@kensio/pinyinjs 0.1.0-beta.0 → 1.0.0

package/dist/orthography/idiom-list.js

@@ -0,0 +1,296 @@
+/**
+ * The 成语 GB/T 16159 writes with a hyphen down the middle.
+ *
+ * 6.3.2: a four-syllable 成语 *that can be read as two disyllables* takes a
+ * connecting hyphen — `fēngpíng-làngjìng` — and one that cannot is written
+ * solid: `bùyìlèhū`, `chūqíbùyì`. The condition is prosodic and structural, and
+ * **nothing in the data tracks it.** Conditioning on both halves being
+ * dictionary words fires on 10,202 of the 22,192 four-character words jieba tags
+ * `i` and is uncorrelated with the standard: it fires on 层出不穷 and not on
+ * 风平浪静, which are the standard's own two examples of the same rule, and it
+ * fires on 精神文明 and 凯旋归来, which want a space rather than a hyphen. No
+ * source carries hyphenated pinyin either, so unlike every other claim this
+ * package makes, a 成语 rule could not be scored even if one were written.
+ *
+ * So: a list, on the same argument as [the spaced word list](./word-list.ts).
+ *
+ * **What is in it, and what is not.** Only 成语 whose two halves are each a
+ * self-contained disyllable — parallel halves (风平｜浪静, 千军｜万马), a
+ * disyllable and its predicate (声势｜浩大), or a disyllabic stem with a
+ * reduplicated tail (小心｜翼翼). Those are the ones where 2+2 is beyond doubt.
+ * Anything with a monosyllabic word breaking the rhythm is left out and stays
+ * solid, which is what the standard does with it anyway: 目不转睛 is 目 ｜
+ * 不转睛, 迫不及待 is 迫 ｜ 不及待, 情不自禁 and 显而易见 and 名副其实 the same.
+ * Left out too is anything this file is not sure of, because a missing entry
+ * keeps today's output and a wrong one breaks it.
+ *
+ * It is not a complete 成语 list and is not meant to become one — 22,192 of them
+ * exist. Measured over 88,866 lines of Tatoeba and zh.wikipedia, the 117 here
+ * cover 264 of the 1,737 four-character `i`-tagged words that turn up, or
+ * 15.2%. Most of what is left is either genuinely not 2+2 (据我所知, 一无所知,
+ * 心不在焉, 难以置信) or not a 成语 at all (非常感谢, 重新启动, 可口可乐), which
+ * is the shape a curated list should have: the cases it declines are mostly
+ * cases it should decline.
+ */
+/**
+ * 简体 forms, grouped by why the halves are halves.
+ */
+const HANS = [
+    // ── Parallel halves: 联合式, the clearest case ─────────────
+    "风平浪静",
+    "山清水秀",
+    "千军万马",
+    "千家万户",
+    "千方百计",
+    "四面八方",
+    "四通八达",
+    "七零八落",
+    "大街小巷",
+    "成千上万",
+    "生死存亡",
+    "你死我活",
+    "翻来覆去",
+    "无影无踪",
+    "无声无息",
+    "自由自在",
+    "自言自语",
+    "自给自足",
+    "全心全意",
+    "一心一意",
+    "平起平坐",
+    "不慌不忙",
+    "所作所为",
+    "咬牙切齿",
+    "手忙脚乱",
+    "胡思乱想",
+    "横冲直撞",
+    "歌功颂德",
+    "耀武扬威",
+    "拨乱反正",
+    "提心吊胆",
+    "前仆后继",
+    "轻描淡写",
+    "斩钉截铁",
+    "讨价还价",
+    "粉身碎骨",
+    "惊天动地",
+    "大惊小怪",
+    "三心二意",
+    "爱憎分明",
+    "光明磊落",
+    "精疲力尽",
+    "新陈代谢",
+    "人杰地灵",
+    "错综复杂",
+    "心甘情愿",
+    "触目惊心",
+    "遍体鳞伤",
+    "气急败坏",
+    "大声疾呼",
+    "欢声雷动",
+    "星罗棋布",
+    "卧薪尝胆",
+    "目瞪口呆",
+    "赤手空拳",
+    "破口大骂",
+    "束手无策",
+    "大惊失色",
+    "手足无措",
+    "针锋相对",
+    "全力以赴",
+    "举世闻名",
+    "全军覆没",
+    "家喻户晓",
+    "想方设法",
+    "独立自主",
+    "不由自主",
+    "自力更生",
+    "艰苦奋斗",
+    "心满意足",
+    "不知不觉",
+    "心旷神怡",
+    "热血沸腾",
+    "光彩夺目",
+    "丰富多彩",
+    "慷慨激昂",
+    "声势浩大",
+    "兴高采烈",
+    "恶有恶报",
+    "开门见山",
+    "粗心大意",
+    "一见钟情",
+    "截然不同",
+    "独一无二",
+    "与众不同",
+    "大错特错",
+    "有朝一日",
+    "吹毛求疵",
+    "沉默寡言",
+    "鸡毛蒜皮",
+    "胡说八道",
+    "了如指掌",
+    "无家可归",
+    "乱七八糟",
+    "引人注目",
+    "大吃一惊",
+    "竭尽全力",
+    "五脏六腑",
+    "直截了当",
+    "出类拔萃",
+    "水落石出",
+    "倾盆大雨",
+    "完美无瑕",
+    "层出不穷",
+    // ── A disyllable with a reduplicated tail: ABCC ────────────
+    "小心翼翼",
+    "威风凛凛",
+    "忠心耿耿",
+    "杀气腾腾",
+    "兴致勃勃",
+    "气喘吁吁",
+    "忧心忡忡",
+    // ── A reduplicated head with a disyllable: AABC ────────────
+    "咄咄逼人",
+    "滔滔不绝",
+    "念念不忘",
+    "源源不断",
+    "息息相关",
+    "面面相觑",
+];
+/**
+ * 繁體 spellings of the same 成语, where they differ.
+ *
+ * Listed rather than converted, because 繁體 is a key in its own right
+ * everywhere else in this package and routing it through 简体 to look something
+ * up is exactly what SCRIPTS-AND-LOCALES.md refuses to do. Every entry is
+ * checked against the dictionary by a test, so a spelling that is not a real
+ * word cannot sit here unnoticed.
+ */
+const HANT = [
+    "風平浪靜",
+    "山清水秀",
+    "千軍萬馬",
+    "千家萬戶",
+    "千方百計",
+    "四面八方",
+    "四通八達",
+    "七零八落",
+    "大街小巷",
+    "成千上萬",
+    "生死存亡",
+    "你死我活",
+    "翻來覆去",
+    "無影無蹤",
+    "無聲無息",
+    "自由自在",
+    "自言自語",
+    "自給自足",
+    "全心全意",
+    "一心一意",
+    "平起平坐",
+    "不慌不忙",
+    "所作所為",
+    "咬牙切齒",
+    "手忙腳亂",
+    "胡思亂想",
+    "橫衝直撞",
+    "歌功頌德",
+    "耀武揚威",
+    "撥亂反正",
+    "提心吊膽",
+    "前仆後繼",
+    "輕描淡寫",
+    "斬釘截鐵",
+    "討價還價",
+    "粉身碎骨",
+    "驚天動地",
+    "大驚小怪",
+    "三心二意",
+    "愛憎分明",
+    "光明磊落",
+    "精疲力盡",
+    "新陳代謝",
+    "人傑地靈",
+    "錯綜複雜",
+    "心甘情願",
+    "觸目驚心",
+    "遍體鱗傷",
+    "氣急敗壞",
+    "大聲疾呼",
+    "歡聲雷動",
+    "星羅棋布",
+    "臥薪嘗膽",
+    "目瞪口呆",
+    "赤手空拳",
+    "破口大罵",
+    "束手無策",
+    "大驚失色",
+    "手足無措",
+    "針鋒相對",
+    "全力以赴",
+    "舉世聞名",
+    "全軍覆沒",
+    "家喻戶曉",
+    "想方設法",
+    "獨立自主",
+    "不由自主",
+    "自力更生",
+    "艱苦奮鬥",
+    "心滿意足",
+    "不知不覺",
+    "心曠神怡",
+    "熱血沸騰",
+    "光彩奪目",
+    "豐富多彩",
+    "慷慨激昂",
+    "聲勢浩大",
+    "興高采烈",
+    "惡有惡報",
+    "開門見山",
+    "粗心大意",
+    "一見鍾情",
+    "截然不同",
+    "獨一無二",
+    "與眾不同",
+    "大錯特錯",
+    "有朝一日",
+    "吹毛求疵",
+    "沉默寡言",
+    "雞毛蒜皮",
+    "胡說八道",
+    "瞭如指掌",
+    "無家可歸",
+    "亂七八糟",
+    "引人注目",
+    "大吃一驚",
+    "竭盡全力",
+    "五臟六腑",
+    "直截了當",
+    "出類拔萃",
+    "水落石出",
+    "傾盆大雨",
+    "完美無瑕",
+    "層出不窮",
+    "小心翼翼",
+    "威風凜凜",
+    "忠心耿耿",
+    "殺氣騰騰",
+    "興致勃勃",
+    "氣喘吁吁",
+    "憂心忡忡",
+    "咄咄逼人",
+    "滔滔不絕",
+    "念念不忘",
+    "源源不斷",
+    "息息相關",
+    "面面相覷",
+];
+/**
+ * Every 成语 written with the 2+2 hyphen, paired across the two scripts.
+ */
+export const HYPHENATED_IDIOMS = HANS.map((hans, at) => ({ hans, hant: HANT[at] ?? "" }));
+/**
+ * Every spelling that takes the hyphen, which is how the rule asks.
+ */
+export const HYPHENATED_IDIOM_FORMS = new Set(HYPHENATED_IDIOMS.flatMap((idiom) => [idiom.hans, idiom.hant]));
+//# sourceMappingURL=idiom-list.js.map

package/dist/orthography/idiom-list.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"idiom-list.js","sourceRoot":"","sources":["../../src/orthography/idiom-list.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AAEH;;GAEG;AACH,MAAM,IAAI,GAAsB;IAC9B,2DAA2D;IAC3D,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IAEN,8DAA8D;IAC9D,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IAEN,8DAA8D;IAC9D,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;CACP,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,IAAI,GAAsB;IAC9B,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;CACP,CAAC;AAgBF;;GAEG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAA6B,IAAI,CAAC,GAAG,CACjE,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,CAC/C,CAAC;AAEF;;GAEG;AACH,MAAM,CAAC,MAAM,sBAAsB,GAAwB,IAAI,GAAG,CAChE,iBAAiB,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAC/D,CAAC"}

package/dist/orthography/idioms.d.ts

@@ -0,0 +1,18 @@
+import { type GroupingRule } from "./rule.js";
+/**
+ * The 成语 hyphen, from the list, since no rule reaches it.
+ *
+ * GB/T 16159 6.3.2 hyphenates a four-syllable 成语 that can be read as two
+ * disyllables and writes the rest solid. Which of the two an idiom is cannot be
+ * derived from anything this package holds — see
+ * [the list](./idiom-list.ts) for the measurement — so the list is the rule's
+ * whole evidence, and an idiom that is not on it is written the way it was
+ * before: solid, which is what the standard does with the ones that cannot be
+ * halved.
+ *
+ * Every entry decodes as a single four-syllable word, so this only ever cuts
+ * something the dictionary already agreed was one word, and cuts it where the
+ * list says.
+ */
+export declare const IDIOM_HYPHENS: GroupingRule;
+//# sourceMappingURL=idioms.d.ts.map

package/dist/orthography/idioms.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"idioms.d.ts","sourceRoot":"","sources":["../../src/orthography/idioms.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,YAAY,EAAa,MAAM,WAAW,CAAC;AAEzD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,aAAa,EAAE,YAM3B,CAAC"}

package/dist/orthography/idioms.js

@@ -0,0 +1,22 @@
+import { HYPHENATED_IDIOM_FORMS } from "./idiom-list.js";
+import { hyphenate } from "./rule.js";
+/**
+ * The 成语 hyphen, from the list, since no rule reaches it.
+ *
+ * GB/T 16159 6.3.2 hyphenates a four-syllable 成语 that can be read as two
+ * disyllables and writes the rest solid. Which of the two an idiom is cannot be
+ * derived from anything this package holds — see
+ * [the list](./idiom-list.ts) for the measurement — so the list is the rule's
+ * whole evidence, and an idiom that is not on it is written the way it was
+ * before: solid, which is what the standard does with the ones that cannot be
+ * halved.
+ *
+ * Every entry decodes as a single four-syllable word, so this only ever cuts
+ * something the dictionary already agreed was one word, and cuts it where the
+ * list says.
+ */
+export const IDIOM_HYPHENS = {
+    name: "idiom-hyphens",
+    apply: (words) => words.flatMap((word) => HYPHENATED_IDIOM_FORMS.has(word.text) ? hyphenate(word, 2) : [word]),
+};
+//# sourceMappingURL=idioms.js.map

package/dist/orthography/idioms.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"idioms.js","sourceRoot":"","sources":["../../src/orthography/idioms.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,sBAAsB,EAAE,MAAM,iBAAiB,CAAC;AACzD,OAAO,EAAqB,SAAS,EAAE,MAAM,WAAW,CAAC;AAEzD;;;;;;;;;;;;;;GAcG;AACH,MAAM,CAAC,MAAM,aAAa,GAAiB;IACzC,IAAI,EAAE,eAAe;IACrB,KAAK,EAAE,CAAC,KAAK,EAAE,EAAE,CACf,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACrB,sBAAsB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CACpE;CACJ,CAAC"}

package/dist/orthography/reduplication.d.ts

@@ -0,0 +1,52 @@
+import { type GroupingRule } from "./rule.js";
+/**
+ * The 重叠 hyphen inside a four-character AABB word.
+ *
+ * GB/T 16159 6.1.3 writes 干干净净 as `gāngān-jìngjìng`: one word, with the
+ * boundary between its halves marked. The shape *is* the evidence — a
+ * four-character word whose first two characters double and whose last two
+ * double is a reduplication and essentially nothing else — so this needs no
+ * tag, which matters given that two thirds of the dictionary carries none.
+ *
+ * Two conditions keep it honest, both sized against the whole dictionary. The
+ * word must have reached the rule as **one** decoded word: 爸爸妈妈 and 哥哥姐姐
+ * decode as two words each and are two words, `bàba māma`, not a reduplication
+ * of 爸妈. And a proper noun is left alone, since a name that happens to be
+ * shaped this way is not a reduplication of anything.
+ *
+ * Fires on 1,541 of the dictionary's 1,590 AABB keys; the 49 it holds back are
+ * all jieba proper-noun tags that CC-CEDICT did not veto — 斯斯文文, 老老少少,
+ * 正正经经 — which are mis-tagged rather than misjudged here, and are already
+ * wrong in their capitals for the same reason.
+ *
+ * Over 711,000 decoded words of Tatoeba and zh.wikipedia it fires 66 times,
+ * and all 66 are reduplications: 时时刻刻, 马马虎虎, 形形色色, 轰轰烈烈. What
+ * it misses is the same shape arriving as two words — 43 spans, of which about
+ * 16 are reduplications the decode happened to split (匆匆忙忙, 来来往往) and
+ * the rest are 爸爸妈妈, 爷爷奶奶 and 爸爸刚刚, where joining would be wrong.
+ * Nothing separates the two halves of that mixture: 匆匆 and 爸爸 are both
+ * words, and so are 匆忙 and 爸妈.
+ */
+export declare const AABB_REDUPLICATION: GroupingRule;
+/**
+ * The 重叠 hyphen between a repeated two-character word: 研究研究.
+ *
+ * GB/T 16159 writes verb reduplication `yánjiū-yánjiū`, and this is the
+ * productive half of 重叠: 研究研究 is not a dictionary entry and never will be,
+ * so it arrives as two decoded words rather than as one word to be split.
+ * Marking the second is all that is needed, and no reading changes.
+ *
+ * The ABAB words the dictionary *does* hold are deliberately not touched: they
+ * decode as one word, and 50 keys are shaped this way with 哔哩哔哩 and
+ * 达姆达姆 among them, where a hyphen would be asserting a reduplication that
+ * is really a transliteration.
+ *
+ * Measured over the same 711,000 decoded words, it fires 54 times and is right
+ * 46 of them — 休息休息, 一天一天, 考虑考虑, 好久好久. The eight misfires are
+ * all one word ending a clause and the same word starting the next
+ * (告诉我们｜我们在哪里, 不喜欢｜喜欢不问问题的), or a noun said twice to
+ * address someone (大象大象). Nothing short of syntax separates those from
+ * 讨论讨论, and the eight are a spacing error where all 54 were one before.
+ */
+export declare const ABAB_REDUPLICATION: GroupingRule;
+//# sourceMappingURL=reduplication.d.ts.map

package/dist/orthography/reduplication.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"reduplication.d.ts","sourceRoot":"","sources":["../../src/orthography/reduplication.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,KAAK,YAAY,EAAa,MAAM,WAAW,CAAC;AAkBzD;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,eAAO,MAAM,kBAAkB,EAAE,YAQhC,CAAC;AAyBF;;;;;;;;;;;;;;;;;;;GAmBG;AACH,eAAO,MAAM,kBAAkB,EAAE,YAShC,CAAC"}

package/dist/orthography/reduplication.js

@@ -0,0 +1,99 @@
+import { toCharacters } from "../script/characters.js";
+import { hyphenate } from "./rule.js";
+/**
+ * Whether four characters repeat as AABB, with A and B different.
+ *
+ * 干干净净 and 高高兴兴 do; 一一得一 does not, and neither does a 叠字 name
+ * like 丁丁 doubled, since A and B have to differ for there to be two halves.
+ */
+function isAabb(characters) {
+    const [first, second, third, fourth] = characters;
+    return (characters.length === 4 &&
+        first === second &&
+        third === fourth &&
+        first !== third);
+}
+/**
+ * The 重叠 hyphen inside a four-character AABB word.
+ *
+ * GB/T 16159 6.1.3 writes 干干净净 as `gāngān-jìngjìng`: one word, with the
+ * boundary between its halves marked. The shape *is* the evidence — a
+ * four-character word whose first two characters double and whose last two
+ * double is a reduplication and essentially nothing else — so this needs no
+ * tag, which matters given that two thirds of the dictionary carries none.
+ *
+ * Two conditions keep it honest, both sized against the whole dictionary. The
+ * word must have reached the rule as **one** decoded word: 爸爸妈妈 and 哥哥姐姐
+ * decode as two words each and are two words, `bàba māma`, not a reduplication
+ * of 爸妈. And a proper noun is left alone, since a name that happens to be
+ * shaped this way is not a reduplication of anything.
+ *
+ * Fires on 1,541 of the dictionary's 1,590 AABB keys; the 49 it holds back are
+ * all jieba proper-noun tags that CC-CEDICT did not veto — 斯斯文文, 老老少少,
+ * 正正经经 — which are mis-tagged rather than misjudged here, and are already
+ * wrong in their capitals for the same reason.
+ *
+ * Over 711,000 decoded words of Tatoeba and zh.wikipedia it fires 66 times,
+ * and all 66 are reduplications: 时时刻刻, 马马虎虎, 形形色色, 轰轰烈烈. What
+ * it misses is the same shape arriving as two words — 43 spans, of which about
+ * 16 are reduplications the decode happened to split (匆匆忙忙, 来来往往) and
+ * the rest are 爸爸妈妈, 爷爷奶奶 and 爸爸刚刚, where joining would be wrong.
+ * Nothing separates the two halves of that mixture: 匆匆 and 爸爸 are both
+ * words, and so are 匆忙 and 爸妈.
+ */
+export const AABB_REDUPLICATION = {
+    name: "aabb-reduplication",
+    apply: (words) => words.flatMap((word) => word.isProperNoun || !isAabb(toCharacters(word.text))
+        ? [word]
+        : hyphenate(word, 2)),
+};
+/**
+ * Whether two words are the same word repeated: 研究研究.
+ *
+ * Two characters each, because a one-character repeat is a different rule —
+ * 看看 is `kànkan`, written solid with a neutral second syllable — and because
+ * a repeated single character is as likely to be two separate words.
+ *
+ * Nothing is asked of the part of speech, which was measured rather than
+ * assumed: excluding noun-tagged repeats removes three misfires — 战争 falling
+ * either side of 克里米亚战争｜战争中, and 大象大象 addressing an elephant
+ * twice — at the cost of four real reduplications, since jieba tags 商量,
+ * 多谢 and 拜托 as nouns. Fewer right answers for a better-looking rule.
+ */
+function isRepeat(word, previous) {
+    return (word.text === previous.text &&
+        word.isKnown &&
+        !word.isProperNoun &&
+        toCharacters(word.text).length === 2 &&
+        word.separator === undefined);
+}
+/**
+ * The 重叠 hyphen between a repeated two-character word: 研究研究.
+ *
+ * GB/T 16159 writes verb reduplication `yánjiū-yánjiū`, and this is the
+ * productive half of 重叠: 研究研究 is not a dictionary entry and never will be,
+ * so it arrives as two decoded words rather than as one word to be split.
+ * Marking the second is all that is needed, and no reading changes.
+ *
+ * The ABAB words the dictionary *does* hold are deliberately not touched: they
+ * decode as one word, and 50 keys are shaped this way with 哔哩哔哩 and
+ * 达姆达姆 among them, where a hyphen would be asserting a reduplication that
+ * is really a transliteration.
+ *
+ * Measured over the same 711,000 decoded words, it fires 54 times and is right
+ * 46 of them — 休息休息, 一天一天, 考虑考虑, 好久好久. The eight misfires are
+ * all one word ending a clause and the same word starting the next
+ * (告诉我们｜我们在哪里, 不喜欢｜喜欢不问问题的), or a noun said twice to
+ * address someone (大象大象). Nothing short of syntax separates those from
+ * 讨论讨论, and the eight are a spacing error where all 54 were one before.
+ */
+export const ABAB_REDUPLICATION = {
+    name: "abab-reduplication",
+    apply: (words) => words.map((word, at) => {
+        const previous = words[at - 1];
+        return previous !== undefined && isRepeat(word, previous)
+            ? { ...word, separator: "-" }
+            : word;
+    }),
+};
+//# sourceMappingURL=reduplication.js.map

package/dist/orthography/reduplication.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"reduplication.js","sourceRoot":"","sources":["../../src/orthography/reduplication.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAqB,SAAS,EAAE,MAAM,WAAW,CAAC;AAEzD;;;;;GAKG;AACH,SAAS,MAAM,CAAC,UAA6B;IAC3C,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,GAAG,UAAU,CAAC;IAClD,OAAO,CACL,UAAU,CAAC,MAAM,KAAK,CAAC;QACvB,KAAK,KAAK,MAAM;QAChB,KAAK,KAAK,MAAM;QAChB,KAAK,KAAK,KAAK,CAChB,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAiB;IAC9C,IAAI,EAAE,oBAAoB;IAC1B,KAAK,EAAE,CAAC,KAAK,EAAE,EAAE,CACf,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACrB,IAAI,CAAC,YAAY,IAAI,CAAC,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnD,CAAC,CAAC,CAAC,IAAI,CAAC;QACR,CAAC,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CACvB;CACJ,CAAC;AAEF;;;;;;;;;;;;GAYG;AACH,SAAS,QAAQ,CAAC,IAAiB,EAAE,QAAqB;IACxD,OAAO,CACL,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,IAAI;QAC3B,IAAI,CAAC,OAAO;QACZ,CAAC,IAAI,CAAC,YAAY;QAClB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC;QACpC,IAAI,CAAC,SAAS,KAAK,SAAS,CAC7B,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAiB;IAC9C,IAAI,EAAE,oBAAoB;IAC1B,KAAK,EAAE,CAAC,KAAK,EAAE,EAAE,CACf,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE;QACrB,MAAM,QAAQ,GAAG,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QAC/B,OAAO,QAAQ,KAAK,SAAS,IAAI,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;YACvD,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,SAAS,EAAE,GAAY,EAAE;YACtC,CAAC,CAAC,IAAI,CAAC;IACX,CAAC,CAAC;CACL,CAAC"}

package/dist/orthography/rule.d.ts

@@ -0,0 +1,50 @@
+import type { DecodedWord } from "../decode/word.js";
+import type { Dictionary } from "../dictionary/dictionary.js";
+/**
+ * One 分词连写 rule, rewriting the words a decode produced.
+ *
+ * Typed and individually testable, which is the point ALGORITHM.md makes about
+ * rules: the old project patched the output string with regexes, so its rules
+ * were order-dependent and could not be tested apart from each other.
+ */
+export interface GroupingRule {
+    readonly name: string;
+    readonly apply: (words: readonly DecodedWord[], dictionary: Dictionary) => readonly DecodedWord[];
+}
+/**
+ * Join two decoded words into one.
+ *
+ * The first word's flags carry, because it is the head: 看了 is the verb 看
+ * with an aspect particle stuck to it, not a new word of its own.
+ */
+export declare function join(head: DecodedWord, tail: DecodedWord): DecodedWord;
+/**
+ * Write a word as two halves with a hyphen between them.
+ *
+ * The 重叠 and 成语 hyphens are the same operation on different evidence: cut
+ * the word, mark the second half, and leave the reading exactly as it was. The
+ * word is returned untouched where its reading cannot be cut — 儿化 has no
+ * halves.
+ */
+export declare function hyphenate(word: DecodedWord, at: number): readonly DecodedWord[];
+/**
+ * Divide a decoded word at every one of a list of character offsets.
+ *
+ * {@link splitAt} generalised, for the proper names whose parts CC-CEDICT marks
+ * more than two of: 上海交通大学 divides at 2 and 4, and cutting only at the
+ * first would leave `Shànghǎi Jiāotōngdàxué`.
+ *
+ * The word comes back whole where its reading cannot be cut — the same 儿化
+ * condition {@link splitAt} has — or where an offset is out of order or outside
+ * the word, since a boundary that does not describe this word cannot be applied
+ * to it.
+ */
+export declare function divideAt(word: DecodedWord, offsets: readonly number[]): readonly DecodedWord[];
+/**
+ * Split a decoded word into two at a character offset.
+ *
+ * Only possible where the reading has one syllable per character, which 儿化
+ * does not: there is no way to cut `wánr` in half.
+ */
+export declare function splitAt(word: DecodedWord, at: number): readonly [DecodedWord, DecodedWord] | undefined;
+//# sourceMappingURL=rule.d.ts.map

package/dist/orthography/rule.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"rule.d.ts","sourceRoot":"","sources":["../../src/orthography/rule.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAG9D;;;;;;GAMG;AACH,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,KAAK,EAAE,CACd,KAAK,EAAE,SAAS,WAAW,EAAE,EAC7B,UAAU,EAAE,UAAU,KACnB,SAAS,WAAW,EAAE,CAAC;CAC7B;AAED;;;;;GAKG;AACH,wBAAgB,IAAI,CAAC,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,WAAW,GAAG,WAAW,CAStE;AAED;;;;;;;GAOG;AACH,wBAAgB,SAAS,CACvB,IAAI,EAAE,WAAW,EACjB,EAAE,EAAE,MAAM,GACT,SAAS,WAAW,EAAE,CAOxB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,QAAQ,CACtB,IAAI,EAAE,WAAW,EACjB,OAAO,EAAE,SAAS,MAAM,EAAE,GACzB,SAAS,WAAW,EAAE,CA2BxB;AAED;;;;;GAKG;AACH,wBAAgB,OAAO,CACrB,IAAI,EAAE,WAAW,EACjB,EAAE,EAAE,MAAM,GACT,SAAS,CAAC,WAAW,EAAE,WAAW,CAAC,GAAG,SAAS,CAsBjD"}

package/dist/orthography/rule.js

@@ -0,0 +1,101 @@
+import { toCharacters } from "../script/characters.js";
+/**
+ * Join two decoded words into one.
+ *
+ * The first word's flags carry, because it is the head: 看了 is the verb 看
+ * with an aspect particle stuck to it, not a new word of its own.
+ */
+export function join(head, tail) {
+    return {
+        text: head.text + tail.text,
+        reading: [...head.reading, ...tail.reading],
+        isProperNoun: head.isProperNoun,
+        partOfSpeech: head.partOfSpeech,
+        isKnown: head.isKnown && tail.isKnown,
+        ...(head.separator !== undefined && { separator: head.separator }),
+    };
+}
+/**
+ * Write a word as two halves with a hyphen between them.
+ *
+ * The 重叠 and 成语 hyphens are the same operation on different evidence: cut
+ * the word, mark the second half, and leave the reading exactly as it was. The
+ * word is returned untouched where its reading cannot be cut — 儿化 has no
+ * halves.
+ */
+export function hyphenate(word, at) {
+    const split = splitAt(word, at);
+    if (split === undefined) {
+        return [word];
+    }
+    const [head, tail] = split;
+    return [head, { ...tail, separator: "-" }];
+}
+/**
+ * Divide a decoded word at every one of a list of character offsets.
+ *
+ * {@link splitAt} generalised, for the proper names whose parts CC-CEDICT marks
+ * more than two of: 上海交通大学 divides at 2 and 4, and cutting only at the
+ * first would leave `Shànghǎi Jiāotōngdàxué`.
+ *
+ * The word comes back whole where its reading cannot be cut — the same 儿化
+ * condition {@link splitAt} has — or where an offset is out of order or outside
+ * the word, since a boundary that does not describe this word cannot be applied
+ * to it.
+ */
+export function divideAt(word, offsets) {
+    const characters = toCharacters(word.text);
+    if (word.reading.length !== characters.length) {
+        return [word];
+    }
+    const cuts = [0, ...offsets, characters.length];
+    const isOrdered = cuts.every((at, index) => index === 0 || at > (cuts[index - 1] ?? 0));
+    if (!isOrdered || (offsets.at(-1) ?? 0) >= characters.length) {
+        return [word];
+    }
+    return cuts.slice(0, -1).map((from, index) => {
+        const to = cuts[index + 1] ?? characters.length;
+        const part = {
+            text: characters.slice(from, to).join(""),
+            reading: word.reading.slice(from, to),
+            isProperNoun: word.isProperNoun,
+            partOfSpeech: word.partOfSpeech,
+            isKnown: word.isKnown,
+        };
+        // Only the head keeps what was written in front of the word: every later
+        // part is written against the part before it rather than against that.
+        return index === 0 && word.separator !== undefined
+            ? { ...part, separator: word.separator }
+            : part;
+    });
+}
+/**
+ * Split a decoded word into two at a character offset.
+ *
+ * Only possible where the reading has one syllable per character, which 儿化
+ * does not: there is no way to cut `wánr` in half.
+ */
+export function splitAt(word, at) {
+    const characters = toCharacters(word.text);
+    if (word.reading.length !== characters.length) {
+        return undefined;
+    }
+    const part = (from, to) => ({
+        text: characters.slice(from, to).join(""),
+        reading: word.reading.slice(from, to),
+        isProperNoun: word.isProperNoun,
+        partOfSpeech: word.partOfSpeech,
+        isKnown: word.isKnown,
+    });
+    const head = part(0, at);
+    return [
+        // Only the head keeps whatever was written in front of the word, since the
+        // tail is now written against the head rather than against what came
+        // before it.
+        word.separator === undefined
+            ? head
+            : { ...head, separator: word.separator },
+        part(at, characters.length),
+    ];
+}
+//# sourceMappingURL=rule.js.map