CHARACTER = "第";
+/**
+ * Whether a 一 is counting something, which is the only time it takes sandhi.
+ *
+ * The rule 一 sandhi is usually stated with — fourth tone flattens it, anything
+ * else raises it — is about the 一 that *counts*. Two shapes of 一 are not
+ * counting and keep the citation tone, and both are standard:
+ *
+ * - **The last digit of a larger number.** 十一月 is `shíyīyuè` and 二十一岁 is
+ *   `èrshíyī suì`. The signal is a numeral word before it with nothing numeric
+ *   after, which is what leaves 一百一十's middle 一 alone: 十 follows it, so it
+ *   is counting the ten.
+ * - **An ordinal.** 第一个 is `dìyī gè`.
+ *
+ * **Answered from the characters wherever the caller has them**, and from the
+ * spellings where it does not. `pinyinjs sandhi shíyī gè` is handed nothing but
+ * pinyin, so there the pass still cannot tell 十 from 时 or 第 from 地 and
+ * `docs/sandhi/` carries what that costs. A conversion knows which 汉字 it read,
+ * and passing them along is what stops 那是一条狗 losing its sandhi to the 是 in
+ * front of the 一.
+ *
+ * The two sides are decided separately, because a run can know one and not the
+ * other: a word whose reading has a different syllable count from its character
+ * count has no character to offer for any one syllable.
+ */
+export function isCounting(syllables, at, characters = []) {
+    const spelt = (index) => {
+        const syllable = syllables[index];
+        return syllable === undefined ? "" : writeSyllableSpelling(syllable);
+    };
+    const before = characters[at - 1];
+    if (before === undefined
+        ? spelt(at - 1) === ORDINAL_PREFIX
+        : before === ORDINAL_CHARACTER) {
+        return false;
+    }
+    if (before === undefined
+        ? !NUMERAL_WORDS.has(spelt(at - 1))
+        : !NUMERAL_CHARACTERS.has(before)) {
+        return true;
+    }
+    const after = characters[at + 1];
+    return after === undefined
+        ? CONTINUES_NUMBER.has(spelt(at + 1))
+        : CONTINUES_NUMBER_CHARACTERS.has(after);
+}
+//# sourceMappingURL=sandhi-tones.js.map

package/dist/decode/sandhi-tones.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandhi-tones.js","sourceRoot":"","sources":["../../src/decode/sandhi-tones.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,EAAiB,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAC/E,OAAO,EAAE,YAAY,EAAa,MAAM,iBAAiB,CAAC;AAE1D;;GAEG;AACH,MAAM,UAAU,IAAI,CAAC,QAAkB;IACrC,OAAO,QAAQ,CAAC,OAAO,KAAK,EAAE,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC3D,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,IAAI,CAAC,QAAkB;IACrC,OAAO,QAAQ,CAAC,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC5D,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,SAA2B;IACtD,IAAI,SAAS,KAAK,SAAS,IAAI,SAAS,KAAK,YAAY,EAAE,CAAC;QAC1D,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACjC,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC;AAEpE;;;;;GAKG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;AAEpE;;GAEG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC;AAEnC;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IACxC,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;CACJ,CAAC,CAAC;AAEH;;GAEG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,GAAG,CAAC;IACjD,GAAG,kBAAkB;IACrB,GAAG;IACH,GAAG;CACJ,CAAC,CAAC;AAEH;;GAEG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAErC;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,UAAU,UAAU,CACxB,SAA8B,EAC9B,EAAU,EACV,UAAU,GAAoC,EAAE;IAEhD,MAAM,KAAK,GAAG,CAAC,KAAa,EAAU,EAAE;QACtC,MAAM,QAAQ,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAClC,OAAO,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,qBAAqB,CAAC,QAAQ,CAAC,CAAC;IACvE,CAAC,CAAC;IACF,MAAM,MAAM,GAAG,UAAU,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;IAClC,IACE,MAAM,KAAK,SAAS;QAClB,CAAC,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,KAAK,cAAc;QAClC,CAAC,CAAC,MAAM,KAAK,iBAAiB,EAChC,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IACE,MAAM,KAAK,SAAS;QAClB,CAAC,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QACnC,CAAC,CAAC,CAAC,kBAAkB,CAAC,GAAG,CAAC,MAAM,CAAC,EACnC,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,KAAK,GAAG,UAAU,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;IACjC,OAAO,KAAK,KAAK,SAAS;QACxB,CAAC,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QACrC,CAAC,CAAC,2BAA2B,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;AAC7C,CAAC"}

package/dist/decode/sandhi.d.ts

@@ -1,4 +1,4 @@
-import { type Syllable } from "../syllable/syllable.js";
+import type { Syllable } from "../syllable/syllable.js";
 /**
  * How much of the sandhi to apply.
  */
@@ -39,14 +39,17 @@ export type SandhiGrouping = readonly (number | readonly number[])[];
  * contextual tone is put back, over the whole syllable array rather than within
  * a word, which is what lets 不 in one word assimilate to a tone in the next.
  *
- * 一 and 不 assimilate to whatever follows them and need nothing but the
- * syllables. Third-tone sandhi is not like that — its domain is the foot rather
- * than the syllable string — so `grouping` says where the words and their parts
- * are. Without one the whole array is taken for a single word, which is what a
- * caller holding nothing but a reading has.
+ * 不 assimilates to whatever follows it and needs nothing but the syllables.
+ * Third-tone sandhi is not like that — its domain is the foot rather than the
+ * syllable string — so `grouping` says where the words and their parts are.
+ * Without one the whole array is taken for a single word, which is what a
+ * caller holding nothing but a reading has. 一 sits between the two: the tone it
+ * takes is settled by the syllables, and whether it takes one at all is a
+ * question about the 汉字, so `characters` carries them where the caller has
+ * them. See {@link isCounting}.
  *
  * Never operates on a string. The old project patched output text with regexes,
  * which is what made its rules order-dependent and untestable.
  */
-export declare function applySandhi(syllables: readonly Syllable[], options?: SandhiOptions, grouping?: SandhiGrouping): readonly Syllable[];
+export declare function applySandhi(syllables: readonly Syllable[], options?: SandhiOptions, grouping?: SandhiGrouping, characters?: readonly (string | undefined)[]): readonly Syllable[];
 //# sourceMappingURL=sandhi.d.ts.map

package/dist/decode/sandhi.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"sandhi.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,QAAQ,EAAyB,MAAM,yBAAyB,CAAC;AAG/E;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IACxB;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;CAC9B;AAED;;;;;;;;;GASG;AACH,MAAM,MAAM,cAAc,GAAG,SAAS,CAAC,MAAM,GAAG,SAAS,MAAM,EAAE,CAAC,EAAE,CAAC;AAmNrE;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,SAAS,QAAQ,EAAE,EAC9B,OAAO,GAAE,aAAkB,EAC3B,QAAQ,CAAC,EAAE,cAAc,GACxB,SAAS,QAAQ,EAAE,CA6BrB"}
+{"version":3,"file":"sandhi.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AAExD;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IACxB;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;CAC9B;AAED;;;;;;;;;GASG;AACH,MAAM,MAAM,cAAc,GAAG,SAAS,CAAC,MAAM,GAAG,SAAS,MAAM,EAAE,CAAC,EAAE,CAAC;AA6HrE;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,SAAS,QAAQ,EAAE,EAC9B,OAAO,GAAE,aAAkB,EAC3B,QAAQ,CAAC,EAAE,cAAc,EACzB,UAAU,CAAC,EAAE,SAAS,CAAC,MAAM,GAAG,SAAS,CAAC,EAAE,GAC3C,SAAS,QAAQ,EAAE,CA6BrB"}

package/dist/decode/sandhi.js

@@ -1,84 +1,4 @@
-import { writeSyllableSpelling } from "../syllable/syllable.js";
-import { NEUTRAL_TONE } from "../tone/tone.js";
-/**
- * Whether a syllable is 一, which is the only thing spelled `yi` that sandhis.
- */
-function isYi(syllable) {
-    return syllable.initial === "" && syllable.final === "i";
-}
-/**
- * Whether a syllable is 不.
- */
-function isBu(syllable) {
-    return syllable.initial === "b" && syllable.final === "u";
-}
-/**
- * The tone 一 takes before a given following tone.
- *
- * Fourth tone flattens it to second; anything else raises it to fourth. Before
- * a neutral tone or at the end of a phrase it keeps its citation first tone,
- * since there is nothing to assimilate to.
- */
-function yiToneBefore(following) {
-    if (following === undefined || following === NEUTRAL_TONE) {
-        return 1;
-    }
-    return following === 4 ? 2 : 4;
-}
-/**
- * The numeral words a 一 can be the last digit of.
- *
- * Toneless, because what is being asked is which word it is rather than what
- * tone it is carrying, and 十 is `shí` in every context it appears in here.
- *
- * **亿 is deliberately absent**, and that is measured rather than assumed: no
- * 一 in 88,866 lines of Tatoeba and zh.wikipedia ends a number in 亿, while 意,
- * 议, 义 and 議 all read `yì`, all precede a 一 that really is counting, and
- * would all lose their sandhi for it — 11 conversions against nothing gained.
- * The other three stay, because a number can end 百一 or 千一 even though this
- * corpus has none, and the one thing they cost is 擺一個.
- */
-const NUMERAL_WORDS = new Set(["shi", "bai", "qian", "wan"]);
-/**
- * The words that carry a number on past a 一, so that it is still counting.
- *
- * 一百一十's second 一 counts the ten it is followed by and takes the sandhi;
- * 二十一's does not, because nothing numeric comes after it.
- */
-const CONTINUES_NUMBER = new Set([...NUMERAL_WORDS, "ling"]);
-/**
- * The ordinal prefix 第.
- */
-const ORDINAL_PREFIX = "di";
-/**
- * Whether a 一 is counting something, which is the only time it takes sandhi.
- *
- * The rule 一 sandhi is usually stated with — fourth tone flattens it, anything
- * else raises it — is about the 一 that *counts*. Two shapes of 一 are not
- * counting and keep the citation tone, and both are standard:
- *
- * - **The last digit of a larger number.** 十一月 is `shíyīyuè` and 二十一岁 is
- *   `èrshíyī suì`. The signal is a numeral word before it with nothing numeric
- *   after, which is what leaves 一百一十's middle 一 alone: 十 follows it, so it
- *   is counting the ten.
- * - **An ordinal.** 第一个 is `dìyī gè`.
- *
- * **Read off the syllables rather than the characters**, because that is all
- * this pass has: `pinyinjs sandhi shíyī gè` never sees a hanzi. So it cannot
- * tell 十 from 时 or 第 from 地, and `docs/sandhi/` carries what that costs —
- * measured, not estimated.
- */
-function isCounting(syllables, at) {
-    const spelt = (index) => {
-        const syllable = syllables[index];
-        return syllable === undefined ? "" : writeSyllableSpelling(syllable);
-    };
-    const before = spelt(at - 1);
-    if (before === ORDINAL_PREFIX) {
-        return false;
-    }
-    return !NUMERAL_WORDS.has(before) || CONTINUES_NUMBER.has(spelt(at + 1));
-}
+import { isBu, isCounting, isYi, yiToneBefore } from "./sandhi-tones.js";
 /**
  * Read a grouping as the junctions it puts in the syllable array.
  *
@@ -183,16 +103,19 @@ function applyThirdTone(syllables, grouping) {
  * contextual tone is put back, over the whole syllable array rather than within
  * a word, which is what lets 不 in one word assimilate to a tone in the next.
  *
- * 一 and 不 assimilate to whatever follows them and need nothing but the
- * syllables. Third-tone sandhi is not like that — its domain is the foot rather
- * than the syllable string — so `grouping` says where the words and their parts
- * are. Without one the whole array is taken for a single word, which is what a
- * caller holding nothing but a reading has.
+ * 不 assimilates to whatever follows it and needs nothing but the syllables.
+ * Third-tone sandhi is not like that — its domain is the foot rather than the
+ * syllable string — so `grouping` says where the words and their parts are.
+ * Without one the whole array is taken for a single word, which is what a
+ * caller holding nothing but a reading has. 一 sits between the two: the tone it
+ * takes is settled by the syllables, and whether it takes one at all is a
+ * question about the 汉字, so `characters` carries them where the caller has
+ * them. See {@link isCounting}.
  *
  * Never operates on a string. The old project patched output text with regexes,
  * which is what made its rules order-dependent and untestable.
  */
-export function applySandhi(syllables, options = {}, grouping) {
+export function applySandhi(syllables, options = {}, grouping, characters) {
     const { yiBu = true, thirdTone = false } = options;
     const applied = thirdTone
         ? applyThirdTone(syllables, grouping)
@@ -202,7 +125,7 @@ export function applySandhi(syllables, options = {}, grouping) {
         if (yiBu && isYi(syllable) && syllable.tone === 1) {
             // A 一 that is not counting keeps its citation tone, so it is left
             // exactly as the dictionary stored it.
-            if (isCounting(applied, at)) {
+            if (isCounting(applied, at, characters)) {
                 applied[at] = { ...syllable, tone: yiToneBefore(following?.tone) };
             }
             continue;