@kensio/pinyinjs 1.2.0 → 1.2.2

package/dist/decode/reading-rules.js

@@ -1,3 +1,5 @@
+import { QUANTITY_CHARACTERS } from "../numerals/characters.js";
+import { toCharacters } from "../script/characters.js";
 import { tagOf, wordEndingAt, wordStartingAt, } from "./rules.js";
 /**
  * Whether a syllable is `děi`, the modal reading of 得.
@@ -100,8 +102,183 @@ export const ATTESTED_ERHUA = {
         return (reading?.at(-1)?.erhua ?? false) ? "forbid" : "keep";
     },
 };
+/**
+ * Whether a syllable is `jiāo`, the verbal reading of 教.
+ */
+function isJiao(syllable) {
+    return (syllable?.initial === "j" && syllable.final === "iao" && syllable.tone === 1);
+}
+/**
+ * What a 教 that is teaching governs: the thing or the person taught.
+ *
+ * A pronoun, a common noun, a name or one of the tags jieba gives a subject —
+ * 教我, 教英语, 教历史, 教孩子, 教玛丽. `ns` is deliberately absent, since the
+ * only 教 with a place tag after it over the corpus is 教青局, an office.
+ */
+const TAUGHT_TAGS = new Set(["r", "n", "nz", "nr", "nt", "ng"]);
+/**
+ * What follows a 教 that is teaching where no object does: aspect and the
+ * complement marker, which only a verb takes.
+ *
+ * 他教了三年书, 他教过我英语, 她教书教得很快乐. Matched as characters rather
+ * than through their tags, because what the dictionary has starting at a 得 is
+ * as likely to be 得很 as the particle itself.
+ */
+const ASPECT = new Set(["了", "过", "過", "着", "著", "得"]);
+/**
+ * The tag prefix jieba gives a particle, which nothing teaching can follow.
+ */
+const PARTICLE_TAG = "u";
+/**
+ * Whether the 教 at a position is teaching something to somebody.
+ *
+ * The object is what says so, which is the same shape the modal 得 rule takes:
+ * a 教 with a pronoun, a noun or a name after it is governing it, and a 教 with
+ * 了, 过, 着 or 得 after it is a verb whatever follows that. A particle in front
+ * of it rules it out, since a verb does not follow one.
+ */
+function isTeachingAt(context, at) {
+    if (tagOf(context, wordEndingAt(context, at)).startsWith(PARTICLE_TAG)) {
+        return false;
+    }
+    return (ASPECT.has(context.characters[at + 1] ?? "") ||
+        TAUGHT_TAGS.has(tagOf(context, wordStartingAt(context, at + 1))));
+}
+/**
+ * 教 read as `jiāo` where it is teaching rather than a religion.
+ *
+ * The dictionary stores 教 as `jiào` with `jiāo` as an alternate, and `jiào` is
+ * right for the compounds — 教育, 教师, 宗教, 主教 — which are words and reach
+ * their reading through the word. What is left is the 教 that stands as a word
+ * of its own, and that one is the verb: 他在北京大学教了三年书 came out
+ * `jiàole sān nián shū`, and 我教英语, 谁教你法语 and 她教我如何游泳 with it.
+ * 教书 is a word and was always `jiāo shū`; nothing carried that across a 了.
+ *
+ * See {@link isTeachingAt} for what decides it. Forcing the single-character
+ * edge is not enough on its own, because a reading spanning two characters
+ * carries its own 教 into the position: 王老师教我们汉语 read `jiào` off 师教,
+ * and 来教我 off 来教. Those are pairs the dictionary holds with no part of
+ * speech and at the cost of a word nothing has ever counted — a reading rather
+ * than a word — so a *tagged* word ending in 教 is left alone, which is every
+ * one that matters: 任教, 宗教, 主教, 佛教, 传教, 执教, 请教, 家教.
+ *
+ * Measured over 88,866 lines of Tatoeba and zh.wikipedia, 181 教 decode as a
+ * word of their own and every one of them read `jiào`. This moves 158 to
+ * `jiāo` and is wrong on three: 统一教创始人, 方法教深思 and 做到了教政分离,
+ * where a nominal compound takes an object's shape, the last of them because
+ * 做到了 is a word and so the 了 in front is not a particle to look at. Of the
+ * 21 it leaves, 8 are right — 诸教中, 教外别传, 董教总, 风教, 教青局 twice, a
+ * 教室 reached through a bad split, and a 教 used as a bare noun — and 13 are
+ * misses, where jieba tags the object something this does not name.
+ */
+export const TEACHING_JIAO = {
+    name: "teaching-jiao",
+    verdictFor: (context) => {
+        const { edge } = context;
+        if (edge.text === "教") {
+            return isJiao(edge.reading[0]) && isTeachingAt(context, edge.from)
+                ? "force"
+                : "keep";
+        }
+        return edge.partOfSpeech === "" &&
+            edge.text.endsWith("教") &&
+            edge.reading.length === edge.to - edge.from &&
+            !isJiao(edge.reading.at(-1)) &&
+            isTeachingAt(context, edge.to - 1)
+            ? "forbid"
+            : "keep";
+    },
+};
+/**
+ * The tag the dictionary gives a 量词.
+ */
+const MEASURE = "q";
+/**
+ * The mark that makes the number after it a position rather than a count.
+ *
+ * 第三集团军 is the Third Army Group and not three 集团军, so the 集 the
+ * dictionary tags `q` is counting nothing and 集团军 is a word like any other.
+ */
+const ORDINAL = "第";
+/**
+ * Where the number counting the character at a position starts, or undefined
+ * where nothing counts it.
+ *
+ * Two things stop a numeral in front of a 量词 from counting it. An ordinal
+ * names a position, so 第十级别 is the tenth 级别 rather than ten 级; and a
+ * numeral inside a longer word belongs to that word, so the 一 of 唯一道路 is
+ * counting no 道 and the 二 of 十二月 no 月.
+ */
+function countedFrom(context, at) {
+    const { characters } = context;
+    let from = at;
+    while (from > 0 && QUANTITY_CHARACTERS.has(characters[from - 1] ?? "")) {
+        from--;
+    }
+    if (from === at || characters[from - 1] === ORDINAL) {
+        return undefined;
+    }
+    const before = wordEndingAt(context, at);
+    const held = before === undefined ? at : at - toCharacters(before).length;
+    return held < from ? undefined : from;
+}
+/**
+ * A 量词 with a number in front of it does not start the next word.
+ *
+ * 三个人 is three people and came out `sān gèrén`, three *personals*, because
+ * 个人 is a common noun and nothing weighed it against the 个 belonging to the
+ * 三 in front of it. The same swallow takes 一杯水 into 杯水, 两家俱乐部 into
+ * 家俱 and 一天天 into 天天; digits reach it too, since a run decoded after a
+ * number sees that number — see `decodeRun` — so 2个人 is the same case.
+ *
+ * What makes it decidable is that the dictionary tags the 量词: 个, 次, 天, 位,
+ * 杯 and 匹 are `q`, while the characters that only look like measure words in
+ * this position are not — 分 is `v`, 部 and 成 are `n`, 年 and 点 are `m`. So
+ * 五分钟, 三部分, 五成分, 五年级 and 三点钟 are untouched, and those are exactly
+ * the words a blanket rule would break. A word that is itself a 量词 is kept as
+ * well, since 分钟 and 点钟 are units rather than something being counted.
+ *
+ * Only the edge *starting at the 量词* is forbidden, so a number and its
+ * measure written solid survive whole where the dictionary has them: 一辈子 and
+ * 一会儿 are entries covering the number too, and the decode still reaches them.
+ *
+ * Measured over the same 88,866 lines of Tatoeba and zh.wikipedia the other
+ * rules were sized against, it forbids 561 edges across 144 shapes — 个人 203
+ * of them, then 杯水, 天一, 口气, 些小 and 家人 — and moves 53 decodes, since
+ * most of what it takes away the cost model was not going to choose anyway.
+ * Three of the 53 are wrong: 一批评, 这一名词 and 六七股灾, where the 一 and the
+ * 六七 count nothing and no tag says so. Exactly one reading changes over the
+ * corpus and it is a fix, 下了两天雨 having read 天雨 as `tiān yù`.
+ */
+export const COUNTED_MEASURE = {
+    name: "counted-measure",
+    verdictFor: (context) => {
+        const { characters, edge } = context;
+        // A reading of a different length from its span says something about how
+        // the characters are read rather than only about where they break: 份儿 is
+        // `fènr` over two characters, and taking the edge away would take the 儿化
+        // with it.
+        if (edge.to - edge.from < 2 ||
+            edge.reading.length !== edge.to - edge.from) {
+            return "keep";
+        }
+        const counted = countedFrom(context, edge.from);
+        if (counted === undefined) {
+            return "keep";
+        }
+        return tagOf(context, characters[edge.from]) === MEASURE &&
+            edge.partOfSpeech !== MEASURE
+            ? "forbid"
+            : "keep";
+    },
+};
 /**
  * The rules the lattice decode applies, in order.
  */
-export const READING_RULES = [MODAL_DE, ATTESTED_ERHUA];
+export const READING_RULES = [
+    MODAL_DE,
+    TEACHING_JIAO,
+    ATTESTED_ERHUA,
+    COUNTED_MEASURE,
+];
 //# sourceMappingURL=reading-rules.js.map

package/dist/decode/reading-rules.js.map

@@ -1 +1 @@
-{"version":3,"file":"reading-rules.js","sourceRoot":"","sources":["../../src/decode/reading-rules.ts"],"names":[],"mappings":"AACA,OAAO,EAGL,KAAK,EACL,YAAY,EACZ,cAAc,GACf,MAAM,YAAY,CAAC;AAEpB;;;;;GAKG;AACH,SAAS,KAAK,CAAC,QAA8B;IAC3C,OAAO,CACL,QAAQ,EAAE,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,IAAI,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,CAC5E,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAE9C;;;;;;GAMG;AACH,MAAM,aAAa,GAAG,eAAe,CAAC;AAEtC;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAa;IAChC,IAAI,EAAE,UAAU;IAChB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACzB,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACjD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/C,OAAO,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC7C,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;YACzC,CAAC,CAAC,OAAO;YACT,CAAC,CAAC,MAAM,CAAC;IACb,CAAC;CACF,CAAC;AAEF;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,CAAC,MAAM,cAAc,GAAa;IACtC,IAAI,EAAE,gBAAgB;IACtB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACrC,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;YACzC,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,wEAAwE;QACxE,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,EAAE,GAAG,CAAC;QACvD,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,OAAO,CAAC;QAC7D,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC;IAC/D,CAAC;CACF,CAAC;AAEF;;GAEG;AACH,MAAM,CAAC,MAAM,aAAa,GAAwB,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAC"}
+{"version":3,"file":"reading-rules.js","sourceRoot":"","sources":["../../src/decode/reading-rules.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAEvD,OAAO,EAGL,KAAK,EACL,YAAY,EACZ,cAAc,GACf,MAAM,YAAY,CAAC;AAEpB;;;;;GAKG;AACH,SAAS,KAAK,CAAC,QAA8B;IAC3C,OAAO,CACL,QAAQ,EAAE,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,IAAI,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,CAC5E,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAE9C;;;;;;GAMG;AACH,MAAM,aAAa,GAAG,eAAe,CAAC;AAEtC;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAa;IAChC,IAAI,EAAE,UAAU;IAChB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACzB,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACjD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/C,OAAO,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC7C,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;YACzC,CAAC,CAAC,OAAO;YACT,CAAC,CAAC,MAAM,CAAC;IACb,CAAC;CACF,CAAC;AAEF;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,CAAC,MAAM,cAAc,GAAa;IACtC,IAAI,EAAE,gBAAgB;IACtB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACrC,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;YACzC,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,wEAAwE;QACxE,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,EAAE,GAAG,CAAC;QACvD,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,OAAO,CAAC;QAC7D,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC;IAC/D,CAAC;CACF,CAAC;AAEF;;GAEG;AACH,SAAS,MAAM,CAAC,QAA8B;IAC5C,OAAO,CACL,QAAQ,EAAE,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,KAAK,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,CAC7E,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;AAEhE;;;;;;;GAOG;AACH,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAEvD;;GAEG;AACH,MAAM,YAAY,GAAG,GAAG,CAAC;AAEzB;;;;;;;GAOG;AACH,SAAS,YAAY,CAAC,OAAoB,EAAE,EAAU;IACpD,IAAI,KAAK,CAAC,OAAO,EAAE,YAAY,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QACvE,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,CACL,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;QAC5C,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,cAAc,CAAC,OAAO,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,CACjE,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,MAAM,CAAC,MAAM,aAAa,GAAa;IACrC,IAAI,EAAE,eAAe;IACrB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACzB,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,EAAE,CAAC;YACtB,OAAO,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC;gBAChE,CAAC,CAAC,OAAO;gBACT,CAAC,CAAC,MAAM,CAAC;QACb,CAAC;QACD,OAAO,IAAI,CAAC,YAAY,KAAK,EAAE;YAC7B,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC;YACvB,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI;YAC3C,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;YAC5B,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC;YAClC,CAAC,CAAC,QAAQ;YACV,CAAC,CAAC,MAAM,CAAC;IACb,CAAC;CACF,CAAC;AAEF;;GAEG;AACH,MAAM,OAAO,GAAG,GAAG,CAAC;AAEpB;;;;;GAKG;AACH,MAAM,OAAO,GAAG,GAAG,CAAC;AAEpB;;;;;;;;GAQG;AACH,SAAS,WAAW,CAAC,OAAoB,EAAE,EAAU;IACnD,MAAM,EAAE,UAAU,EAAE,GAAG,OAAO,CAAC;IAC/B,IAAI,IAAI,GAAG,EAAE,CAAC;IACd,OAAO,IAAI,GAAG,CAAC,IAAI,mBAAmB,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC;QACvE,IAAI,EAAE,CAAC;IACT,CAAC;IACD,IAAI,IAAI,KAAK,EAAE,IAAI,UAAU,CAAC,IAAI,GAAG,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;QACpD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IACzC,MAAM,IAAI,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAC1E,OAAO,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC;AACxC,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,MAAM,CAAC,MAAM,eAAe,GAAa;IACvC,IAAI,EAAE,iBAAiB;IACvB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACrC,yEAAyE;QACzE,yEAAyE;QACzE,yEAAyE;QACzE,WAAW;QACX,IACE,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,GAAG,CAAC;YACvB,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,EAC3C,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;YAC1B,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,OAAO,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,OAAO;YACtD,IAAI,CAAC,YAAY,KAAK,OAAO;YAC7B,CAAC,CAAC,QAAQ;YACV,CAAC,CAAC,MAAM,CAAC;IACb,CAAC;CACF,CAAC;AAEF;;GAEG;AACH,MAAM,CAAC,MAAM,aAAa,GAAwB;IAChD,QAAQ;IACR,aAAa;IACb,cAAc;IACd,eAAe;CAChB,CAAC"}

package/dist/dictionary/assertions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../src/dictionary/assertions.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAElD;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,kEAAkE;IAClE,QAAQ,CAAC,KAAK,EAAE,CAAC,UAAU,EAAE,eAAe,KAAK,MAAM,GAAG,SAAS,CAAC;CACrE;AAED;;GAEG;AACH,qBAAa,eAAe;;IAG1B,QAAQ,CAAC,OAAO,EAAE,SAAS,eAAe,EAAE,CAAC;IAE7C;;;;;OAKG;IACH,YAAY,OAAO,EAAE,SAAS,eAAe,EAAE,EAc9C;IAeD;;OAEG;IACH,GAAG,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAE7C;IAED;;OAEG;IACH,OAAO,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAKxC;IAED;;OAEG;IACH,iBAAiB,IAAI,WAAW,CAAC,MAAM,CAAC,CAUvC;IAED;;;;;;;;OAQG;IACH,eAAe,IAAI,WAAW,CAAC,MAAM,CAAC,CAUrC;CACF;AAoBD;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,gBAAgB,EAAE,SAAS,cAAc,EAgHrD,CAAC;AAEF;;GAEG;AACH,wBAAgB,UAAU,CACxB,OAAO,EAAE,SAAS,eAAe,EAAE,GAClC,SAAS,MAAM,EAAE,CAKnB"}
+{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../src/dictionary/assertions.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAElD;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,kEAAkE;IAClE,QAAQ,CAAC,KAAK,EAAE,CAAC,UAAU,EAAE,eAAe,KAAK,MAAM,GAAG,SAAS,CAAC;CACrE;AAED;;GAEG;AACH,qBAAa,eAAe;;IAG1B,QAAQ,CAAC,OAAO,EAAE,SAAS,eAAe,EAAE,CAAC;IAE7C;;;;;OAKG;IACH,YAAY,OAAO,EAAE,SAAS,eAAe,EAAE,EAc9C;IAeD;;OAEG;IACH,GAAG,CAAC,IAAI,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAE7C;IAED;;OAEG;IACH,OAAO,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAKxC;IAED;;OAEG;IACH,iBAAiB,IAAI,WAAW,CAAC,MAAM,CAAC,CAUvC;IAED;;;;;;;;OAQG;IACH,eAAe,IAAI,WAAW,CAAC,MAAM,CAAC,CAUrC;CACF;AAoBD;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,gBAAgB,EAAE,SAAS,cAAc,EAwIrD,CAAC;AAEF;;GAEG;AACH,wBAAgB,UAAU,CACxB,OAAO,EAAE,SAAS,eAAe,EAAE,GAClC,SAAS,MAAM,EAAE,CAKnB"}

package/dist/dictionary/assertions.js

@@ -170,6 +170,26 @@ export const BUILD_ASSERTIONS = [
                 : `no longer proper nouns: ${wrong.join(", ")}`;
         },
     },
+    {
+        // The gap composeLocaleDeltas closes, and the two ways it must not close
+        // it. 運行狀況 contains the marked word 行狀 and cuts elsewhere; 一个个地
+        // ends in a character whose delta is a different sense of 地 entirely.
+        description: "垃圾分類 inherits 垃圾's 國語 reading, 運行狀況 inherits nothing",
+        check: (dictionary) => {
+            const taiwan = (word) => dictionary
+                .get(word)
+                ?.readings.tw?.map((syllable) => writeSyllable(syllable))
+                .join(" ");
+            const composed = taiwan("垃圾分类");
+            if (composed !== "lè sè fēn lèi") {
+                return `垃圾分类 reads ${composed ?? "nothing"} in zh-TW, expected lè sè fēn lèi`;
+            }
+            const wrong = ["运行状况", "一个个地", "相亲相爱"].filter((word) => taiwan(word) !== undefined);
+            return wrong.length === 0
+                ? undefined
+                : `composed a zh-TW reading for: ${wrong.join(", ")}`;
+        },
+    },
     {
         description: "头发 derives 頭髮 rather than 頭發",
         check: (dictionary) => {

package/dist/dictionary/assertions.js.map

@@ -1 +1 @@
-{"version":3,"file":"assertions.js","sourceRoot":"","sources":["../../src/dictionary/assertions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AACzD,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAChF,OAAO,EAAiB,aAAa,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AAYjD;;GAEG;AACH,MAAM,OAAO,eAAe;IACjB,MAAM,CAAuC;IAE7C,OAAO,CAA6B;IAE7C;;;;;OAKG;IACH,YAAY,OAAmC;QAC7C,MAAM,KAAK,GAAG,IAAI,GAAG,EAA2B,CAAC;QACjD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QAC/B,CAAC;QACD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,KAAK,MAAM,IAAI,IAAI,gBAAgB,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC3C,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;oBACrB,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;gBACzB,CAAC;YACH,CAAC;QACH,CAAC;QACD,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;IAED;;OAEG;IACH,CAAC,SAAS;QACR,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjC,KAAK,CAAC,CAAC;gBACL,KAAK,CAAC,QAAQ,CAAC,EAAE;gBACjB,KAAK,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE;gBACvB,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,EAAE,CAAC;aAC5B,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;IACH,GAAG,CAAC,IAAY;QACd,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IAED;;OAEG;IACH,OAAO,CAAC,IAAY;QAClB,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACpC,OAAO,KAAK,KAAK,SAAS;YACxB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC7E,CAAC;IAED;;OAEG;IACH,iBAAiB;QACf,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;QACpC,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,CAAC;YACvC,KAAK,MAAM,QAAQ,IAAI,OAAO,EAAE,CAAC;gBAC/B,SAAS,CAAC,GAAG,CACX,aAAa,CAAC,EAAE,GAAG,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,CAC9D,CAAC;YACJ,CAAC;QACH,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;;;;;;;OAQG;IACH,eAAe;QACb,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAC;QACrC,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,CAAC;YACvC,KAAK,MAAM,QAAQ,IAAI,OAAO,EAAE,CAAC;gBAC/B,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC9B,UAAU,CAAC,GAAG,CAAC,aAAa,CAAC,EAAE,GAAG,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;gBAC/D,CAAC;YACH,CAAC;QACH,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;CACF;AAED;;GAEG;AACH,SAAS,KAAK,CAAC,IAAY,EAAE,QAAgB,EAAE,GAAW;IACxD,OAAO;QACL,WAAW,EAAE,GAAG,IAAI,MAAM,QAAQ,KAAK,GAAG,GAAG;QAC7C,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YACxC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,OAAO,GAAG,IAAI,iCAAiC,CAAC;YAClD,CAAC;YACD,OAAO,MAAM,KAAK,QAAQ;gBACxB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,GAAG,IAAI,UAAU,MAAM,cAAc,QAAQ,EAAE,CAAC;QACtD,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAA8B;IACzD,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,iCAAiC,CAAC;IACtD,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,+BAA+B,CAAC;IACrD,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,8CAA8C,CAAC;IACnE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,6BAA6B,CAAC;IACnD,KAAK,CAAC,MAAM,EAAE,gBAAgB,EAAE,yBAAyB,CAAC;IAC1D,KAAK,CAAC,MAAM,EAAE,gBAAgB,EAAE,+BAA+B,CAAC;IAChE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,oCAAoC,CAAC;IAC3D,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,oBAAoB,CAAC;IAC7C,KAAK,CAAC,IAAI,EAAE,YAAY,EAAE,oBAAoB,CAAC;IAC/C,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,oCAAoC,CAAC;IAC3D,KAAK,CAAC,IAAI,EAAE,SAAS,EAAE,oCAAoC,CAAC;IAC5D,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,kDAAkD,CAAC;IACzE,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,qCAAqC,CAAC;IAC9D,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,8BAA8B,CAAC;IACvD,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,iDAAiD,CAAC;IAC1E,0EAA0E;IAC1E,wEAAwE;IACxE,0EAA0E;IAC1E,4EAA4E;IAC5E,uDAAuD;IACvD,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,yCAAyC,CAAC;IAC3D,KAAK,CAAC,GAAG,EAAE,KAAK,EAAE,+CAA+C,CAAC;IAClE;QACE,WAAW,EAAE,iDAAiD;QAC9D,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE,CACzD,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI;YACzC,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,mCAAmC;KAC1C;IACD;QACE,qEAAqE;QACrE,2EAA2E;QAC3E,WAAW,EAAE,wCAAwC;QACrD,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,KAAK,GAAG,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,MAAM,CAC3C,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI,CACtD,CAAC;YACF,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,2BAA2B,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACpD,CAAC;KACF;IACD;QACE,kEAAkE;QAClE,6DAA6D;QAC7D,WAAW,EAAE,mCAAmC;QAChD,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,KAAK,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,MAAM,CACvC,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI,CACtD,CAAC;YACF,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,2BAA2B,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACpD,CAAC;KACF;IACD;QACE,WAAW,EAAE,8BAA8B;QAC3C,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,IAAI,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC;YACxC,OAAO,IAAI,KAAK,IAAI;gBAClB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,cAAc,IAAI,IAAI,SAAS,eAAe,CAAC;QACrD,CAAC;KACF;IACD;QACE,4EAA4E;QAC5E,sEAAsE;QACtE,0EAA0E;QAC1E,0EAA0E;QAC1E,4EAA4E;QAC5E,kCAAkC;QAClC,WAAW,EAAE,gDAAgD;QAC7D,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,OAAO,GAAG,CAAC,GAAG,UAAU,CAAC,iBAAiB,EAAE,CAAC,CAAC,MAAM,CACxD,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC,CAClD,CAAC;YACF,OAAO,OAAO,CAAC,MAAM,KAAK,CAAC;gBACzB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,gBAAgB,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,uCAAuC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACxG,CAAC;KACF;IACD;QACE,yEAAyE;QACzE,uEAAuE;QACvE,0EAA0E;QAC1E,oEAAoE;QACpE,WAAW,EAAE,sDAAsD;QACnE,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,UAAU,GAAG,CAAC,GAAG,UAAU,CAAC,eAAe,EAAE,CAAC,CAAC;YACrD,OAAO,UAAU,CAAC,MAAM,KAAK,CAAC;gBAC5B,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,gBAAgB,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,uDAAuD,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAC9H,CAAC;KACF;IACD;QACE,WAAW,EAAE,0DAA0D;QACvE,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;gBACvC,MAAM,UAAU,GAAG,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC9C,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CACpC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,KAAK,KAAK,IAAI,CACtC,CAAC,MAAM,CAAC;gBACT,iEAAiE;gBACjE,iEAAiE;gBACjE,IAAI,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,GAAG,KAAK,GAAG,UAAU,EAAE,CAAC;oBAClD,OAAO,GAAG,KAAK,CAAC,IAAI,QAAQ,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CAAC,kBAAkB,MAAM,CAAC,UAAU,CAAC,aAAa,CAAC;gBAChH,CAAC;YACH,CAAC;YACD,OAAO;QACT,CAAC;KACF;CACF,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,UAAU,CACxB,OAAmC;IAEnC,MAAM,UAAU,GAAG,IAAI,eAAe,CAAC,OAAO,CAAC,CAAC;IAChD,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CACxC,SAAS,CAAC,KAAK,CAAC,UAAU,CAAC,CAC5B,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC;AAC/C,CAAC"}
+{"version":3,"file":"assertions.js","sourceRoot":"","sources":["../../src/dictionary/assertions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AACzD,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAChF,OAAO,EAAiB,aAAa,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AAYjD;;GAEG;AACH,MAAM,OAAO,eAAe;IACjB,MAAM,CAAuC;IAE7C,OAAO,CAA6B;IAE7C;;;;;OAKG;IACH,YAAY,OAAmC;QAC7C,MAAM,KAAK,GAAG,IAAI,GAAG,EAA2B,CAAC;QACjD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QAC/B,CAAC;QACD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,KAAK,MAAM,IAAI,IAAI,gBAAgB,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC3C,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;oBACrB,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;gBACzB,CAAC;YACH,CAAC;QACH,CAAC;QACD,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QACpB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;IAED;;OAEG;IACH,CAAC,SAAS;QACR,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjC,KAAK,CAAC,CAAC;gBACL,KAAK,CAAC,QAAQ,CAAC,EAAE;gBACjB,KAAK,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE;gBACvB,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,EAAE,CAAC;aAC5B,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;IACH,GAAG,CAAC,IAAY;QACd,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IAED;;OAEG;IACH,OAAO,CAAC,IAAY;QAClB,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACpC,OAAO,KAAK,KAAK,SAAS;YACxB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC7E,CAAC;IAED;;OAEG;IACH,iBAAiB;QACf,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;QACpC,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,CAAC;YACvC,KAAK,MAAM,QAAQ,IAAI,OAAO,EAAE,CAAC;gBAC/B,SAAS,CAAC,GAAG,CACX,aAAa,CAAC,EAAE,GAAG,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,CAC9D,CAAC;YACJ,CAAC;QACH,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;;;;;;;OAQG;IACH,eAAe;QACb,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAC;QACrC,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,SAAS,EAAE,EAAE,CAAC;YACvC,KAAK,MAAM,QAAQ,IAAI,OAAO,EAAE,CAAC;gBAC/B,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC9B,UAAU,CAAC,GAAG,CAAC,aAAa,CAAC,EAAE,GAAG,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;gBAC/D,CAAC;YACH,CAAC;QACH,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;CACF;AAED;;GAEG;AACH,SAAS,KAAK,CAAC,IAAY,EAAE,QAAgB,EAAE,GAAW;IACxD,OAAO;QACL,WAAW,EAAE,GAAG,IAAI,MAAM,QAAQ,KAAK,GAAG,GAAG;QAC7C,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YACxC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,OAAO,GAAG,IAAI,iCAAiC,CAAC;YAClD,CAAC;YACD,OAAO,MAAM,KAAK,QAAQ;gBACxB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,GAAG,IAAI,UAAU,MAAM,cAAc,QAAQ,EAAE,CAAC;QACtD,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAA8B;IACzD,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,iCAAiC,CAAC;IACtD,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,+BAA+B,CAAC;IACrD,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,8CAA8C,CAAC;IACnE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,6BAA6B,CAAC;IACnD,KAAK,CAAC,MAAM,EAAE,gBAAgB,EAAE,yBAAyB,CAAC;IAC1D,KAAK,CAAC,MAAM,EAAE,gBAAgB,EAAE,+BAA+B,CAAC;IAChE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,oCAAoC,CAAC;IAC3D,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,oBAAoB,CAAC;IAC7C,KAAK,CAAC,IAAI,EAAE,YAAY,EAAE,oBAAoB,CAAC;IAC/C,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,oCAAoC,CAAC;IAC3D,KAAK,CAAC,IAAI,EAAE,SAAS,EAAE,oCAAoC,CAAC;IAC5D,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,kDAAkD,CAAC;IACzE,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,qCAAqC,CAAC;IAC9D,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,8BAA8B,CAAC;IACvD,KAAK,CAAC,IAAI,EAAE,UAAU,EAAE,iDAAiD,CAAC;IAC1E,0EAA0E;IAC1E,wEAAwE;IACxE,0EAA0E;IAC1E,4EAA4E;IAC5E,uDAAuD;IACvD,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,yCAAyC,CAAC;IAC3D,KAAK,CAAC,GAAG,EAAE,KAAK,EAAE,+CAA+C,CAAC;IAClE;QACE,WAAW,EAAE,iDAAiD;QAC9D,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE,CACzD,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI;YACzC,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,mCAAmC;KAC1C;IACD;QACE,qEAAqE;QACrE,2EAA2E;QAC3E,WAAW,EAAE,wCAAwC;QACrD,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,KAAK,GAAG,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,MAAM,CAC3C,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI,CACtD,CAAC;YACF,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,2BAA2B,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACpD,CAAC;KACF;IACD;QACE,kEAAkE;QAClE,6DAA6D;QAC7D,WAAW,EAAE,mCAAmC;QAChD,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,KAAK,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,MAAM,CACvC,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,YAAY,KAAK,IAAI,CACtD,CAAC;YACF,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,2BAA2B,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACpD,CAAC;KACF;IACD;QACE,yEAAyE;QACzE,gEAAgE;QAChE,sEAAsE;QACtE,WAAW,EACT,sDAAsD;QACxD,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,MAAM,GAAG,CAAC,IAAY,EAAsB,EAAE,CAClD,UAAU;iBACP,GAAG,CAAC,IAAI,CAAC;gBACV,EAAE,QAAQ,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;iBACxD,IAAI,CAAC,GAAG,CAAC,CAAC;YACf,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC;YAChC,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;gBACjC,OAAO,cAAc,QAAQ,IAAI,SAAS,mCAAmC,CAAC;YAChF,CAAC;YACD,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC,MAAM,CAC3C,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,SAAS,CACrC,CAAC;YACF,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,iCAAiC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAC1D,CAAC;KACF;IACD;QACE,WAAW,EAAE,8BAA8B;QAC3C,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,IAAI,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC;YACxC,OAAO,IAAI,KAAK,IAAI;gBAClB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,cAAc,IAAI,IAAI,SAAS,eAAe,CAAC;QACrD,CAAC;KACF;IACD;QACE,4EAA4E;QAC5E,sEAAsE;QACtE,0EAA0E;QAC1E,0EAA0E;QAC1E,4EAA4E;QAC5E,kCAAkC;QAClC,WAAW,EAAE,gDAAgD;QAC7D,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,OAAO,GAAG,CAAC,GAAG,UAAU,CAAC,iBAAiB,EAAE,CAAC,CAAC,MAAM,CACxD,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC,CAClD,CAAC;YACF,OAAO,OAAO,CAAC,MAAM,KAAK,CAAC;gBACzB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,gBAAgB,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,uCAAuC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACxG,CAAC;KACF;IACD;QACE,yEAAyE;QACzE,uEAAuE;QACvE,0EAA0E;QAC1E,oEAAoE;QACpE,WAAW,EAAE,sDAAsD;QACnE,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,MAAM,UAAU,GAAG,CAAC,GAAG,UAAU,CAAC,eAAe,EAAE,CAAC,CAAC;YACrD,OAAO,UAAU,CAAC,MAAM,KAAK,CAAC;gBAC5B,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,gBAAgB,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,uDAAuD,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAC9H,CAAC;KACF;IACD;QACE,WAAW,EAAE,0DAA0D;QACvE,KAAK,EAAE,CAAC,UAA2B,EAAsB,EAAE;YACzD,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;gBACvC,MAAM,UAAU,GAAG,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC9C,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CACpC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,KAAK,KAAK,IAAI,CACtC,CAAC,MAAM,CAAC;gBACT,iEAAiE;gBACjE,iEAAiE;gBACjE,IAAI,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,GAAG,KAAK,GAAG,UAAU,EAAE,CAAC;oBAClD,OAAO,GAAG,KAAK,CAAC,IAAI,QAAQ,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CAAC,kBAAkB,MAAM,CAAC,UAAU,CAAC,aAAa,CAAC;gBAChH,CAAC;YACH,CAAC;YACD,OAAO;QACT,CAAC;KACF;CACF,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,UAAU,CACxB,OAAmC;IAEnC,MAAM,UAAU,GAAG,IAAI,eAAe,CAAC,OAAO,CAAC,CAAC;IAChD,OAAO,gBAAgB,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CACxC,SAAS,CAAC,KAAK,CAAC,UAAU,CAAC,CAC5B,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC;AAC/C,CAAC"}

package/dist/dictionary/locale.d.ts

@@ -0,0 +1,55 @@
+import { type DictionaryEntry } from "./entry.js";
+/**
+ * Compounds whose constituent's 國語 reading must not be carried over.
+ *
+ * The composition below only fires where the compound reads its constituent
+ * exactly as that constituent's own entry does, which rules out most of the ways
+ * it could be wrong. What it cannot rule out is a homograph: two words spelled
+ * and read identically in 普通话, of which only one shifts in 國語. 相亲 is
+ * `xiāngqīn` either way when it means "mutually close" and `xiàngqīn` in Taiwan
+ * when it means a matchmaking meeting, and nothing in the data distinguishes the
+ * two senses.
+ *
+ * **Keep this small**, for the same reason
+ * {@link import("./overrides.js").READING_OVERRIDES} is kept small. Three
+ * exclusions against the 100 compounds the composition settles is the measured
+ * ratio; a list growing much past that would mean the alignment check has
+ * stopped carrying its weight and the rule itself needs revisiting.
+ */
+export declare const LOCALE_COMPOSITION_EXCLUSIONS: readonly {
+    readonly word: string;
+    readonly reason: string;
+}[];
+/**
+ * What the composition pass changed.
+ */
+export interface LocaleComposition {
+    readonly entries: readonly DictionaryEntry[];
+    /** Compounds given a zh-TW delta they had no source for. */
+    readonly composed: number;
+}
+/**
+ * Give a compound the 國語 reading its constituents already carry.
+ *
+ * The gap this closes: `zh-TW` is a delta over `zh-CN`, and a source marks that
+ * delta on whichever headword it happened to list. CC-CEDICT marks 垃圾 and
+ * 垃圾桶 but not 垃圾分类, and because the decoder prefers the longest word it
+ * finds, 垃圾分类 is decoded whole and 垃圾's delta is never consulted — so a
+ * locale switch that visibly works on 垃圾 appears to do nothing on 垃圾分类.
+ * Patching the entries a source missed would be endless; the compounds are
+ * open-class and the marked words are not.
+ *
+ * Three things keep it from overreaching, and all three are load-bearing:
+ *
+ * 1. **Segmentation, not substring.** 运行状况 contains 行状, which carries a
+ *    delta, but cuts as 运行 + 状况 and never as 运 + 行状 + 况.
+ * 2. **The constituent must be read here as its own entry reads it**, or the
+ *    compound means something else by those characters.
+ * 3. **Only whole words contribute**, never single characters — see
+ *    {@link composedReading} for what allowing them costs.
+ *
+ * What survives all three is a homograph the data cannot see, which is what
+ * {@link LOCALE_COMPOSITION_EXCLUSIONS} is for.
+ */
+export declare function composeLocaleDeltas(entries: readonly DictionaryEntry[]): LocaleComposition;
+//# sourceMappingURL=locale.d.ts.map

package/dist/dictionary/locale.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"locale.d.ts","sourceRoot":"","sources":["../../src/dictionary/locale.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,KAAK,eAAe,EAAiB,MAAM,YAAY,CAAC;AAGjE;;;;;;;;;;;;;;;;GAgBG;AACH,eAAO,MAAM,6BAA6B,EAAE,SAAS;IACnD,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB,EAeA,CAAC;AA6PF;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,OAAO,EAAE,SAAS,eAAe,EAAE,CAAC;IAC7C,4DAA4D;IAC5D,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;CAC3B;AAYD;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,mBAAmB,CACjC,OAAO,EAAE,SAAS,eAAe,EAAE,GAClC,iBAAiB,CA8DnB"}

package/dist/dictionary/locale.js

@@ -0,0 +1,323 @@
+import { isSingleCharacter, toCharacters } from "../script/characters.js";
+import { isSameReading } from "./entry.js";
+import { FrequencyTable } from "./frequency-table.js";
+/**
+ * Compounds whose constituent's 國語 reading must not be carried over.
+ *
+ * The composition below only fires where the compound reads its constituent
+ * exactly as that constituent's own entry does, which rules out most of the ways
+ * it could be wrong. What it cannot rule out is a homograph: two words spelled
+ * and read identically in 普通话, of which only one shifts in 國語. 相亲 is
+ * `xiāngqīn` either way when it means "mutually close" and `xiàngqīn` in Taiwan
+ * when it means a matchmaking meeting, and nothing in the data distinguishes the
+ * two senses.
+ *
+ * **Keep this small**, for the same reason
+ * {@link import("./overrides.js").READING_OVERRIDES} is kept small. Three
+ * exclusions against the 100 compounds the composition settles is the measured
+ * ratio; a list growing much past that would mean the alignment check has
+ * stopped carrying its weight and the rule itself needs revisiting.
+ */
+export const LOCALE_COMPOSITION_EXCLUSIONS = [
+    {
+        word: "相亲相爱",
+        reason: "相亲 here is 相 + 亲 in the reciprocal sense, `xiāngqīn`, not the matchmaking word that reads `xiàngqīn` in 國語.",
+    },
+    {
+        word: "腹背相亲",
+        reason: "The same reciprocal 相亲, so the same reading in both locales.",
+    },
+    {
+        word: "洋为中用",
+        reason: "中 is 中国 and 用 is its own verb: 洋为中用 is not built on the word 中用, which is what reads `zhòngyòng` in 國語.",
+    },
+];
+/**
+ * The exclusions as a set, for the composition to consult.
+ */
+const EXCLUDED = new Set(LOCALE_COMPOSITION_EXCLUSIONS.map((exclusion) => exclusion.word));
+/**
+ * What each word in a segmentation costs on top of its frequency.
+ *
+ * The same charge, and for the same reason, as `READING_CHARGE` in
+ * `src/decode/lattice.ts`: it orders candidate segmentations by fewest words
+ * first and frequency second. That ordering is what decides 皮夹克 is 皮 + 夹克
+ * rather than 皮夹 + 克, which is the difference between leaving `pí jiákè`
+ * alone and mistaking it for the wallet 皮夹 that reads `pí jiá` in 國語.
+ *
+ * Duplicated rather than imported because the decoder sits above the dictionary
+ * and must not be depended on from inside it. The two are kept in step by
+ * `src/dictionary/locale.test.ts`, which asserts the segmentations this value
+ * produces rather than the value itself.
+ */
+const WORD_CHARGE = 16;
+/**
+ * Every key an entry answers to, under either script.
+ */
+function keysOf(entry) {
+    return [entry.hans, entry.hant, ...(entry.hantVariants ?? [])];
+}
+/**
+ * A compound's constituent words, and the readings they carry.
+ *
+ * Indexed by every key an entry claims, so that a 繁體 headword segments into
+ * 繁體 constituents. Headwords are set first and never displaced, which is the
+ * same precedence the artifact and the build assertions use: 发 and 髮 both claim
+ * 发, and the entry the key names has to win.
+ */
+class ConstituentIndex {
+    #byKey;
+    #cost;
+    /**
+     * Index entries by key, quantising their frequencies the way the decoder will.
+     */
+    constructor(entries) {
+        const byKey = new Map();
+        for (const entry of entries) {
+            byKey.set(entry.hans, entry);
+        }
+        for (const entry of entries) {
+            for (const form of keysOf(entry)) {
+                if (!byKey.has(form)) {
+                    byKey.set(form, entry);
+                }
+            }
+        }
+        // Costs come from the same quantised table the artifact ships, so that a
+        // segmentation decided here is the one the decoder would have reached.
+        const table = FrequencyTable.build(entries.map((entry) => entry.frequency));
+        const byEntry = new Map();
+        for (const [at, entry] of entries.entries()) {
+            byEntry.set(entry, table.costOf(at));
+        }
+        const cost = new Map();
+        for (const [key, entry] of byKey) {
+            /* c8 ignore next -- every key was indexed from one of these entries */
+            cost.set(key, (byEntry.get(entry) ?? 0) + WORD_CHARGE);
+        }
+        this.#byKey = byKey;
+        this.#cost = cost;
+    }
+    /**
+     * Relax every word starting at one position, improving the costs it reaches.
+     *
+     * One step of the shortest-path scan, lifted out of its enclosing loop.
+     */
+    #relaxFrom(word, characters, from, best, previous) {
+        /* c8 ignore next -- the caller only relaxes from a position in range */
+        const reached = best[from] ?? Infinity;
+        let candidate = "";
+        for (let to = from; to < characters.length; to++) {
+            /* c8 ignore next -- the loop condition keeps the index in range */
+            candidate += characters[to] ?? "";
+            // The whole word is not a segmentation of itself.
+            const entry = candidate === word ? undefined : this.#byKey.get(candidate);
+            const total = reached + this.costOf(candidate);
+            /* c8 ignore next -- `to + 1` is at most `length`, which `best` holds */
+            if (entry !== undefined && total < (best[to + 1] ?? Infinity)) {
+                best[to + 1] = total;
+                previous[to + 1] = { entry, text: candidate, from, to: to + 1 };
+            }
+        }
+    }
+    /**
+     * What a key costs a segmentation that uses it.
+     */
+    costOf(key) {
+        /* c8 ignore next -- only asked about keys the index just returned */
+        return this.#cost.get(key) ?? WORD_CHARGE;
+    }
+    /**
+     * The likeliest way to cut a word into *other* dictionary words.
+     *
+     * The whole word is excluded from its own segmentation, which is the point:
+     * 垃圾分类 is a dictionary word, so a decoder handed the text would stop there
+     * and never ask what it is made of. This asks anyway.
+     *
+     * Returns undefined when no path of known words covers the word, which is
+     * common — most compounds contain a character no shorter word claims.
+     */
+    segment(word) {
+        const characters = toCharacters(word);
+        const length = characters.length;
+        const best = Array.from({ length: length + 1 }, () => Infinity);
+        const previous = Array.from({ length: length + 1 });
+        best[0] = 0;
+        for (let from = 0; from < length; from++) {
+            /* c8 ignore next -- the loop condition keeps the index in range */
+            if ((best[from] ?? Infinity) !== Infinity) {
+                this.#relaxFrom(word, characters, from, best, previous);
+            }
+        }
+        /* c8 ignore next -- `best` is built one longer than the word */
+        if ((best[length] ?? Infinity) === Infinity) {
+            return undefined;
+        }
+        const path = [];
+        let at = length;
+        while (at > 0) {
+            const step = previous[at];
+            /* c8 ignore next 3 -- a finite cost at `at` means a step reached it */
+            if (step === undefined) {
+                return undefined;
+            }
+            path.unshift(step);
+            at = step.from;
+        }
+        return path;
+    }
+}
+/**
+ * The 國語 reading a compound inherits from its constituents, or undefined.
+ *
+ * Only a **multi-character** constituent may contribute. A character's delta
+ * reaches every compound that character appears in, and surviving the merge's
+ * sense test — see `isOwnSense` — does not make one locale-wide. 期 carries
+ * `qí`, which 國語 really does read everywhere, but 會 carries `huǐ`, and
+ * propagating it would rewrite 三合会 as `sānhéhuǐ`. Measured on the full
+ * dictionary, allowing characters to contribute composes 3,743 entries against
+ * the 101 that multi-character words compose.
+ */
+function composedReading(entry, path, settled) {
+    const reading = entry.readings.cn;
+    const composed = [];
+    let isContributed = false;
+    for (const part of path) {
+        const span = reading.slice(part.from, part.to);
+        // A constituent an earlier round composed counts as carrying a delta, which
+        // is what lets 中看不中用 inherit through 不中用.
+        const taiwan = settled.get(part.entry.hans) ?? part.entry.readings.tw;
+        if (taiwan === undefined || isSingleCharacter(part.text)) {
+            composed.push(...span);
+            continue;
+        }
+        // The compound has to read this constituent the way the constituent's own
+        // entry reads it. Where it does not, the compound is using another sense —
+        // 浑身解数 is `xiè shù` and 解数 is a word, but a compound reading `jiě shù`
+        // there would be a different word wearing the same characters, and its
+        // 國語 reading says nothing about this one.
+        if (!isSameReading(part.entry.readings.cn, span)) {
+            return undefined;
+        }
+        composed.push(...taiwan);
+        isContributed = true;
+    }
+    return isContributed ? composed : undefined;
+}
+/**
+ * Whether a compound can be asked what it is made of at all.
+ *
+ * A reading of one syllable per character is what lets a character span be cut
+ * out of the compound's reading and replaced. 儿化 folds two characters into one
+ * syllable and punctuation reads as none, so neither lines up and both are left
+ * alone.
+ */
+function isComposable(entry) {
+    const characters = toCharacters(entry.hans);
+    return (characters.length > 2 &&
+        entry.readings.cn.length === characters.length &&
+        !EXCLUDED.has(entry.hans) &&
+        !EXCLUDED.has(entry.hant));
+}
+/**
+ * Whether any delta-carrying word appears inside this one.
+ *
+ * A cheap filter ahead of the segmentation: only a few hundred words carry a
+ * delta, against 461,623 entries, and enumerating one word's substrings costs
+ * far less than segmenting it.
+ */
+function hasMarkedConstituent(word, marked) {
+    const characters = toCharacters(word);
+    return characters.some((_, from) => {
+        let candidate = "";
+        return characters.slice(from).some((character) => {
+            candidate += character;
+            return candidate !== word && marked.has(candidate);
+        });
+    });
+}
+/**
+ * How many times the pass may feed its own output back in.
+ *
+ * 中看不中用 needs two rounds: 不中用 is itself a compound with no delta of its
+ * own, so only once it has been composed can the phrase containing it inherit
+ * one. Nothing measured needs a third, and the bound is what stops a cycle in
+ * the data from turning into a loop here.
+ */
+const ROUNDS = 4;
+/**
+ * Give a compound the 國語 reading its constituents already carry.
+ *
+ * The gap this closes: `zh-TW` is a delta over `zh-CN`, and a source marks that
+ * delta on whichever headword it happened to list. CC-CEDICT marks 垃圾 and
+ * 垃圾桶 but not 垃圾分类, and because the decoder prefers the longest word it
+ * finds, 垃圾分类 is decoded whole and 垃圾's delta is never consulted — so a
+ * locale switch that visibly works on 垃圾 appears to do nothing on 垃圾分类.
+ * Patching the entries a source missed would be endless; the compounds are
+ * open-class and the marked words are not.
+ *
+ * Three things keep it from overreaching, and all three are load-bearing:
+ *
+ * 1. **Segmentation, not substring.** 运行状况 contains 行状, which carries a
+ *    delta, but cuts as 运行 + 状况 and never as 运 + 行状 + 况.
+ * 2. **The constituent must be read here as its own entry reads it**, or the
+ *    compound means something else by those characters.
+ * 3. **Only whole words contribute**, never single characters — see
+ *    {@link composedReading} for what allowing them costs.
+ *
+ * What survives all three is a homograph the data cannot see, which is what
+ * {@link LOCALE_COMPOSITION_EXCLUSIONS} is for.
+ */
+export function composeLocaleDeltas(entries) {
+    const index = new ConstituentIndex(entries);
+    // Only a compound containing a delta-carrying word can gain one, and there
+    // are a few hundred such words against 461,623 entries. Enumerating each
+    // entry's substrings against that set costs far less than segmenting every
+    // entry would.
+    const marked = new Set();
+    for (const entry of entries) {
+        if (entry.readings.tw !== undefined && !isSingleCharacter(entry.hans)) {
+            for (const key of keysOf(entry)) {
+                marked.add(key);
+            }
+        }
+    }
+    // Candidates are fixed after the first sweep: composing a delta never makes a
+    // new compound eligible, only changes what an eligible one inherits.
+    const candidates = entries.filter((entry) => entry.readings.tw === undefined &&
+        isComposable(entry) &&
+        hasMarkedConstituent(entry.hans, marked));
+    const settled = new Map();
+    /**
+     * Settle one compound, reporting whether that changed what it reads.
+     */
+    const didCompose = (entry) => {
+        const path = index.segment(entry.hans);
+        const reading = path === undefined ? undefined : composedReading(entry, path, settled);
+        // A composition that comes back identical to the 普通话 reading is not a
+        // delta at all, and one identical to the last round's is not news.
+        if (reading === undefined || isSameReading(reading, entry.readings.cn)) {
+            return false;
+        }
+        const held = settled.get(entry.hans);
+        if (held !== undefined && isSameReading(held, reading)) {
+            return false;
+        }
+        settled.set(entry.hans, reading);
+        return true;
+    };
+    for (let round = 0; round < ROUNDS; round++) {
+        const changed = candidates.filter((entry) => didCompose(entry)).length;
+        if (changed === 0) {
+            break;
+        }
+    }
+    const composed = entries.map((entry) => {
+        const reading = settled.get(entry.hans);
+        return reading === undefined
+            ? entry
+            : { ...entry, readings: { ...entry.readings, tw: reading } };
+    });
+    return { entries: composed, composed: settled.size };
+}
+//# sourceMappingURL=locale.js.map