@kensio/pinyinjs 0.1.0-beta.0 → 1.0.0

package/dist/decode/decode.js.map

@@ -1 +1 @@
-{"version":3,"file":"decode.js","sourceRoot":"","sources":["../../src/decode/decode.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAmB,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,YAAY,EAAE,SAAS,EAAgB,MAAM,cAAc,CAAC;AACrE,OAAO,EACL,SAAS,EACT,eAAe,EAGf,YAAY,EACZ,OAAO,GACR,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAGtE;;;;;;;GAOG;AACH,MAAM,UAAU,cAAc,CAC5B,OAAgB,EAChB,UAA6B;IAE7B,MAAM,IAAI,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC;IAChC,MAAM,KAAK,GAAkB,EAAE,CAAC;IAEhC,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,CAAC,GAAG,YAAY,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;YAClD,SAAS;QACX,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,WAAW,CAAC,EAAE,CAAC;YAChE,KAAK,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;QAC/B,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,aAAa,CAAC,OAAgB;IAC5C,OAAO,YAAY,CAAC,OAAO,EAAE,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,GAAG,CACzE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CACpB,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,UAAU,CACjB,KAAsB,EACtB,UAA6B;IAE7B,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACvD,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAChE,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,IAAI,GAAW,EAAE,CAAC;IAEtB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3C,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClB,IAAI,GAAG,EAAE,CAAC;QACZ,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACpB,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,SAAS,QAAQ,CACf,UAAsB,EACtB,OAAgB,EAChB,KAA6B;IAE7B,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1B,wEAAwE;IACxE,IAAI,KAAK,KAAK,SAAS,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACpE,MAAM,KAAK,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAEtC,OAAO;QACL,IAAI;QACJ,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;QACnD,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,KAAK;QAC1C,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,EAAE;QACvC,OAAO,EAAE,KAAK,KAAK,SAAS;KAC7B,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,SAAS,CACvB,UAAsB,EACtB,GAAW;IAEX,MAAM,OAAO,GAAG,YAAY,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;IAC9C,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAChE,OAAO,UAAU,CAAC,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAC7D,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC,CACrC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,eAAe,CAC7B,UAAsB,EACtB,GAAW;IAEX,MAAM,OAAO,GAAG,YAAY,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;IAC9C,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAE/D,OAAO,UAAU,CAAa,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC3E,IAAI,EAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC;QAC1C,UAAU,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACjC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;YACtB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,YAAY,EAAE,IAAI,CAAC,YAAY;SAChC,CAAC,CAAC,CACJ;KACF,CAAC,CAAC,CAAC;AACN,CAAC"}
+{"version":3,"file":"decode.js","sourceRoot":"","sources":["../../src/decode/decode.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAmB,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,YAAY,EAAE,SAAS,EAAgB,MAAM,cAAc,CAAC;AACrE,OAAO,EACL,SAAS,EACT,eAAe,EAGf,YAAY,EACZ,OAAO,GACR,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACnD,OAAO,EAAE,cAAc,EAAiB,MAAM,YAAY,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAGtE;;;;;;;GAOG;AACH,MAAM,UAAU,cAAc,CAC5B,OAAgB,EAChB,UAA6B;IAE7B,MAAM,IAAI,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC;IAChC,MAAM,KAAK,GAAkB,EAAE,CAAC;IAEhC,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,CAAC,GAAG,YAAY,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;YAClD,SAAS;QACX,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,WAAW,CAAC,EAAE,CAAC;YAChE,KAAK,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;QAC/B,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,aAAa,CAAC,OAAgB;IAC5C,OAAO,YAAY,CAAC,OAAO,EAAE,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,GAAG,CACzE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CACpB,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,UAAU,CACjB,KAAsB,EACtB,UAA6B;IAE7B,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACvD,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAChE,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,IAAI,GAAW,EAAE,CAAC;IAEtB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3C,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClB,IAAI,GAAG,EAAE,CAAC;QACZ,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACpB,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,SAAS,QAAQ,CACf,UAAsB,EACtB,OAAgB,EAChB,KAA6B;IAE7B,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1B,wEAAwE;IACxE,IAAI,KAAK,KAAK,SAAS,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IACjE,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACpE,MAAM,KAAK,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAEtC,OAAO;QACL,IAAI;QACJ,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;QACnD,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,KAAK;QAC1C,YAAY,EAAE,KAAK,EAAE,YAAY,IAAI,EAAE;QACvC,OAAO,EAAE,KAAK,KAAK,SAAS;KAC7B,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,YAAY,CACnB,UAAsB,EACtB,GAAW,EACX,KAA0B;IAE1B,OAAO,cAAc,CAAC,YAAY,CAAC,UAAU,EAAE,GAAG,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,CAAC;AAC1E,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,SAAS,CACvB,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa;IAE1C,MAAM,OAAO,GAAG,YAAY,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,CAAC,CAAC;IACrD,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAChE,OAAO,UAAU,CAAC,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAC7D,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC,CACrC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,eAAe,CAC7B,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa;IAE1C,MAAM,OAAO,GAAG,YAAY,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,CAAC,CAAC;IACrD,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAE/D,OAAO,UAAU,CAAa,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC3E,IAAI,EAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC;QAC1C,UAAU,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACjC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;YACtB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,YAAY,EAAE,IAAI,CAAC,YAAY;SAChC,CAAC,CAAC,CACJ;KACF,CAAC,CAAC,CAAC;AACN,CAAC"}

package/dist/decode/reading-rules.d.ts

@@ -0,0 +1,56 @@
+import { type EdgeRule } from "./rules.js";
+/**
+ * 得 read as `děi` where it is a modal rather than a particle.
+ *
+ * One character, three readings, and the dictionary can only carry a default:
+ * 得 is stored `de` — the structural particle, which is far the commonest — with
+ * `dé` and `děi` as alternates. Nothing in the cost model can prefer an
+ * alternate at a position no word covers, so 我得走了 came out `wǒ de zǒule`
+ * and every other modal 得 with it. This is the case ROADMAP.md was waiting for
+ * before building stage 4: neither harness could see it, since the gold corpus
+ * had only 觉得 and CPP's 40 得 cases are labelled `dé` or `de` and never `děi`.
+ *
+ * The context is what separates the three, and only one side of it carries
+ * information. The particle 得 always follows the verb or adjective it attaches
+ * to (说得好, 跑得快, 累得要命), so a 得 with a *pronoun, adverb or time word*
+ * in front of it is not that particle. Requiring a verb phrase after it as well
+ * keeps 得了感冒 and 得了奖 — `dé`, and a real reading of the same character —
+ * out of it.
+ *
+ * Conditioning on what precedes rather than on what follows was measured, not
+ * assumed: "the word before is not a verb" fires 309 times over 88,866 lines of
+ * Tatoeba and zh.wikipedia and is wrong about four times in five, because
+ * jieba tags 下 as `f`, 学 as `n` and 飞 as nothing at all, and 雨下得很大 is a
+ * particle whatever the tag says. Naming the small closed set of things that
+ * can precede a modal is the same rule from the other side, and over the same
+ * text it fires 126 times and is right 122: the four misses are literary 得
+ * read `dé` after an adverb — 未必得喜悦, 暂得于己, 仅得约, 未得迁还家乡 —
+ * which nothing short of semantics separates from 必须得承认.
+ */
+export declare const MODAL_DE: EdgeRule;
+/**
+ * 儿 does not stand on its own where the dictionary attests the 儿化.
+ *
+ * 儿化 is a per-word dictionary fact taken from CC-CEDICT's explicit `r5`
+ * token, and 2,009 of the 2,067 keys ending in 儿 carry it. What that leaves is
+ * an edge: 这边儿, 上边儿 and 旁边儿 are all listed and 那边儿 is not, so
+ * 我得跑到那边儿去 read the 儿 as a syllable of its own — `nàbian ér` — and the
+ * decode was choosing between 那边 plus 儿 and 那 plus 边儿 on frequency alone.
+ *
+ * This forbids the bare 儿 edge when the character in front of it makes an
+ * attested 儿化 word, so the only paths left are the ones that absorb it. That
+ * asserts nothing the dictionary does not already say — 边儿 is `biānr` because
+ * CC-CEDICT says so — which is what separates it from synthesising 儿化 with a
+ * suffix rule, the thing ORTHOGRAPHY.md rejects and the old project got wrong.
+ *
+ * Measured over the same 88,866 lines it fires 7 times, and all 7 are right:
+ * 照片儿, 出门儿, 热点儿, 辣味儿, 晚会儿, 针管儿, 碍事儿. It is deliberately
+ * silent on the 20 bare 儿 with no attested 儿化 behind them, which are 马儿,
+ * 门铃儿 and a row of Mongol names transliterated with 儿.
+ */
+export declare const ATTESTED_ERHUA: EdgeRule;
+/**
+ * The rules the lattice decode applies, in order.
+ */
+export declare const READING_RULES: readonly EdgeRule[];
+//# sourceMappingURL=reading-rules.d.ts.map

package/dist/decode/reading-rules.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"reading-rules.d.ts","sourceRoot":"","sources":["../../src/decode/reading-rules.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,QAAQ,EAId,MAAM,YAAY,CAAC;AAgCpB;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,eAAO,MAAM,QAAQ,EAAE,QActB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;GAmBG;AACH,eAAO,MAAM,cAAc,EAAE,QAY5B,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,aAAa,EAAE,SAAS,QAAQ,EAA+B,CAAC"}

package/dist/decode/reading-rules.js

@@ -0,0 +1,107 @@
+import { tagOf, wordEndingAt, wordStartingAt, } from "./rules.js";
+/**
+ * Whether a syllable is `děi`, the modal reading of 得.
+ *
+ * Compared structurally rather than by spelling it out, since a syllable is
+ * stored as its parts and writing one back is the formatting stage's job.
+ */
+function isDei(syllable) {
+    return (syllable?.initial === "d" && syllable.final === "ei" && syllable.tone === 3);
+}
+/**
+ * What stands in front of a modal 得: a subject, an adverb, or a time word.
+ *
+ * 我得走, 你就得去, 我今天得走. The particle 得 cannot follow any of these,
+ * because it attaches to the predicate it modifies — which is exactly what
+ * makes this the discriminating side of the context.
+ */
+const SUBJECT_TAGS = new Set(["r", "d", "t"]);
+/**
+ * What follows a modal 得: the verb phrase it governs.
+ *
+ * Prepositions and adverbs are in because 我得在八月, 我得先走 and 我得快点 are
+ * all modal, and what follows the 得 in each is the front of a verb phrase
+ * rather than a verb.
+ */
+const GOVERNED_TAGS = /^(?:v|p|d|a)/u;
+/**
+ * 得 read as `děi` where it is a modal rather than a particle.
+ *
+ * One character, three readings, and the dictionary can only carry a default:
+ * 得 is stored `de` — the structural particle, which is far the commonest — with
+ * `dé` and `děi` as alternates. Nothing in the cost model can prefer an
+ * alternate at a position no word covers, so 我得走了 came out `wǒ de zǒule`
+ * and every other modal 得 with it. This is the case ROADMAP.md was waiting for
+ * before building stage 4: neither harness could see it, since the gold corpus
+ * had only 觉得 and CPP's 40 得 cases are labelled `dé` or `de` and never `děi`.
+ *
+ * The context is what separates the three, and only one side of it carries
+ * information. The particle 得 always follows the verb or adjective it attaches
+ * to (说得好, 跑得快, 累得要命), so a 得 with a *pronoun, adverb or time word*
+ * in front of it is not that particle. Requiring a verb phrase after it as well
+ * keeps 得了感冒 and 得了奖 — `dé`, and a real reading of the same character —
+ * out of it.
+ *
+ * Conditioning on what precedes rather than on what follows was measured, not
+ * assumed: "the word before is not a verb" fires 309 times over 88,866 lines of
+ * Tatoeba and zh.wikipedia and is wrong about four times in five, because
+ * jieba tags 下 as `f`, 学 as `n` and 飞 as nothing at all, and 雨下得很大 is a
+ * particle whatever the tag says. Naming the small closed set of things that
+ * can precede a modal is the same rule from the other side, and over the same
+ * text it fires 126 times and is right 122: the four misses are literary 得
+ * read `dé` after an adverb — 未必得喜悦, 暂得于己, 仅得约, 未得迁还家乡 —
+ * which nothing short of semantics separates from 必须得承认.
+ */
+export const MODAL_DE = {
+    name: "modal-de",
+    verdictFor: (context) => {
+        const { edge } = context;
+        if (edge.text !== "得" || !isDei(edge.reading[0])) {
+            return "keep";
+        }
+        const before = wordEndingAt(context, edge.from);
+        const after = wordStartingAt(context, edge.to);
+        return SUBJECT_TAGS.has(tagOf(context, before)) &&
+            GOVERNED_TAGS.test(tagOf(context, after))
+            ? "force"
+            : "keep";
+    },
+};
+/**
+ * 儿 does not stand on its own where the dictionary attests the 儿化.
+ *
+ * 儿化 is a per-word dictionary fact taken from CC-CEDICT's explicit `r5`
+ * token, and 2,009 of the 2,067 keys ending in 儿 carry it. What that leaves is
+ * an edge: 这边儿, 上边儿 and 旁边儿 are all listed and 那边儿 is not, so
+ * 我得跑到那边儿去 read the 儿 as a syllable of its own — `nàbian ér` — and the
+ * decode was choosing between 那边 plus 儿 and 那 plus 边儿 on frequency alone.
+ *
+ * This forbids the bare 儿 edge when the character in front of it makes an
+ * attested 儿化 word, so the only paths left are the ones that absorb it. That
+ * asserts nothing the dictionary does not already say — 边儿 is `biānr` because
+ * CC-CEDICT says so — which is what separates it from synthesising 儿化 with a
+ * suffix rule, the thing ORTHOGRAPHY.md rejects and the old project got wrong.
+ *
+ * Measured over the same 88,866 lines it fires 7 times, and all 7 are right:
+ * 照片儿, 出门儿, 热点儿, 辣味儿, 晚会儿, 针管儿, 碍事儿. It is deliberately
+ * silent on the 20 bare 儿 with no attested 儿化 behind them, which are 马儿,
+ * 门铃儿 and a row of Mongol names transliterated with 儿.
+ */
+export const ATTESTED_ERHUA = {
+    name: "attested-erhua",
+    verdictFor: (context) => {
+        const { characters, edge } = context;
+        if (edge.text !== "儿" || edge.from === 0) {
+            return "keep";
+        }
+        /* c8 ignore next -- the edge starts inside the run it was built from */
+        const absorbed = `${characters[edge.from - 1] ?? ""}儿`;
+        const reading = context.dictionary.lookup(absorbed)?.reading;
+        return (reading?.at(-1)?.erhua ?? false) ? "forbid" : "keep";
+    },
+};
+/**
+ * The rules the lattice decode applies, in order.
+ */
+export const READING_RULES = [MODAL_DE, ATTESTED_ERHUA];
+//# sourceMappingURL=reading-rules.js.map

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

@@ -0,0 +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"}

package/dist/decode/rules.d.ts

@@ -0,0 +1,66 @@
+import type { Dictionary } from "../dictionary/dictionary.js";
+import type { Lattice, LatticeEdge } from "./lattice.js";
+/**
+ * What a rule can say about one edge.
+ *
+ * `forbid` takes the edge out of the lattice; `force` takes every *rival* out,
+ * meaning every other edge covering exactly the same characters. Forcing settles
+ * how a stretch is read without saying anything about how it is segmented, so
+ * the words that overlap it are left to the decode.
+ */
+export type EdgeVerdict = "keep" | "forbid" | "force";
+/**
+ * What a rule is given to judge an edge by.
+ *
+ * The characters of the whole run rather than of the edge, because the
+ * interesting rules are about context: 得 is `de` after a verb and `děi` before
+ * one, and neither is visible from the character on its own.
+ */
+export interface EdgeContext {
+    readonly dictionary: Dictionary;
+    readonly characters: readonly string[];
+    readonly edge: LatticeEdge;
+}
+/**
+ * One typed rule over the lattice's edges — stage 4 of ALGORITHM.md.
+ *
+ * The old project tweaked its output string with regexes, so its rules ran in
+ * an order nobody chose and could not be tested apart from each other. A rule
+ * here is a function from one edge to one verdict: it can be read on its own,
+ * tested on its own, and measured on its own.
+ */
+export interface EdgeRule {
+    readonly name: string;
+    readonly verdictFor: (context: EdgeContext) => EdgeVerdict;
+}
+/**
+ * The word ending at a position, longest first, or undefined.
+ *
+ * What a rule needs when it asks about the word *before* an edge. The decode
+ * has not run yet, so there is no settled answer; the longest dictionary word
+ * ending here is the closest thing available, and it is what the decode will
+ * usually choose, since the reading charge prefers fewer words.
+ */
+export declare function wordEndingAt(context: EdgeContext, at: number, longest?: number): string | undefined;
+/**
+ * The word starting at a position, longest first, or undefined.
+ */
+export declare function wordStartingAt(context: EdgeContext, at: number, longest?: number): string | undefined;
+/**
+ * The part of speech the dictionary gives a word, or the empty string.
+ */
+export declare function tagOf(context: EdgeContext, word: string | undefined): string;
+/**
+ * Apply the rules to a lattice, returning the lattice they leave.
+ *
+ * Stage 4 of the pipeline: the rules run over the built lattice and before any
+ * decoding, so what they change is what the decode gets to choose between
+ * rather than what it has already chosen. Nothing is rewritten — an edge is
+ * kept or it is not — which is what keeps a rule from inventing a reading no
+ * source attests.
+ *
+ * The lattice is returned unchanged if the rules would leave the run with no
+ * path through it.
+ */
+export declare function applyEdgeRules(lattice: Lattice, dictionary: Dictionary, rules: readonly EdgeRule[]): Lattice;
+//# sourceMappingURL=rules.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"rules.d.ts","sourceRoot":"","sources":["../../src/decode/rules.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAC9D,OAAO,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAEzD;;;;;;;GAOG;AACH,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,QAAQ,GAAG,OAAO,CAAC;AAEtD;;;;;;GAMG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,UAAU,EAAE,UAAU,CAAC;IAChC,QAAQ,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;IACvC,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;CAC5B;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,QAAQ;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,UAAU,EAAE,CAAC,OAAO,EAAE,WAAW,KAAK,WAAW,CAAC;CAC5D;AAED;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAC1B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,MAAM,GAAG,SAAS,CAYpB;AAED;;GAEG;AACH,wBAAgB,cAAc,CAC5B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,MAAM,GAAG,SAAS,CAUpB;AAED;;GAEG;AACH,wBAAgB,KAAK,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CAI5E;AA+DD;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAC5B,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,UAAU,EACtB,KAAK,EAAE,SAAS,QAAQ,EAAE,GACzB,OAAO,CAsCT"}

package/dist/decode/rules.js

@@ -0,0 +1,138 @@
+/**
+ * The word ending at a position, longest first, or undefined.
+ *
+ * What a rule needs when it asks about the word *before* an edge. The decode
+ * has not run yet, so there is no settled answer; the longest dictionary word
+ * ending here is the closest thing available, and it is what the decode will
+ * usually choose, since the reading charge prefers fewer words.
+ */
+export function wordEndingAt(context, at, longest = 4) {
+    for (let length = longest; length > 0; length--) {
+        const from = at - length;
+        if (from < 0) {
+            continue;
+        }
+        const text = context.characters.slice(from, at).join("");
+        if (context.dictionary.lookup(text) !== undefined) {
+            return text;
+        }
+    }
+    return undefined;
+}
+/**
+ * The word starting at a position, longest first, or undefined.
+ */
+export function wordStartingAt(context, at, longest = 4) {
+    for (let length = longest; length > 0; length--) {
+        // Nothing starts at the end of a run, which is where a 得 closing a
+        // sentence asks from.
+        const text = context.characters.slice(at, at + length).join("");
+        if (text !== "" && context.dictionary.lookup(text) !== undefined) {
+            return text;
+        }
+    }
+    return undefined;
+}
+/**
+ * The part of speech the dictionary gives a word, or the empty string.
+ */
+export function tagOf(context, word) {
+    return word === undefined
+        ? ""
+        : (context.dictionary.lookup(word)?.partOfSpeech ?? "");
+}
+/**
+ * Whether every position can still be left, and the end still reached.
+ *
+ * A rule that forbids an edge can strand a stretch of the run — that is the one
+ * way a rule can do real damage, since a lattice with no path through it has no
+ * decode at all. Checked rather than reasoned about, because the rules are meant
+ * to be extensible and somebody else's rule is not this file's to trust.
+ */
+function isTraversable(lattice) {
+    const length = lattice.characters.length;
+    const reaches = Array.from({ length: length + 1 }, () => false);
+    reaches[length] = true;
+    for (let at = length - 1; at >= 0; at--) {
+        /* c8 ignore next 3 -- every index here is a position of the run itself */
+        reaches[at] = (lattice.edges[at] ?? []).some((edge) => reaches[edge.to] ?? false);
+    }
+    /* c8 ignore next -- a run always has a position 0 */
+    return reaches[0] ?? false;
+}
+/**
+ * The verdict the first rule with an opinion gives an edge.
+ */
+function verdictFor(rules, context) {
+    for (const rule of rules) {
+        const verdict = rule.verdictFor(context);
+        if (verdict !== "keep") {
+            return verdict;
+        }
+    }
+    return "keep";
+}
+/**
+ * Forbid every edge a forced one covers the same characters as.
+ *
+ * All the edges leaving a position are given together, so covering the same
+ * characters is a question about where they end. A forced edge never forbids
+ * another forced one: two rules forcing the same span disagree, and the decode
+ * choosing between the two readings is a better answer than neither.
+ */
+function forbidRivals(edges, forced, forbidden) {
+    const settled = new Set(edges.filter((edge) => forced.has(edge)).map((edge) => edge.to));
+    for (const rival of edges) {
+        if (settled.has(rival.to) && !forced.has(rival)) {
+            forbidden.add(rival);
+        }
+    }
+}
+/**
+ * Apply the rules to a lattice, returning the lattice they leave.
+ *
+ * Stage 4 of the pipeline: the rules run over the built lattice and before any
+ * decoding, so what they change is what the decode gets to choose between
+ * rather than what it has already chosen. Nothing is rewritten — an edge is
+ * kept or it is not — which is what keeps a rule from inventing a reading no
+ * source attests.
+ *
+ * The lattice is returned unchanged if the rules would leave the run with no
+ * path through it.
+ */
+export function applyEdgeRules(lattice, dictionary, rules) {
+    if (rules.length === 0) {
+        return lattice;
+    }
+    const forbidden = new Set();
+    const forced = new Set();
+    for (const edges of lattice.edges) {
+        for (const edge of edges) {
+            const verdict = verdictFor(rules, {
+                dictionary,
+                characters: lattice.characters,
+                edge,
+            });
+            if (verdict === "forbid") {
+                forbidden.add(edge);
+            }
+            else if (verdict === "force") {
+                forced.add(edge);
+            }
+        }
+    }
+    // Rivals are forbidden after every verdict is in, so that two rules forcing
+    // the same span leave both readings standing rather than neither.
+    for (const edges of lattice.edges) {
+        forbidRivals(edges, forced, forbidden);
+    }
+    if (forbidden.size === 0) {
+        return lattice;
+    }
+    const ruled = {
+        characters: lattice.characters,
+        edges: lattice.edges.map((edges) => edges.filter((edge) => !forbidden.has(edge))),
+    };
+    return isTraversable(ruled) ? ruled : lattice;
+}
+//# sourceMappingURL=rules.js.map

package/dist/decode/rules.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rules.js","sourceRoot":"","sources":["../../src/decode/rules.ts"],"names":[],"mappings":"AAuCA;;;;;;;GAOG;AACH,MAAM,UAAU,YAAY,CAC1B,OAAoB,EACpB,EAAU,EACV,OAAO,GAAG,CAAC;IAEX,KAAK,IAAI,MAAM,GAAG,OAAO,EAAE,MAAM,GAAG,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC;QAChD,MAAM,IAAI,GAAG,EAAE,GAAG,MAAM,CAAC;QACzB,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC;YACb,SAAS;QACX,CAAC;QACD,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACzD,IAAI,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,SAAS,EAAE,CAAC;YAClD,OAAO,IAAI,CAAC;QACd,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,cAAc,CAC5B,OAAoB,EACpB,EAAU,EACV,OAAO,GAAG,CAAC;IAEX,KAAK,IAAI,MAAM,GAAG,OAAO,EAAE,MAAM,GAAG,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC;QAChD,mEAAmE;QACnE,sBAAsB;QACtB,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,EAAE,EAAE,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAChE,IAAI,IAAI,KAAK,EAAE,IAAI,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,SAAS,EAAE,CAAC;YACjE,OAAO,IAAI,CAAC;QACd,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,KAAK,CAAC,OAAoB,EAAE,IAAwB;IAClE,OAAO,IAAI,KAAK,SAAS;QACvB,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,YAAY,IAAI,EAAE,CAAC,CAAC;AAC5D,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,aAAa,CAAC,OAAgB;IACrC,MAAM,MAAM,GAAG,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC;IACzC,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC;IAChE,OAAO,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC;IACvB,KAAK,IAAI,EAAE,GAAG,MAAM,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC;QACxC,0EAA0E;QAC1E,OAAO,CAAC,EAAE,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAC1C,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,CACpC,CAAC;IACJ,CAAC;IACD,qDAAqD;IACrD,OAAO,OAAO,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC;AAC7B,CAAC;AAED;;GAEG;AACH,SAAS,UAAU,CACjB,KAA0B,EAC1B,OAAoB;IAEpB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;QACzC,IAAI,OAAO,KAAK,MAAM,EAAE,CAAC;YACvB,OAAO,OAAO,CAAC;QACjB,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;;;;GAOG;AACH,SAAS,YAAY,CACnB,KAA6B,EAC7B,MAAgC,EAChC,SAA2B;IAE3B,MAAM,OAAO,GAAG,IAAI,GAAG,CACrB,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,CAChE,CAAC;IACF,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;YAChD,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,cAAc,CAC5B,OAAgB,EAChB,UAAsB,EACtB,KAA0B;IAE1B,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAe,CAAC;IACzC,MAAM,MAAM,GAAG,IAAI,GAAG,EAAe,CAAC;IAEtC,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QAClC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,OAAO,GAAG,UAAU,CAAC,KAAK,EAAE;gBAChC,UAAU;gBACV,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,IAAI;aACL,CAAC,CAAC;YACH,IAAI,OAAO,KAAK,QAAQ,EAAE,CAAC;gBACzB,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YACtB,CAAC;iBAAM,IAAI,OAAO,KAAK,OAAO,EAAE,CAAC;gBAC/B,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YACnB,CAAC;QACH,CAAC;IACH,CAAC;IAED,4EAA4E;IAC5E,kEAAkE;IAClE,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QAClC,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;IACzC,CAAC;IAED,IAAI,SAAS,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,MAAM,KAAK,GAAY;QACrB,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,KAAK,EAAE,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACjC,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAC7C;KACF,CAAC;IACF,OAAO,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC;AAChD,CAAC"}

package/dist/decode/sandhi.d.ts

@@ -1,4 +1,4 @@
-import type { Syllable } from "../syllable/syllable.js";
+import { type Syllable } from "../syllable/syllable.js";
 /**
  * How much of the sandhi to apply.
  */

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

@@ -1 +1 @@
-{"version":3,"file":"sandhi.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AAGxD;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IACxB;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;CAC9B;AA8BD;;;;;;;;;;;GAWG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,SAAS,QAAQ,EAAE,EAC9B,OAAO,GAAE,aAAkB,GAC1B,SAAS,QAAQ,EAAE,CA6BrB"}
+{"version":3,"file":"sandhi.d.ts","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,QAAQ,EAAyB,MAAM,yBAAyB,CAAC;AAG/E;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IACxB;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;CAC9B;AAwFD;;;;;;;;;;;GAWG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,SAAS,QAAQ,EAAE,EAC9B,OAAO,GAAE,aAAkB,GAC1B,SAAS,QAAQ,EAAE,CAiCrB"}

package/dist/decode/sandhi.js

@@ -1,3 +1,4 @@
+import { writeSyllableSpelling } from "../syllable/syllable.js";
 import { NEUTRAL_TONE } from "../tone/tone.js";
 /**
  * Whether a syllable is 一, which is the only thing spelled `yi` that sandhis.
@@ -24,6 +25,60 @@ function yiToneBefore(following) {
     }
     return following === 4 ? 2 : 4;
 }
+/**
+ * The numeral words a 一 can be the last digit of.
+ *
+ * Toneless, because what is being asked is which word it is rather than what
+ * tone it is carrying, and 十 is `shí` in every context it appears in here.
+ *
+ * **亿 is deliberately absent**, and that is measured rather than assumed: no
+ * 一 in 88,866 lines of Tatoeba and zh.wikipedia ends a number in 亿, while 意,
+ * 议, 义 and 議 all read `yì`, all precede a 一 that really is counting, and
+ * would all lose their sandhi for it — 11 conversions against nothing gained.
+ * The other three stay, because a number can end 百一 or 千一 even though this
+ * corpus has none, and the one thing they cost is 擺一個.
+ */
+const NUMERAL_WORDS = new Set(["shi", "bai", "qian", "wan"]);
+/**
+ * The words that carry a number on past a 一, so that it is still counting.
+ *
+ * 一百一十's second 一 counts the ten it is followed by and takes the sandhi;
+ * 二十一's does not, because nothing numeric comes after it.
+ */
+const CONTINUES_NUMBER = new Set([...NUMERAL_WORDS, "ling"]);
+/**
+ * The ordinal prefix 第.
+ */
+const ORDINAL_PREFIX = "di";
+/**
+ * Whether a 一 is counting something, which is the only time it takes sandhi.
+ *
+ * The rule 一 sandhi is usually stated with — fourth tone flattens it, anything
+ * else raises it — is about the 一 that *counts*. Two shapes of 一 are not
+ * counting and keep the citation tone, and both are standard:
+ *
+ * - **The last digit of a larger number.** 十一月 is `shíyīyuè` and 二十一岁 is
+ *   `èrshíyī suì`. The signal is a numeral word before it with nothing numeric
+ *   after, which is what leaves 一百一十's middle 一 alone: 十 follows it, so it
+ *   is counting the ten.
+ * - **An ordinal.** 第一个 is `dìyī gè`.
+ *
+ * **Read off the syllables rather than the characters**, because that is all
+ * this pass has: `pinyinjs sandhi shíyī gè` never sees a hanzi. So it cannot
+ * tell 十 from 时 or 第 from 地, and `docs/sandhi/` carries what that costs —
+ * measured, not estimated.
+ */
+function isCounting(syllables, at) {
+    const spelt = (index) => {
+        const syllable = syllables[index];
+        return syllable === undefined ? "" : writeSyllableSpelling(syllable);
+    };
+    const before = spelt(at - 1);
+    if (before === ORDINAL_PREFIX) {
+        return false;
+    }
+    return !NUMERAL_WORDS.has(before) || CONTINUES_NUMBER.has(spelt(at + 1));
+}
 /**
  * Apply the tone sandhi that standard orthography writes.
  *
@@ -42,7 +97,11 @@ export function applySandhi(syllables, options = {}) {
     for (const [at, syllable] of applied.entries()) {
         const following = applied[at + 1];
         if (yiBu && isYi(syllable) && syllable.tone === 1) {
-            applied[at] = { ...syllable, tone: yiToneBefore(following?.tone) };
+            // A 一 that is not counting keeps its citation tone, so it is left
+            // exactly as the dictionary stored it.
+            if (isCounting(applied, at)) {
+                applied[at] = { ...syllable, tone: yiToneBefore(following?.tone) };
+            }
             continue;
         }
         // 不 flattens to second tone before a fourth, and is otherwise unchanged.

package/dist/decode/sandhi.js.map

@@ -1 +1 @@
-{"version":3,"file":"sandhi.js","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAa,MAAM,iBAAiB,CAAC;AAuB1D;;GAEG;AACH,SAAS,IAAI,CAAC,QAAkB;IAC9B,OAAO,QAAQ,CAAC,OAAO,KAAK,EAAE,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC3D,CAAC;AAED;;GAEG;AACH,SAAS,IAAI,CAAC,QAAkB;IAC9B,OAAO,QAAQ,CAAC,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC5D,CAAC;AAED;;;;;;GAMG;AACH,SAAS,YAAY,CAAC,SAA2B;IAC/C,IAAI,SAAS,KAAK,SAAS,IAAI,SAAS,KAAK,YAAY,EAAE,CAAC;QAC1D,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACjC,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,WAAW,CACzB,SAA8B,EAC9B,OAAO,GAAkB,EAAE;IAE3B,MAAM,EAAE,IAAI,GAAG,IAAI,EAAE,SAAS,GAAG,KAAK,EAAE,GAAG,OAAO,CAAC;IACnD,MAAM,OAAO,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IAE/B,KAAK,MAAM,CAAC,EAAE,EAAE,QAAQ,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QAC/C,MAAM,SAAS,GAAG,OAAO,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QAElC,IAAI,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;YAClD,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,YAAY,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,CAAC;YACnE,SAAS;QACX,CAAC;QACD,yEAAyE;QACzE,IACE,IAAI;YACJ,IAAI,CAAC,QAAQ,CAAC;YACd,QAAQ,CAAC,IAAI,KAAK,CAAC;YACnB,SAAS,EAAE,IAAI,KAAK,CAAC,EACrB,CAAC;YACD,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;YACvC,SAAS;QACX,CAAC;QACD,2EAA2E;QAC3E,0EAA0E;QAC1E,IAAI,SAAS,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,IAAI,SAAS,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,IAAI,KAAK,CAAC,EAAE,CAAC;YACtE,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QACzC,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}
+{"version":3,"file":"sandhi.js","sourceRoot":"","sources":["../../src/decode/sandhi.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAC/E,OAAO,EAAE,YAAY,EAAa,MAAM,iBAAiB,CAAC;AAuB1D;;GAEG;AACH,SAAS,IAAI,CAAC,QAAkB;IAC9B,OAAO,QAAQ,CAAC,OAAO,KAAK,EAAE,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC3D,CAAC;AAED;;GAEG;AACH,SAAS,IAAI,CAAC,QAAkB;IAC9B,OAAO,QAAQ,CAAC,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,CAAC;AAC5D,CAAC;AAED;;;;;;GAMG;AACH,SAAS,YAAY,CAAC,SAA2B;IAC/C,IAAI,SAAS,KAAK,SAAS,IAAI,SAAS,KAAK,YAAY,EAAE,CAAC;QAC1D,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACjC,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC;AAE7D;;;;;GAKG;AACH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;AAE7D;;GAEG;AACH,MAAM,cAAc,GAAG,IAAI,CAAC;AAE5B;;;;;;;;;;;;;;;;;GAiBG;AACH,SAAS,UAAU,CAAC,SAA8B,EAAE,EAAU;IAC5D,MAAM,KAAK,GAAG,CAAC,KAAa,EAAU,EAAE;QACtC,MAAM,QAAQ,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAClC,OAAO,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,qBAAqB,CAAC,QAAQ,CAAC,CAAC;IACvE,CAAC,CAAC;IACF,MAAM,MAAM,GAAG,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;IAC7B,IAAI,MAAM,KAAK,cAAc,EAAE,CAAC;QAC9B,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;AAC3E,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,WAAW,CACzB,SAA8B,EAC9B,OAAO,GAAkB,EAAE;IAE3B,MAAM,EAAE,IAAI,GAAG,IAAI,EAAE,SAAS,GAAG,KAAK,EAAE,GAAG,OAAO,CAAC;IACnD,MAAM,OAAO,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IAE/B,KAAK,MAAM,CAAC,EAAE,EAAE,QAAQ,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QAC/C,MAAM,SAAS,GAAG,OAAO,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QAElC,IAAI,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;YAClD,kEAAkE;YAClE,uCAAuC;YACvC,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC;gBAC5B,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,YAAY,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,CAAC;YACrE,CAAC;YACD,SAAS;QACX,CAAC;QACD,yEAAyE;QACzE,IACE,IAAI;YACJ,IAAI,CAAC,QAAQ,CAAC;YACd,QAAQ,CAAC,IAAI,KAAK,CAAC;YACnB,SAAS,EAAE,IAAI,KAAK,CAAC,EACrB,CAAC;YACD,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;YACvC,SAAS;QACX,CAAC;QACD,2EAA2E;QAC3E,0EAA0E;QAC1E,IAAI,SAAS,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,IAAI,SAAS,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,IAAI,KAAK,CAAC,EAAE,CAAC;YACtE,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QACzC,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/decode/word.d.ts

@@ -13,7 +13,25 @@ export interface DecodedWord {
     readonly partOfSpeech: string;
     /** Whether a dictionary entry was found, rather than a character fallback. */
     readonly isKnown: boolean;
+    /**
+     * What is written in front of this word, where it is not a space.
+     *
+     * Only 分词连写 sets it, and only to a hyphen: the two halves of 干干净净
+     * are written `gāngān-jìngjìng`, one orthographic word with a boundary
+     * inside it rather than two words. Keeping them two decoded words and
+     * recording the mark between them is what lets the halves carry their own
+     * readings and capitals — see {@link WordSeparator}.
+     */
+    readonly separator?: WordSeparator;
 }
+/**
+ * A mark GB/T 16159 writes between two words instead of a space.
+ *
+ * A hyphen is the only one, and it does not mean the same thing as a space: a
+ * hyphenated pair is read back as a single word, which is why
+ * `splitSyllables` treats it as a syllable boundary rather than a word one.
+ */
+export type WordSeparator = "-";
 /**
  * A decoded word together with how settled each of its syllables was.
  */

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

@@ -1 +1 @@
-{"version":3,"file":"word.d.ts","sourceRoot":"","sources":["../../src/decode/word.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AACxD,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AAEzD;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,gDAAgD;IAChD,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,OAAO,EAAE,SAAS,QAAQ,EAAE,CAAC;IACtC,qDAAqD;IACrD,QAAQ,CAAC,YAAY,EAAE,OAAO,CAAC;IAC/B,qDAAqD;IACrD,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,8EAA8E;IAC9E,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;IAC3B;;;;;;OAMG;IACH,QAAQ,CAAC,UAAU,EAAE,SAAS,iBAAiB,EAAE,CAAC;CACnD"}
+{"version":3,"file":"word.d.ts","sourceRoot":"","sources":["../../src/decode/word.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,yBAAyB,CAAC;AACxD,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AAEzD;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,gDAAgD;IAChD,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,OAAO,EAAE,SAAS,QAAQ,EAAE,CAAC;IACtC,qDAAqD;IACrD,QAAQ,CAAC,YAAY,EAAE,OAAO,CAAC;IAC/B,qDAAqD;IACrD,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,8EAA8E;IAC9E,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;IAC1B;;;;;;;;OAQG;IACH,QAAQ,CAAC,SAAS,CAAC,EAAE,aAAa,CAAC;CACpC;AAED;;;;;;GAMG;AACH,MAAM,MAAM,aAAa,GAAG,GAAG,CAAC;AAEhC;;GAEG;AACH,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAC;IAC3B;;;;;;OAMG;IACH,QAAQ,CAAC,UAAU,EAAE,SAAS,iBAAiB,EAAE,CAAC;CACnD"}

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

@@ -1 +1 @@
-{"version":3,"file":"artifact.d.ts","sourceRoot":"","sources":["../../src/dictionary/artifact.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,QAAQ,EAEd,MAAM,yBAAyB,CAAC;AACjC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAyBlD;;;;;;;GAOG;AACH,MAAM,WAAW,kBAAkB;IACjC,8DAA8D;IAC9D,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,sCAAsC;IACtC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,kDAAkD;IAClD,QAAQ,CAAC,WAAW,EAAE,UAAU,CAAC;CAClC;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,SAAS,EAAE,SAAS,QAAQ,EAAE,GAAG,MAAM,CAIpE;AAED;;GAEG;AACH,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,QAAQ,EAAE,GAAG,SAAS,CAQ3E;AAYD;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,eAAe,GAAG,SAAS,MAAM,EAAE,CAE1E;AA+GD;;;;;;;;;;;;GAYG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,SAAS,eAAe,EAAE,GAClC,kBAAkB,CAkCpB;AAED;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,EAAE,kBAAkB,GAC3B,SAAS,eAAe,EAAE,CAyD5B;AAED;;;;;;;;;GASG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,SAAS,eAAe,EAAE,EACnC,QAAQ,EAAE,kBAAkB,GAC3B,MAAM,GAAG,SAAS,CAkBpB"}
+{"version":3,"file":"artifact.d.ts","sourceRoot":"","sources":["../../src/dictionary/artifact.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,QAAQ,EAEd,MAAM,yBAAyB,CAAC;AACjC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAkDlD;;;;;;;GAOG;AACH,MAAM,WAAW,kBAAkB;IACjC,8DAA8D;IAC9D,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,sCAAsC;IACtC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,kDAAkD;IAClD,QAAQ,CAAC,WAAW,EAAE,UAAU,CAAC;CAClC;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,SAAS,EAAE,SAAS,QAAQ,EAAE,GAAG,MAAM,CAIpE;AAED;;GAEG;AACH,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,QAAQ,EAAE,GAAG,SAAS,CAQ3E;AAYD;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,eAAe,GAAG,SAAS,MAAM,EAAE,CAE1E;AA+GD;;;;;;;;;;;;GAYG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,SAAS,eAAe,EAAE,GAClC,kBAAkB,CAqCpB;AAED;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,EAAE,kBAAkB,GAC3B,SAAS,eAAe,EAAE,CA2D5B;AAED;;;;;;;;;GASG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,SAAS,eAAe,EAAE,EACnC,QAAQ,EAAE,kBAAkB,GAC3B,MAAM,GAAG,SAAS,CAkBpB"}

package/dist/dictionary/artifact.js

@@ -19,6 +19,28 @@ const ALTERNATE = ",";
  * The flag marking a proper noun.
  */
 const PROPER_NOUN_FLAG = "p";
+/**
+ * Where a proper name divides, written after the proper-noun flag.
+ *
+ * `p` alone is a proper noun whose parts CC-CEDICT does not divide; `p1` is one
+ * that divides after the first character, `p2.4` after the second and fourth.
+ * The positions are decimal and dot-separated rather than one digit each,
+ * because the longest word carrying any is 19 characters.
+ *
+ * It rides in the flags column rather than taking a column of its own because
+ * it is only ever present on a proper noun, and only 8,205 of 723,139 keys
+ * carry one — a sixth column would cost a separator on every line to say
+ * nothing on almost all of them.
+ */
+const NAME_BOUNDARIES = /^p(?<at>\d+(?:\.\d+)*)$/u;
+const BOUNDARY_SEPARATOR = ".";
+/**
+ * The boundaries a flags column states, empty where it states none.
+ */
+function nameBoundariesIn(flags) {
+    const at = NAME_BOUNDARIES.exec(flags)?.groups?.["at"];
+    return at === undefined ? [] : at.split(BOUNDARY_SEPARATOR).map(Number);
+}
 /**
  * Write a reading in tone-numbered notation.
  *
@@ -176,7 +198,10 @@ export function buildArtifact(entries) {
             isDerivable(key, reading, defaults) ? "" : reading,
             entry.readings.tw === undefined ? "" : encodeReading(entry.readings.tw),
             entry.partOfSpeech,
-            entry.isProperNoun ? PROPER_NOUN_FLAG : "",
+            entry.isProperNoun
+                ? PROPER_NOUN_FLAG +
+                    (entry.nameBoundaries ?? []).join(BOUNDARY_SEPARATOR)
+                : "",
             (entry.alternates ?? [])
                 .map((alternate) => encodeReading(alternate))
                 .join(ALTERNATE),
@@ -215,6 +240,7 @@ export function readArtifact(artifact) {
     for (let at = 0; at < index.size; at++) {
         const key = index.keyAt(at);
         const [reading = "", taiwan = "", partOfSpeech = "", flags = "", alternates = "",] = (lines[at] ?? "").split(COLUMN);
+        const nameBoundaries = nameBoundariesIn(flags);
         const stored = reading === ""
             ? toCharacters(key)
                 .map((character) => defaults.get(character) ?? "")
@@ -234,6 +260,7 @@ export function readArtifact(artifact) {
             frequency: frequencies.bucketOf(at),
             partOfSpeech,
             isProperNoun: flags.includes(PROPER_NOUN_FLAG),
+            ...(nameBoundaries.length > 0 && { nameBoundaries }),
             ...(alternates !== "" && {
                 alternates: alternates
                     .split(ALTERNATE)