@kensio/pinyinjs 1.7.4 → 1.7.6

package/dist/decode/chang-rule.d.ts

@@ -0,0 +1,37 @@
+import { type EdgeRule } from "./rules.js";
+/**
+ * 长 read as `cháng` where an adverb of degree measures it.
+ *
+ * The character is stored `zhǎng` with `cháng` as an alternate, and the data
+ * says so about the character in isolation: Unihan ranks the readings
+ * `zhǎng(1879) cháng(1179)` by corpus count and names `zhǎng` in `kMandarin`.
+ * That ranking is doing real work — 署长, 团长, 公安局长, 总会长, 审理长 and
+ * the rest reach a bare 长 at the end of a title and read it correctly — so the
+ * default is not the thing to change. What is left over is the 长 that is an
+ * adjective, and nothing in the cost model can prefer an alternate at a
+ * position no word covers: 这篇文章不太长 came out `bú tài zhǎng`.
+ *
+ * Three contexts settle it, and the first of them needs only one side. A 长
+ * that is *growing* is a verb, and a degree adverb cannot modify one — there is
+ * no 很长 meaning it grew a lot — so 很, 太, 最, 多 and their kin in front of a
+ * bare 长 make it the adjective whatever follows. {@link DEGREE} is that set.
+ *
+ * The other two need both sides, and they are in `chang-context.ts`:
+ * {@link isPredicated} for 那座桥不长, {@link isAttributive} for 他有长头发.
+ * What they have in common is that the near side alone would be wrong. 不长 is
+ * 不长在耳朵外面 as readily as it is 不长, and a noun after a 长 is 长知识 as
+ * readily as 长头发.
+ *
+ * Measured over the same 88,866 lines of Tatoeba and zh.wikipedia the other
+ * rules were sized against, 260 长 decode as a word of their own. The degree
+ * adverbs move 75 of them to `cháng`, all 75 correctly; the two contexts added
+ * since move 8 more, also all correct — 那座桥不长, 生命不长, 射程不长, 还长,
+ * 生得又瘦又长, 他有长头发, 兔子有长耳朵 and 那位长胡子德国人. On CPP's 40
+ * hand-labelled 长 the character stands at 90.00%, against 85.00% with no rule
+ * and 87.50% with the degree adverbs alone.
+ *
+ * The last piece is {@link GROWN_INTO}, which pushes the opposite way on the one
+ * shape the sources get wrong.
+ */
+export declare const ADJECTIVAL_CHANG: EdgeRule;
+//# sourceMappingURL=chang-rule.d.ts.map

package/dist/decode/chang-rule.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"chang-rule.d.ts","sourceRoot":"","sources":["../../src/decode/chang-rule.ts"],"names":[],"mappings":"AASA,OAAO,EAAoB,KAAK,QAAQ,EAAgB,MAAM,YAAY,CAAC;AA2H3E;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,eAAO,MAAM,gBAAgB,EAAE,QAmB9B,CAAC"}

package/dist/decode/chang-rule.js

@@ -0,0 +1,169 @@
+import { isAttributive, isPredicated } from "./chang-context.js";
+import { wordEndingAt } from "./rules.js";
+/**
+ * Whether a syllable is `cháng`, the stative reading of 长.
+ */
+function isChang(syllable) {
+    return (syllable?.initial === "ch" &&
+        syllable.final === "ang" &&
+        syllable.tone === 2);
+}
+/**
+ * Adverbs that measure a quality rather than modify an action.
+ *
+ * The discriminating side of the 长 context, and the reason the set is written
+ * out rather than taken from a tag: jieba calls 很 `zg` and 多 `m`, so the two
+ * commonest members are not adverbs as far as the dictionary is concerned,
+ * while 已经, 也, 却 and 就 — which sit in front of the *verb* 长 — are `d`
+ * along with the ones wanted here. The tag names the wrong set in both
+ * directions, so the set is named directly.
+ *
+ * Everything here is degree or comparison. 还 and 不 are deliberately absent
+ * even though 距离还长 and 那座桥不长 are both `cháng`: they modify verbs as
+ * readily as qualities, and 还长有毛 and 胡子不长在前额上 are the same two
+ * characters around the verb.
+ */
+const DEGREE = new Set([
+    "一样",
+    "十分",
+    "多",
+    "多么",
+    "好",
+    "如此",
+    "尽量",
+    "很",
+    "格外",
+    "极",
+    "极其",
+    "比较",
+    "略",
+    "相当",
+    "真",
+    "稍",
+    "稍微",
+    "越",
+    "越来越",
+    "这么",
+    "这样",
+    "那么",
+    "那样",
+    "更",
+    "最",
+    "有多",
+    "特别",
+    "老",
+    "够",
+    "挺",
+    "太",
+    "非常",
+]);
+/**
+ * What shows a 长 to be growing even with a degree word in front of it.
+ *
+ * {@link DEGREE} already excludes nearly everything that precedes the verb, so
+ * this only has to settle the three shapes reachable from both sets: 真长得很快
+ * and 越长得快, where 得 marks the complement; 长着, where the aspect does; and
+ * 越长越高, where the correlative makes the 长 the thing that increases.
+ *
+ * 了 and 的 are pointedly *not* here, and that is the whole use of conditioning
+ * on the adverb first. Following a bare 长 either would suggest the verb — 长了
+ * 一个大包, 长的很像 — but following an adverb and a 长 they are the sentence
+ * particle and the attributive: 时间太长了 and 很长的道路 are both `cháng`, and
+ * an earlier draft of this rule that guarded on them read both as `zhǎng`.
+ */
+const GROWING = new Set(["得", "着", "越"]);
+/**
+ * What growing produces, as the far half of 越长越X.
+ *
+ * 越长越高, 越长越大, 越长越胖: each names a dimension that increases *because*
+ * something grew, so the 长 in front of it is the verb. The adjectives that go
+ * with a length or a duration are pointedly absent — 时间越长越好, 队伍越长越慢
+ * — because those are the cases the sources already read correctly.
+ */
+const GROWN_INTO = new Set(["高", "大", "胖", "壮", "结实"]);
+/**
+ * Whether an edge is the 越长 of a 越长越X saying something grew.
+ *
+ * This exists because of a defect in the sources rather than a gap in them.
+ * 越长 is a key, read `yuè cháng`, and it is the *only* one of its shape: 越大,
+ * 越高, 越好, 越快 and 越多 are all absent, so 越高越好 decodes as two words and
+ * 越长越高 reaches for a word nothing else in the paradigm has. It carries no
+ * part of speech, which is how the dictionary holds a reading somebody asserted
+ * rather than a word anybody counted — the same category {@link TEACHING_JIAO}
+ * declines to trust — and it comes from one source, `large_pinyin.txt`, with
+ * CC-CEDICT holding no such entry.
+ *
+ * Taking the edge away leaves the character's own default, which is `zhǎng`,
+ * and leaves `cháng` standing as a rival about one bucket dearer. That last
+ * part is the point of forbidding rather than forcing: 越长越X is genuinely
+ * ambiguous — 孩子越长越漂亮 grows and 头发越长越漂亮 lengthens, and only the
+ * subject says which — so the decode should answer with the likelier reading
+ * and still report itself as guessing. A forced edge could not.
+ *
+ * Unlike the rest of this file the shape has no corpus behind it: 越长 occurs
+ * three times in the 88,866 lines, and all three are 越来越长 or 说的越长. The
+ * word list is therefore small on purpose, and everything outside it is left as
+ * the sources have it.
+ */
+function isGrowthCorrelative(context) {
+    const { characters, edge } = context;
+    return (edge.text === "越长" &&
+        edge.partOfSpeech === "" &&
+        characters[edge.to] === "越" &&
+        GROWN_INTO.has(characters[edge.to + 1] ?? ""));
+}
+/**
+ * 长 read as `cháng` where an adverb of degree measures it.
+ *
+ * The character is stored `zhǎng` with `cháng` as an alternate, and the data
+ * says so about the character in isolation: Unihan ranks the readings
+ * `zhǎng(1879) cháng(1179)` by corpus count and names `zhǎng` in `kMandarin`.
+ * That ranking is doing real work — 署长, 团长, 公安局长, 总会长, 审理长 and
+ * the rest reach a bare 长 at the end of a title and read it correctly — so the
+ * default is not the thing to change. What is left over is the 长 that is an
+ * adjective, and nothing in the cost model can prefer an alternate at a
+ * position no word covers: 这篇文章不太长 came out `bú tài zhǎng`.
+ *
+ * Three contexts settle it, and the first of them needs only one side. A 长
+ * that is *growing* is a verb, and a degree adverb cannot modify one — there is
+ * no 很长 meaning it grew a lot — so 很, 太, 最, 多 and their kin in front of a
+ * bare 长 make it the adjective whatever follows. {@link DEGREE} is that set.
+ *
+ * The other two need both sides, and they are in `chang-context.ts`:
+ * {@link isPredicated} for 那座桥不长, {@link isAttributive} for 他有长头发.
+ * What they have in common is that the near side alone would be wrong. 不长 is
+ * 不长在耳朵外面 as readily as it is 不长, and a noun after a 长 is 长知识 as
+ * readily as 长头发.
+ *
+ * Measured over the same 88,866 lines of Tatoeba and zh.wikipedia the other
+ * rules were sized against, 260 长 decode as a word of their own. The degree
+ * adverbs move 75 of them to `cháng`, all 75 correctly; the two contexts added
+ * since move 8 more, also all correct — 那座桥不长, 生命不长, 射程不长, 还长,
+ * 生得又瘦又长, 他有长头发, 兔子有长耳朵 and 那位长胡子德国人. On CPP's 40
+ * hand-labelled 长 the character stands at 90.00%, against 85.00% with no rule
+ * and 87.50% with the degree adverbs alone.
+ *
+ * The last piece is {@link GROWN_INTO}, which pushes the opposite way on the one
+ * shape the sources get wrong.
+ */
+export const ADJECTIVAL_CHANG = {
+    name: "adjectival-chang",
+    verdictFor: (context) => {
+        const { characters, edge } = context;
+        if (isGrowthCorrelative(context)) {
+            return "forbid";
+        }
+        if (edge.text !== "长" || !isChang(edge.reading[0])) {
+            return "keep";
+        }
+        if (GROWING.has(characters[edge.to] ?? "")) {
+            return "keep";
+        }
+        if (isPredicated(context) || isAttributive(context)) {
+            return "force";
+        }
+        const before = wordEndingAt(context, edge.from);
+        return before !== undefined && DEGREE.has(before) ? "force" : "keep";
+    },
+};
+//# sourceMappingURL=chang-rule.js.map

package/dist/decode/chang-rule.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"chang-rule.js","sourceRoot":"","sources":["../../src/decode/chang-rule.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACjE,OAAO,EAAmC,YAAY,EAAE,MAAM,YAAY,CAAC;AAE3E;;GAEG;AACH,SAAS,OAAO,CAAC,QAA8B;IAC7C,OAAO,CACL,QAAQ,EAAE,OAAO,KAAK,IAAI;QAC1B,QAAQ,CAAC,KAAK,KAAK,KAAK;QACxB,QAAQ,CAAC,IAAI,KAAK,CAAC,CACpB,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC;IACrB,IAAI;IACJ,IAAI;IACJ,GAAG;IACH,IAAI;IACJ,GAAG;IACH,IAAI;IACJ,IAAI;IACJ,GAAG;IACH,IAAI;IACJ,GAAG;IACH,IAAI;IACJ,IAAI;IACJ,GAAG;IACH,IAAI;IACJ,GAAG;IACH,GAAG;IACH,IAAI;IACJ,GAAG;IACH,KAAK;IACL,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,GAAG;IACH,GAAG;IACH,IAAI;IACJ,IAAI;IACJ,GAAG;IACH,GAAG;IACH,GAAG;IACH,GAAG;IACH,IAAI;CACL,CAAC,CAAC;AAEH;;;;;;;;;;;;;GAaG;AACH,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAEzC;;;;;;;GAOG;AACH,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC;AAEvD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,SAAS,mBAAmB,CAAC,OAAoB;IAC/C,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;IACrC,OAAO,CACL,IAAI,CAAC,IAAI,KAAK,IAAI;QAClB,IAAI,CAAC,YAAY,KAAK,EAAE;QACxB,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,GAAG;QAC3B,UAAU,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAC9C,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAa;IACxC,IAAI,EAAE,kBAAkB;IACxB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACrC,IAAI,mBAAmB,CAAC,OAAO,CAAC,EAAE,CAAC;YACjC,OAAO,QAAQ,CAAC;QAClB,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACnD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC;YAC3C,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,IAAI,YAAY,CAAC,OAAO,CAAC,IAAI,aAAa,CAAC,OAAO,CAAC,EAAE,CAAC;YACpD,OAAO,OAAO,CAAC;QACjB,CAAC;QACD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,OAAO,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC;IACvE,CAAC;CACF,CAAC"}

package/dist/decode/confidence.d.ts

@@ -52,9 +52,17 @@ export interface ScoredUnit extends ReadingUnit, ReadingConfidence {
  * only be had by rejecting a word the dictionary attests.
  *
  * **Measured** against the CPP dataset's 20,139 hand-labelled polyphones: a
- * reading this reports uncertain is wrong 27.2% of the time, against 4.5% for
- * one backed by a word and 1.5% for a locked one. It lands on 18.7% of the
- * syllables of everyday text and 13.2% of encyclopedic text — see ROADMAP.md.
+ * reading this reports uncertain is wrong 19.9% of the time, against 4.4% for
+ * one backed by a word and 1.5% for a locked one. It lands on 17.6% of the
+ * syllables of everyday text and 12.6% of encyclopedic text — see ROADMAP.md.
+ *
+ * What it is not is a graded score, and the threshold is not the lever it
+ * looks like. A character's second reading costs one `ALTERNATE_PENALTY`
+ * more than its first whatever the real odds are, so nearly every uncertain
+ * reading is uncertain by exactly that margin: moving the threshold anywhere
+ * between 1.5 and 6 moves the share of everyday text flagged from 16.5% to
+ * 17.6%. The flag says a word did not cover the character. It does not say the
+ * decode was close.
  */
 export declare function isUncertain(confidence: ReadingConfidence): boolean;
 /**

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

@@ -1 +1 @@
-{"version":3,"file":"confidence.d.ts","sourceRoot":"","sources":["../../src/decode/confidence.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AACxD,OAAO,EAEL,KAAK,OAAO,EAGb,MAAM,cAAc,CAAC;AACtB,OAAO,EAEL,KAAK,iBAAiB,EACtB,KAAK,WAAW,EAIjB,MAAM,cAAc,CAAC;AAGtB;;;;;;GAMG;AACH,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,OAAO,EAAE,SAAS,QAAQ,EAAE,CAAC;IACtC;;;;;;OAMG;IACH,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC;;;;;;OAMG;IACH,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;IAC3B,gDAAgD;IAChD,QAAQ,CAAC,YAAY,EAAE,SAAS,kBAAkB,EAAE,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,UAAW,SAAQ,WAAW,EAAE,iBAAiB;CAAG;AAErE;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,WAAW,CAAC,UAAU,EAAE,iBAAiB,GAAG,OAAO,CAElE;AAiLD;;;;;;;;;;;GAWG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,iBAAiB,GAC5B,SAAS,UAAU,EAAE,CAuBvB"}
+{"version":3,"file":"confidence.d.ts","sourceRoot":"","sources":["../../src/decode/confidence.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AACxD,OAAO,EAAa,KAAK,OAAO,EAAkB,MAAM,cAAc,CAAC;AACvE,OAAO,EAEL,KAAK,iBAAiB,EACtB,KAAK,WAAW,EAEjB,MAAM,cAAc,CAAC;AAGtB;;;;;;GAMG;AACH,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,OAAO,EAAE,SAAS,QAAQ,EAAE,CAAC;IACtC;;;;;;OAMG;IACH,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;CACvB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC;;;;;;OAMG;IACH,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;IAC3B,gDAAgD;IAChD,QAAQ,CAAC,YAAY,EAAE,SAAS,kBAAkB,EAAE,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,UAAW,SAAQ,WAAW,EAAE,iBAAiB;CAAG;AAErE;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,WAAW,CAAC,UAAU,EAAE,iBAAiB,GAAG,OAAO,CAElE;AACD;;;;;;;;;;;GAWG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,iBAAiB,GAC5B,SAAS,UAAU,EAAE,CAuBvB"}

package/dist/decode/confidence.js

@@ -1,6 +1,6 @@
-import { cutPoints, READING_CHARGE, } from "./lattice.js";
-import { isSettled, settledUnits, unitKey, unitsOf, } from "./locking.js";
-import { readingCost, shortestPath } from "./viterbi.js";
+import { cutPoints, READING_CHARGE } from "./lattice.js";
+import { isSettled, settledUnits, } from "./locking.js";
+import { scoreStretch } from "./alternatives.js";
 /**
  * Whether a reading was settled by something weaker than dictionary evidence.
  *
@@ -12,128 +12,21 @@ import { readingCost, shortestPath } from "./viterbi.js";
  * only be had by rejecting a word the dictionary attests.
  *
  * **Measured** against the CPP dataset's 20,139 hand-labelled polyphones: a
- * reading this reports uncertain is wrong 27.2% of the time, against 4.5% for
- * one backed by a word and 1.5% for a locked one. It lands on 18.7% of the
- * syllables of everyday text and 13.2% of encyclopedic text — see ROADMAP.md.
+ * reading this reports uncertain is wrong 19.9% of the time, against 4.4% for
+ * one backed by a word and 1.5% for a locked one. It lands on 17.6% of the
+ * syllables of everyday text and 12.6% of encyclopedic text — see ROADMAP.md.
+ *
+ * What it is not is a graded score, and the threshold is not the lever it
+ * looks like. A character's second reading costs one `ALTERNATE_PENALTY`
+ * more than its first whatever the real odds are, so nearly every uncertain
+ * reading is uncertain by exactly that margin: moving the threshold anywhere
+ * between 1.5 and 6 moves the share of everyday text flagged from 16.5% to
+ * 17.6%. The flag says a word did not cover the character. It does not say the
+ * decode was close.
  */
 export function isUncertain(confidence) {
     return (confidence.alternatives[0]?.cost ?? Infinity) < READING_CHARGE;
 }
-/**
- * The edges leaving one position.
- */
-function edgesAt(lattice, at) {
-    /* c8 ignore next -- at is always a position of the run */
-    return lattice.edges[at] ?? [];
-}
-/**
- * Every edge lying wholly inside a stretch.
- */
-function edgesInside(lattice, from, to) {
-    const inside = [];
-    for (let at = from; at < to; at++) {
-        inside.push(...edgesAt(lattice, at).filter((edge) => edge.to <= to));
-    }
-    return inside;
-}
-/**
- * Cheapest cost of reaching each position in a stretch from its start.
- */
-function forwardCosts(lattice, from, to, costOf) {
-    const best = Array.from({ length: to - from + 1 }, () => Infinity);
-    best[0] = 0;
-    for (let at = from; at < to; at++) {
-        /* c8 ignore next 2 -- every position of the stretch is reached and in range */
-        const sofar = best[at - from] ?? Infinity;
-        for (const edge of edgesAt(lattice, at)) {
-            /* c8 ignore next -- the target of an edge inside the stretch is in range */
-            const rival = best[edge.to - from] ?? Infinity;
-            if (edge.to <= to && sofar + costOf(edge) < rival) {
-                best[edge.to - from] = sofar + costOf(edge);
-            }
-        }
-    }
-    return best;
-}
-/**
- * Cheapest cost of reaching the end of a stretch from each position in it.
- */
-function backwardCosts(lattice, from, to, costOf) {
-    const best = Array.from({ length: to - from + 1 }, () => Infinity);
-    best[to - from] = 0;
-    for (let at = to - 1; at >= from; at--) {
-        let cheapest = Infinity;
-        for (const edge of edgesAt(lattice, at)) {
-            /* c8 ignore next -- the target of an edge inside the stretch is in range */
-            const onwards = best[edge.to - from] ?? Infinity;
-            if (edge.to <= to && costOf(edge) + onwards < cheapest) {
-                cheapest = costOf(edge) + onwards;
-            }
-        }
-        best[at - from] = cheapest;
-    }
-    return best;
-}
-/**
- * Price every distinct reading a stretch offers.
- *
- * A claim is priced by the cheapest path through the whole stretch that makes
- * it, which is what makes the figures comparable: a reading can be expensive
- * because it forces expensive readings around it as much as because it is
- * itself unlikely. One forward sweep and one backward sweep give that for every
- * edge at once, which is the same work the decode already does in one
- * direction.
- */
-function scoreClaims(lattice, from, to, costOf) {
-    const forward = forwardCosts(lattice, from, to, costOf);
-    const backward = backwardCosts(lattice, from, to, costOf);
-    const claims = new Map();
-    for (const edge of edgesInside(lattice, from, to)) {
-        /* c8 ignore next 2 -- both ends of an edge in the stretch are in range */
-        const through = (forward[edge.from - from] ?? Infinity) +
-            costOf(edge) +
-            (backward[edge.to - from] ?? Infinity);
-        for (const unit of unitsOf(edge)) {
-            const key = unitKey(unit);
-            const held = claims.get(key);
-            if (held === undefined || through < held.cost) {
-                claims.set(key, { unit, cost: through });
-            }
-        }
-    }
-    /* c8 ignore next -- every position of a stretch is reachable from its start */
-    return { claims, best: forward[to - from] ?? Infinity };
-}
-/**
- * What a unit beat: the claims overlapping its span, other than its own.
- */
-function alternativesTo(stretch, unit) {
-    const key = unitKey(unit);
-    return [...stretch.claims.values()]
-        .filter((claim) => unitKey(claim.unit) !== key &&
-        claim.unit.from < unit.to &&
-        unit.from < claim.unit.to)
-        .map((claim) => ({
-        from: claim.unit.from,
-        to: claim.unit.to,
-        reading: claim.unit.reading,
-        cost: claim.cost - stretch.best,
-    }))
-        .toSorted((left, right) => left.cost - right.cost);
-}
-/**
- * Decode one stretch's readings, keeping what the cheapest path rejected.
- */
-function scoreStretch(lattice, projection, from, to) {
-    const stretch = scoreClaims(lattice, from, to, readingCost);
-    return shortestPath(lattice, from, to, readingCost)
-        .flatMap((edge) => unitsOf(edge))
-        .map((unit) => ({
-        ...unit,
-        isLocked: projection.locked[unit.from] !== undefined,
-        alternatives: alternativesTo(stretch, unit),
-    }));
-}
 /**
  * Decode the run's readings, keeping what each choice rejected.
  *

package/dist/decode/confidence.js.map

@@ -1 +1 @@
-{"version":3,"file":"confidence.js","sourceRoot":"","sources":["../../src/decode/confidence.ts"],"names":[],"mappings":"AACA,OAAO,EACL,SAAS,EAGT,cAAc,GACf,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,SAAS,EAGT,YAAY,EACZ,OAAO,EACP,OAAO,GACR,MAAM,cAAc,CAAC;AACtB,OAAO,EAAe,WAAW,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AA4CtE;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,WAAW,CAAC,UAA6B;IACvD,OAAO,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,IAAI,QAAQ,CAAC,GAAG,cAAc,CAAC;AACzE,CAAC;AAkBD;;GAEG;AACH,SAAS,OAAO,CAAC,OAAgB,EAAE,EAAU;IAC3C,0DAA0D;IAC1D,OAAO,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;AACjC,CAAC;AAED;;GAEG;AACH,SAAS,WAAW,CAClB,OAAgB,EAChB,IAAY,EACZ,EAAU;IAEV,MAAM,MAAM,GAAkB,EAAE,CAAC;IACjC,KAAK,IAAI,EAAE,GAAG,IAAI,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC;QAClC,MAAM,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IACvE,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,SAAS,YAAY,CACnB,OAAgB,EAChB,IAAY,EACZ,EAAU,EACV,MAAc;IAEd,MAAM,IAAI,GAAa,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,EAAE,GAAG,IAAI,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,CAAC;IAC7E,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAEZ,KAAK,IAAI,EAAE,GAAG,IAAI,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,EAAE,EAAE,CAAC;QAClC,+EAA+E;QAC/E,MAAM,KAAK,GAAG,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,QAAQ,CAAC;QAC1C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC;YACxC,4EAA4E;YAC5E,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,QAAQ,CAAC;YAC/C,IAAI,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,GAAG,KAAK,EAAE,CAAC;gBAClD,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC;YAC9C,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;GAEG;AACH,SAAS,aAAa,CACpB,OAAgB,EAChB,IAAY,EACZ,EAAU,EACV,MAAc;IAEd,MAAM,IAAI,GAAa,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,EAAE,GAAG,IAAI,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,CAAC;IAC7E,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC;IAEpB,KAAK,IAAI,EAAE,GAAG,EAAE,GAAG,CAAC,EAAE,EAAE,IAAI,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC;QACvC,IAAI,QAAQ,GAAG,QAAQ,CAAC;QACxB,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC;YACxC,4EAA4E;YAC5E,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,QAAQ,CAAC;YACjD,IAAI,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,OAAO,GAAG,QAAQ,EAAE,CAAC;gBACvD,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC;YACpC,CAAC;QACH,CAAC;QACD,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,QAAQ,CAAC;IAC7B,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;;;;GASG;AACH,SAAS,WAAW,CAClB,OAAgB,EAChB,IAAY,EACZ,EAAU,EACV,MAAc;IAEd,MAAM,OAAO,GAAG,YAAY,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;IACxD,MAAM,QAAQ,GAAG,aAAa,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;IAC1D,MAAM,MAAM,GAAG,IAAI,GAAG,EAAuB,CAAC;IAE9C,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,CAAC;QAClD,0EAA0E;QAC1E,MAAM,OAAO,GACX,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,QAAQ,CAAC;YACvC,MAAM,CAAC,IAAI,CAAC;YACZ,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,QAAQ,CAAC,CAAC;QACzC,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC;YACjC,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;YAC1B,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC7B,IAAI,IAAI,KAAK,SAAS,IAAI,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;gBAC9C,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YAC3C,CAAC;QACH,CAAC;IACH,CAAC;IAED,+EAA+E;IAC/E,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,QAAQ,EAAE,CAAC;AAC1D,CAAC;AAED;;GAEG;AACH,SAAS,cAAc,CACrB,OAAsB,EACtB,IAAiB;IAEjB,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;SAChC,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG;QAC3B,KAAK,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,EAAE;QACzB,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,CAC5B;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QACf,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI;QACrB,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE;QACjB,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,OAAO;QAC3B,IAAI,EAAE,KAAK,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI;KAChC,CAAC,CAAC;SACF,QAAQ,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC;AACvD,CAAC;AAED;;GAEG;AACH,SAAS,YAAY,CACnB,OAAgB,EAChB,UAA6B,EAC7B,IAAY,EACZ,EAAU;IAEV,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,WAAW,CAAC,CAAC;IAC5D,OAAO,YAAY,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,WAAW,CAAC;SAChD,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;SAChC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACd,GAAG,IAAI;QACP,QAAQ,EAAE,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,SAAS;QACpD,YAAY,EAAE,cAAc,CAAC,OAAO,EAAE,IAAI,CAAC;KAC5C,CAAC,CAAC,CAAC;AACR,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,aAAa,CAC3B,OAAgB,EAChB,UAA6B;IAE7B,MAAM,IAAI,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC;IAChC,MAAM,KAAK,GAAiB,EAAE,CAAC;IAE/B,KAAK,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QAC3C,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC3B,IAAI,EAAE,KAAK,SAAS,EAAE,CAAC;YACrB,MAAM;QACR,CAAC;QACD,IAAI,SAAS,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CACR,GAAG,YAAY,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;gBACnD,GAAG,IAAI;gBACP,QAAQ,EAAE,IAAI;gBACd,YAAY,EAAE,EAAE;aACjB,CAAC,CAAC,CACJ,CAAC;YACF,SAAS;QACX,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;IAC7D,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}
+{"version":3,"file":"confidence.js","sourceRoot":"","sources":["../../src/decode/confidence.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,SAAS,EAAgB,cAAc,EAAE,MAAM,cAAc,CAAC;AACvE,OAAO,EACL,SAAS,EAGT,YAAY,GACb,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AA4CjD;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,MAAM,UAAU,WAAW,CAAC,UAA6B;IACvD,OAAO,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,IAAI,QAAQ,CAAC,GAAG,cAAc,CAAC;AACzE,CAAC;AACD;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,aAAa,CAC3B,OAAgB,EAChB,UAA6B;IAE7B,MAAM,IAAI,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC;IAChC,MAAM,KAAK,GAAiB,EAAE,CAAC;IAE/B,KAAK,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QAC3C,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC3B,IAAI,EAAE,KAAK,SAAS,EAAE,CAAC;YACrB,MAAM;QACR,CAAC;QACD,IAAI,SAAS,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CACR,GAAG,YAAY,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;gBACnD,GAAG,IAAI;gBACP,QAAQ,EAAE,IAAI;gBACd,YAAY,EAAE,EAAE;aACjB,CAAC,CAAC,CACJ,CAAC;YACF,SAAS;QACX,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;IAC7D,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}

package/dist/decode/convert-options.d.ts

@@ -0,0 +1,116 @@
+/**
+ * What a conversion is asked for, and the shape of one piece of its output.
+ *
+ * Everything here describes the conversion rather than performing any of it.
+ */
+import type { ApostropheStyle } from "../orthography/apostrophe.js";
+import type { CapitalStyle } from "../orthography/capitals.js";
+import type { PunctuationStyle } from "../orthography/punctuation.js";
+import type { Locale } from "../script/script.js";
+import type { Syllable, ToneNotation } from "../syllable/syllable.js";
+import type { ReadingConfidence } from "./confidence.js";
+import type { ReadingHints } from "./hints.js";
+import type { SandhiOptions } from "./sandhi.js";
+/**
+ * How a conversion should be carried out and written.
+ */
+export interface ConvertOptions {
+    /** Which reading standard to use. Defaults to `zh-CN`. */
+    readonly locale?: Locale;
+    /** How tones are written. Defaults to diacritics. */
+    readonly notation?: ToneNotation;
+    /** Which tone sandhi to apply. */
+    readonly sandhi?: SandhiOptions;
+    /** When the 隔音符号 is written. Defaults to `always`. */
+    readonly apostrophe?: ApostropheStyle;
+    /** Which capitals are written. Defaults to `auto`. */
+    readonly capitals?: CapitalStyle;
+    /** Whether Chinese punctuation is rewritten. Defaults to `latin`. */
+    readonly punctuation?: PunctuationStyle;
+    /** Whether GB/T 16159 word grouping is applied. Defaults to true. */
+    readonly grouping?: boolean;
+    /**
+     * What to do with the digits in a text. Defaults to `read`.
+     *
+     * `read` says them: 我有3个 is `wǒ yǒu sān gè` and 1998年 is
+     * `yī jiǔ jiǔ bā nián`. `keep` leaves every digit exactly as it was written,
+     * which is what this did before there was anything to read them with.
+     */
+    readonly numbers?: NumberStyle;
+    /**
+     * Readings the caller asserts, over whatever the sources say.
+     *
+     * No rule can settle every polyphone, and some texts are genuinely ambiguous
+     * — 孩子越长越漂亮 grows where 头发越长越漂亮 lengthens — so an application
+     * that knows its own content can say what this one could only guess at.
+     *
+     * ```ts
+     * convert(dictionary, "这篇文章不太长。", { readings: { 太长: "tài cháng" } });
+     * ```
+     *
+     * A hint displaces whatever covered exactly the characters it names, and
+     * competes normally with everything else, so a bare `长` hint leaves 校长 as
+     * `xiàozhǎng` — the longer word is still the better reading of that stretch.
+     * See {@link ReadingHints}.
+     */
+    readonly readings?: ReadingHints;
+}
+/**
+ * What a conversion does with the digits it meets.
+ */
+export type NumberStyle = "read" | "keep";
+/**
+ * One piece of a conversion: a syllable, or the text between two of them.
+ *
+ * A conversion is assembled piece by piece and joined at the very end, so that
+ * a caller wanting to render each syllable separately — colouring its tone,
+ * marking it uncertain — is not left trying to find the syllables again in a
+ * finished string. {@link convert} is this, joined.
+ */
+export interface ConvertedPiece {
+    readonly text: string;
+    /** The syllable this piece writes, or undefined where it writes none. */
+    readonly syllable: Syllable | undefined;
+    /**
+     * The characters this piece reads, where it names any of its own.
+     *
+     * What an annotation needs and a plain conversion throws away: to write 银行
+     * with `yín` over 银 and `háng` over 行, something has to say which characters
+     * each syllable is a reading *of*, and that is not the syllable's position.
+     * A reading is not one syllable per character — 玩儿 is two characters and
+     * the single syllable `wánr` — and a read number is not made of characters at
+     * all in the same way, 95% being `bǎifēnzhījiǔshíwǔ` over eight syllables and
+     * three written characters.
+     *
+     * So a piece either names the characters it reads or continues the ones the
+     * piece before it named, and the two are told apart by the syllable rather
+     * than by a second flag:
+     *
+     * | `syllable` | `source` | The piece |
+     * | --- | --- | --- |
+     * | undefined | undefined | writes no reading: a space, or a non-Han run |
+     * | set | set | reads exactly those characters |
+     * | set | undefined | reads on into the characters named before it |
+     *
+     * {@link import("../format/html.js").toAnnotatedHtml} is what consumes it.
+     */
+    readonly source: string | undefined;
+    /**
+     * How settled that syllable was, where the decode reported it.
+     *
+     * Only {@link convertPieces} fills this in, and only for a syllable the
+     * lattice decoded: the greedy baseline cannot say what it rejected, and a
+     * Taiwan reading that differs in length from its mainland form cannot be
+     * lined up with it.
+     */
+    readonly confidence: ReadingConfidence | undefined;
+}
+/**
+ * The orthographic choices a conversion has settled, rather than defaulted.
+ */
+export interface Written {
+    readonly notation: ToneNotation;
+    readonly apostrophe: ApostropheStyle;
+    readonly capitals: CapitalStyle;
+}
+//# sourceMappingURL=convert-options.d.ts.map

package/dist/decode/convert-options.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"convert-options.d.ts","sourceRoot":"","sources":["../../src/decode/convert-options.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,4BAA4B,CAAC;AAC/D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,+BAA+B,CAAC;AACtE,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,qBAAqB,CAAC;AAClD,OAAO,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AACzD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC/C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEjD;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,0DAA0D;IAC1D,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,qDAAqD;IACrD,QAAQ,CAAC,QAAQ,CAAC,EAAE,YAAY,CAAC;IACjC,kCAAkC;IAClC,QAAQ,CAAC,MAAM,CAAC,EAAE,aAAa,CAAC;IAChC,sDAAsD;IACtD,QAAQ,CAAC,UAAU,CAAC,EAAE,eAAe,CAAC;IACtC,sDAAsD;IACtD,QAAQ,CAAC,QAAQ,CAAC,EAAE,YAAY,CAAC;IACjC,qEAAqE;IACrE,QAAQ,CAAC,WAAW,CAAC,EAAE,gBAAgB,CAAC;IACxC,qEAAqE;IACrE,QAAQ,CAAC,QAAQ,CAAC,EAAE,OAAO,CAAC;IAC5B;;;;;;OAMG;IACH,QAAQ,CAAC,OAAO,CAAC,EAAE,WAAW,CAAC;IAC/B;;;;;;;;;;;;;;;OAeG;IACH,QAAQ,CAAC,QAAQ,CAAC,EAAE,YAAY,CAAC;CAClC;AAED;;GAEG;AACH,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,MAAM,CAAC;AAE1C;;;;;;;GAOG;AACH,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,yEAAyE;IACzE,QAAQ,CAAC,QAAQ,EAAE,QAAQ,GAAG,SAAS,CAAC;IACxC;;;;;;;;;;;;;;;;;;;;;;OAsBG;IACH,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IACpC;;;;;;;OAOG;IACH,QAAQ,CAAC,UAAU,EAAE,iBAAiB,GAAG,SAAS,CAAC;CACpD;AAED;;GAEG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,CAAC,QAAQ,EAAE,YAAY,CAAC;IAChC,QAAQ,CAAC,UAAU,EAAE,eAAe,CAAC;IACrC,QAAQ,CAAC,QAAQ,EAAE,YAAY,CAAC;CACjC"}

package/dist/decode/convert-options.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=convert-options.js.map

package/dist/decode/convert-options.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"convert-options.js","sourceRoot":"","sources":["../../src/decode/convert-options.ts"],"names":[],"mappings":""}

package/dist/decode/convert-pipeline.d.ts

@@ -0,0 +1,18 @@
+/**
+ * The conversion pipeline: runs in, written pieces out.
+ *
+ * One function, because the phases depend on each other in a fixed order and
+ * each needs what the one before it produced — the numbers are read before
+ * anything is decoded, the Han runs are decoded before anything is written, and
+ * the capitals and punctuation are applied over the whole text rather than over
+ * a run. Which decoder does the middle of it is the caller's to choose, which is
+ * what makes the six entry points in `convert.ts` six lines each.
+ */
+import type { Dictionary } from "../dictionary/dictionary.js";
+import type { ConvertedPiece, ConvertOptions } from "./pieces.js";
+import type { Decode } from "./decoders.js";
+/**
+ * Run the pipeline over a text with a given decoder.
+ */
+export declare function convertWith(decode: Decode, dictionary: Dictionary, text: string, options: ConvertOptions): readonly ConvertedPiece[];
+//# sourceMappingURL=convert-pipeline.d.ts.map

package/dist/decode/convert-pipeline.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"convert-pipeline.d.ts","sourceRoot":"","sources":["../../src/decode/convert-pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAU9D,OAAO,KAAK,EAAE,cAAc,EAAE,cAAc,EAAW,MAAM,aAAa,CAAC;AAE3E,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,eAAe,CAAC;AAI5C;;GAEG;AACH,wBAAgB,WAAW,CACzB,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,UAAU,EACtB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,cAAc,GACtB,SAAS,cAAc,EAAE,CAiF3B"}

package/dist/decode/convert-pipeline.js

@@ -0,0 +1,67 @@
+import { capitaliseSentenceParts, isSentence, } from "../orthography/capitals.js";
+import { toLatinPunctuationParts } from "../orthography/punctuation.js";
+import { readNumbersIn } from "../numerals/text.js";
+import { toCharacters } from "../script/characters.js";
+import { hintsWithin, resolveHints } from "./hints.js";
+import { splitRuns, surroundingCharacters } from "./runs.js";
+import { numeralBefore, surrounding, writeNumbers } from "./numbers.js";
+import { wordsOf } from "./run-words.js";
+import { writeRun, rewrite } from "./write-run.js";
+/**
+ * Run the pipeline over a text with a given decoder.
+ */
+export function convertWith(decode, dictionary, text, options) {
+    const { locale = "zh-CN", notation = "marks", apostrophe = "always", capitals = "auto", punctuation = "latin", grouping = true, numbers = "read", sandhi, readings, } = options;
+    const written = { notation, apostrophe, capitals };
+    const converted = [];
+    const runs = [...splitRuns(text)];
+    // Parsed once for the whole text rather than per run, so that a malformed
+    // hint is reported whether or not a run happens to reach it.
+    const hints = readings === undefined ? undefined : resolveHints(readings);
+    // Where each run starts, in code points, so a caller counting positions from
+    // the start of the text can be met where the run puts them.
+    const starts = [];
+    let start = 0;
+    for (const run of runs) {
+        starts.push(start);
+        start += toCharacters(run.text).length;
+    }
+    // Read before anything is decoded, because a number needs to know what
+    // follows it — 1998年 is a year and 3个 is a count — and the character that
+    // decides it is there for the reading without a decode of its own.
+    const read = runs.map((run, at) => run.isHan || numbers === "keep"
+        ? []
+        : readNumbersIn(run.text, surroundingCharacters(runs, at)));
+    // Decoded before anything is written, because a number's sandhi needs the
+    // syllable after it, and because a Han run needs the number in front of it:
+    // 2个人 is `liǎng gè rén` and 个人 on its own is the word `gèrén`.
+    const decoded = runs.map((run, at) => {
+        if (!run.isHan) {
+            return [];
+        }
+        const within = hints === undefined
+            ? undefined
+            : hintsWithin(hints, starts[at] ?? 0, toCharacters(run.text).length);
+        const said = numeralBefore(read[at - 1] ?? []);
+        return wordsOf(decode(dictionary, run.text, said, within), dictionary, grouping);
+    });
+    for (const [at, run] of runs.entries()) {
+        const words = decoded[at] ?? [];
+        if (run.isHan) {
+            converted.push(...writeRun(dictionary, words, locale, written, sandhi));
+            continue;
+        }
+        converted.push(...writeNumbers(run.text, read[at] ?? [], surrounding(runs, decoded, at), written, { sandhi }));
+    }
+    let pieces = converted;
+    // Both of these read the whole conversion rather than one run: a sentence
+    // capital belongs to whichever run happens to start the sentence, and a mark
+    // needs to know whether anything follows it before it takes a space.
+    if (capitals === "auto" && isSentence(text)) {
+        pieces = rewrite(pieces, capitaliseSentenceParts);
+    }
+    return punctuation === "latin"
+        ? rewrite(pieces, toLatinPunctuationParts)
+        : pieces;
+}
+//# sourceMappingURL=convert-pipeline.js.map

package/dist/decode/convert-pipeline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"convert-pipeline.js","sourceRoot":"","sources":["../../src/decode/convert-pipeline.ts"],"names":[],"mappings":"AAWA,OAAO,EACL,uBAAuB,EACvB,UAAU,GACX,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AACvD,OAAO,EAAE,SAAS,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC;AAE7D,OAAO,EAAE,aAAa,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAExE,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AAEnD;;GAEG;AACH,MAAM,UAAU,WAAW,CACzB,MAAc,EACd,UAAsB,EACtB,IAAY,EACZ,OAAuB;IAEvB,MAAM,EACJ,MAAM,GAAG,OAAO,EAChB,QAAQ,GAAG,OAAO,EAClB,UAAU,GAAG,QAAQ,EACrB,QAAQ,GAAG,MAAM,EACjB,WAAW,GAAG,OAAO,EACrB,QAAQ,GAAG,IAAI,EACf,OAAO,GAAG,MAAM,EAChB,MAAM,EACN,QAAQ,GACT,GAAG,OAAO,CAAC;IACZ,MAAM,OAAO,GAAY,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;IAC5D,MAAM,SAAS,GAAqB,EAAE,CAAC;IAEvC,MAAM,IAAI,GAAG,CAAC,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;IAClC,0EAA0E;IAC1E,6DAA6D;IAC7D,MAAM,KAAK,GAAG,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC1E,6EAA6E;IAC7E,4DAA4D;IAC5D,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnB,KAAK,IAAI,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;IACzC,CAAC;IACD,uEAAuE;IACvE,0EAA0E;IAC1E,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE,CAChC,GAAG,CAAC,KAAK,IAAI,OAAO,KAAK,MAAM;QAC7B,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,IAAI,EAAE,qBAAqB,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAC7D,CAAC;IACF,0EAA0E;IAC1E,4EAA4E;IAC5E,+DAA+D;IAC/D,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE;QACnC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;YACf,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,MAAM,MAAM,GACV,KAAK,KAAK,SAAS;YACjB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC;QACzE,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QAC/C,OAAO,OAAO,CACZ,MAAM,CAAC,UAAU,EAAE,GAAG,CAAC,IAAI,EAAE,IAAI,EAAE,MAAM,CAAC,EAC1C,UAAU,EACV,QAAQ,CACT,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,KAAK,MAAM,CAAC,EAAE,EAAE,GAAG,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;QAChC,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;YACd,SAAS,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC;YACxE,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CACZ,GAAG,YAAY,CACb,GAAG,CAAC,IAAI,EACR,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,EACd,WAAW,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,EAC9B,OAAO,EACP,EAAE,MAAM,EAAE,CACX,CACF,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAA8B,SAAS,CAAC;IAElD,0EAA0E;IAC1E,6EAA6E;IAC7E,qEAAqE;IACrE,IAAI,QAAQ,KAAK,MAAM,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QAC5C,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC,CAAC;IACpD,CAAC;IACD,OAAO,WAAW,KAAK,OAAO;QAC5B,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC;QAC1C,CAAC,CAAC,MAAM,CAAC;AACb,CAAC"}