@kensio/pinyinjs 1.7.5 → 1.7.7

package/dist/decode/convert-pipeline.js

@@ -4,8 +4,10 @@ import { readNumbersIn } from "../numerals/text.js";
 import { toCharacters } from "../script/characters.js";
 import { hintsWithin, resolveHints } from "./hints.js";
 import { splitRuns, surroundingCharacters } from "./runs.js";
-import { numeralBefore, surrounding, writeNumbers } from "./numbers.js";
-import { writeRun, rewrite, wordsOf } from "./write-run.js";
+import { surrounding, writeNumbers } from "./numbers.js";
+import { contextAround } from "./run-context.js";
+import { wordsOf } from "./run-words.js";
+import { writeRun, rewrite } from "./write-run.js";
 /**
  * Run the pipeline over a text with a given decoder.
  */
@@ -41,8 +43,7 @@ export function convertWith(decode, dictionary, text, options) {
         const within = hints === undefined
             ? undefined
             : hintsWithin(hints, starts[at] ?? 0, toCharacters(run.text).length);
-        const said = numeralBefore(read[at - 1] ?? []);
-        return wordsOf(decode(dictionary, run.text, said, within), dictionary, grouping);
+        return wordsOf(decode(dictionary, run.text, contextAround(runs, read, at), within), dictionary, grouping);
     });
     for (const [at, run] of runs.entries()) {
         const words = decoded[at] ?? [];

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

@@ -1 +1 @@
-{"version":3,"file":"convert-pipeline.js","sourceRoot":"","sources":["../../src/decode/convert-pipeline.ts"],"names":[],"mappings":"AAWA,OAAO,EACL,uBAAuB,EACvB,UAAU,GACX,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AACvD,OAAO,EAAE,SAAS,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC;AAE7D,OAAO,EAAE,aAAa,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAExE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AAE5D;;GAEG;AACH,MAAM,UAAU,WAAW,CACzB,MAAc,EACd,UAAsB,EACtB,IAAY,EACZ,OAAuB;IAEvB,MAAM,EACJ,MAAM,GAAG,OAAO,EAChB,QAAQ,GAAG,OAAO,EAClB,UAAU,GAAG,QAAQ,EACrB,QAAQ,GAAG,MAAM,EACjB,WAAW,GAAG,OAAO,EACrB,QAAQ,GAAG,IAAI,EACf,OAAO,GAAG,MAAM,EAChB,MAAM,EACN,QAAQ,GACT,GAAG,OAAO,CAAC;IACZ,MAAM,OAAO,GAAY,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;IAC5D,MAAM,SAAS,GAAqB,EAAE,CAAC;IAEvC,MAAM,IAAI,GAAG,CAAC,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;IAClC,0EAA0E;IAC1E,6DAA6D;IAC7D,MAAM,KAAK,GAAG,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC1E,6EAA6E;IAC7E,4DAA4D;IAC5D,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnB,KAAK,IAAI,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;IACzC,CAAC;IACD,uEAAuE;IACvE,0EAA0E;IAC1E,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE,CAChC,GAAG,CAAC,KAAK,IAAI,OAAO,KAAK,MAAM;QAC7B,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,IAAI,EAAE,qBAAqB,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAC7D,CAAC;IACF,0EAA0E;IAC1E,4EAA4E;IAC5E,+DAA+D;IAC/D,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE;QACnC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;YACf,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,MAAM,MAAM,GACV,KAAK,KAAK,SAAS;YACjB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC;QACzE,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QAC/C,OAAO,OAAO,CACZ,MAAM,CAAC,UAAU,EAAE,GAAG,CAAC,IAAI,EAAE,IAAI,EAAE,MAAM,CAAC,EAC1C,UAAU,EACV,QAAQ,CACT,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,KAAK,MAAM,CAAC,EAAE,EAAE,GAAG,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;QAChC,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;YACd,SAAS,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC;YACxE,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CACZ,GAAG,YAAY,CACb,GAAG,CAAC,IAAI,EACR,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,EACd,WAAW,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,EAC9B,OAAO,EACP,EAAE,MAAM,EAAE,CACX,CACF,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAA8B,SAAS,CAAC;IAElD,0EAA0E;IAC1E,6EAA6E;IAC7E,qEAAqE;IACrE,IAAI,QAAQ,KAAK,MAAM,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QAC5C,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC,CAAC;IACpD,CAAC;IACD,OAAO,WAAW,KAAK,OAAO;QAC5B,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC;QAC1C,CAAC,CAAC,MAAM,CAAC;AACb,CAAC"}
+{"version":3,"file":"convert-pipeline.js","sourceRoot":"","sources":["../../src/decode/convert-pipeline.ts"],"names":[],"mappings":"AAWA,OAAO,EACL,uBAAuB,EACvB,UAAU,GACX,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AACvD,OAAO,EAAE,SAAS,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAC;AAE7D,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AACzD,OAAO,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEjD,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AAEnD;;GAEG;AACH,MAAM,UAAU,WAAW,CACzB,MAAc,EACd,UAAsB,EACtB,IAAY,EACZ,OAAuB;IAEvB,MAAM,EACJ,MAAM,GAAG,OAAO,EAChB,QAAQ,GAAG,OAAO,EAClB,UAAU,GAAG,QAAQ,EACrB,QAAQ,GAAG,MAAM,EACjB,WAAW,GAAG,OAAO,EACrB,QAAQ,GAAG,IAAI,EACf,OAAO,GAAG,MAAM,EAChB,MAAM,EACN,QAAQ,GACT,GAAG,OAAO,CAAC;IACZ,MAAM,OAAO,GAAY,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;IAC5D,MAAM,SAAS,GAAqB,EAAE,CAAC;IAEvC,MAAM,IAAI,GAAG,CAAC,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;IAClC,0EAA0E;IAC1E,6DAA6D;IAC7D,MAAM,KAAK,GAAG,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC1E,6EAA6E;IAC7E,4DAA4D;IAC5D,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnB,KAAK,IAAI,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;IACzC,CAAC;IACD,uEAAuE;IACvE,0EAA0E;IAC1E,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE,CAChC,GAAG,CAAC,KAAK,IAAI,OAAO,KAAK,MAAM;QAC7B,CAAC,CAAC,EAAE;QACJ,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,IAAI,EAAE,qBAAqB,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAC7D,CAAC;IACF,0EAA0E;IAC1E,4EAA4E;IAC5E,+DAA+D;IAC/D,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,EAAE;QACnC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;YACf,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,MAAM,MAAM,GACV,KAAK,KAAK,SAAS;YACjB,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC;QACzE,OAAO,OAAO,CACZ,MAAM,CAAC,UAAU,EAAE,GAAG,CAAC,IAAI,EAAE,aAAa,CAAC,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC,EAAE,MAAM,CAAC,EACnE,UAAU,EACV,QAAQ,CACT,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,KAAK,MAAM,CAAC,EAAE,EAAE,GAAG,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;QAChC,IAAI,GAAG,CAAC,KAAK,EAAE,CAAC;YACd,SAAS,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC;YACxE,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CACZ,GAAG,YAAY,CACb,GAAG,CAAC,IAAI,EACR,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,EACd,WAAW,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,EAC9B,OAAO,EACP,EAAE,MAAM,EAAE,CACX,CACF,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAA8B,SAAS,CAAC;IAElD,0EAA0E;IAC1E,6EAA6E;IAC7E,qEAAqE;IACrE,IAAI,QAAQ,KAAK,MAAM,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QAC5C,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC,CAAC;IACpD,CAAC;IACD,OAAO,WAAW,KAAK,OAAO;QAC5B,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,uBAAuB,CAAC;QAC1C,CAAC,CAAC,MAAM,CAAC;AACb,CAAC"}

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

@@ -0,0 +1,74 @@
+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;
+/**
+ * A verb and a 得 in front of a complement is not a word reading `dé`.
+ *
+ * 他算得上一个作家 came out `suàn dé shàng`, because 算得 is a key jieba counted
+ * and the phrase corpus reads it `suàn dé`. In a potential complement the 得 is
+ * the particle: 算得上, 吃得了, 买得起 and 看得住 are all `de`, and the reading
+ * the key carries is the one 算得 has standing alone, which is not what a
+ * complement leaves it doing.
+ *
+ * Both sides are needed and neither is enough. A verb before the 得 rules out
+ * 只得上山 and 不得下车, where the 得 belongs to an adverb and the character
+ * after it starts a verb phrase of its own; the complement after it rules out
+ * 取得上级批准, where 上 is the first character of 上级 rather than a complement.
+ * That second test is why the complement has to be the whole of the word
+ * starting there.
+ *
+ * Over the 88,866 lines it changes 5 conversions and all 5 are 算得上, which is
+ * the only word of this shape the corpus holds. CPP does not move.
+ */
+export declare const POTENTIAL_DE: EdgeRule;
+/**
+ * A word beginning with 的 does not start where the particle 的 does.
+ *
+ * The structural particle attaches to the modifier in front of it and the head
+ * follows, so a dictionary key spanning that 的 and the head's first character
+ * is describing some other sentence: 没有人知道他的真名字 came out
+ * `tā dí zhēn míngzi` on the strength of 的真, 图书馆的卡 on 的卡, and 你說的對
+ * came out as one word, `deduì`.
+ *
+ * Only a key **no source has tagged** is forbidden, which is the same line
+ * {@link TEACHING_JIAO} draws for a word ending in 教. 的确, 的士 and 的哥 are
+ * words jieba counted and tagged, and every one of them can genuinely begin
+ * where this fires: 我的确知道, 我要一辆的士, 那个的哥. That leaves 我的哥哥们
+ * reading `wǒ dī gēgē men`, since 的哥 is a taxi driver and jieba has counted
+ * them. What this does reach is the untagged tail — a reading somebody asserted
+ * rather than a word anybody counted — where the 的 is the particle every time.
+ *
+ * Measured over the 88,866 lines of Tatoeba and zh.wikipedia it forbids edges
+ * in 40 runs: 的筆 13, 的這 11, 的真 6, 的對 4, 的卡 4 and 的歷 once. Every one
+ * is a correction, and half of them are corrections to the spacing as much as
+ * to the reading — 你說的對 was one word `deduì` and is now two.
+ */
+export declare const PARTICLE_DE: EdgeRule;
+//# sourceMappingURL=de-rule.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"de-rule.d.ts","sourceRoot":"","sources":["../../src/decode/de-rule.ts"],"names":[],"mappings":"AAUA,OAAO,EAEL,KAAK,QAAQ,EAKd,MAAM,YAAY,CAAC;AAgCpB;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,eAAO,MAAM,QAAQ,EAAE,QActB,CAAC;AAkDF;;;;;;;;;;;;;;;;;;GAkBG;AACH,eAAO,MAAM,YAAY,EAAE,QAoB1B,CAAC;AAWF;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,eAAO,MAAM,WAAW,EAAE,QAczB,CAAC"}

package/dist/decode/de-rule.js

@@ -0,0 +1,199 @@
+/**
+ * The two rules about 得 and 的, which are one character each and three jobs.
+ *
+ * 得 is stored as the structural particle with `dé` and `děi` as alternates,
+ * and 的 is stored as the particle full stop. Both are among the commonest
+ * characters in the language, so the words they begin and end are numerous and
+ * mostly wrong.
+ */
+import { toCharacters } from "../script/characters.js";
+import { tagOf, wordEndingAt, wordStartingAt, wordsStartingAt, } 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";
+    },
+};
+/**
+ * What a potential complement is made of, after the 得 that marks it.
+ *
+ * 算得上, 买得起, 看得住, 说得过去. The complement is a resultative or a
+ * directional, and this is the closed set of the ones that take a 得 in front of
+ * them rather than a verb of their own.
+ *
+ * 了 and 过 are deliberately absent, and for the same reason: both are aspect
+ * markers as well as complements, and the aspect is the commoner by a wide
+ * margin. Over 88,866 lines of Tatoeba and zh.wikipedia, including 了 broke 70
+ * conversions and fixed none — 获得了, 取得了 and 赢得了 all became 获 得 了 —
+ * and 过 broke 赢得过 twice. 吃得了 and 说得过去 are the shapes that gives up.
+ */
+const COMPLEMENTS = new Set(["上", "起", "下", "住"]);
+/**
+ * The tag prefix jieba gives a verb, which is what a complement attaches to.
+ */
+const VERB_TAG = "v";
+/**
+ * Whether a syllable is `dé`, the reading 得 takes as a verb of its own.
+ */
+function isDe(syllable) {
+    return (syllable?.initial === "d" && syllable.final === "e" && syllable.tone === 2);
+}
+/**
+ * Whether a word starting at the complement takes it away from the 得.
+ *
+ * The one thing that separates 算得上一个作家 from 取得上级批准, since both put a
+ * verb, a 得 and a 上 in the same order. 上级 is a word and nothing follows it
+ * that starts another, so the 上 belongs to it; 上道 is also a word, but 道歉
+ * follows and is the better claim on the 道, so the 上 is left to the 得.
+ *
+ * Only multi-character words are asked, because every character is a word of its
+ * own and a test that counted those would answer yes everywhere.
+ */
+function isSwallowed(context, at) {
+    const longer = (from) => wordsStartingAt(context, from).some((word) => toCharacters(word).length > 1);
+    return longer(at) && !longer(at + 1);
+}
+/**
+ * A verb and a 得 in front of a complement is not a word reading `dé`.
+ *
+ * 他算得上一个作家 came out `suàn dé shàng`, because 算得 is a key jieba counted
+ * and the phrase corpus reads it `suàn dé`. In a potential complement the 得 is
+ * the particle: 算得上, 吃得了, 买得起 and 看得住 are all `de`, and the reading
+ * the key carries is the one 算得 has standing alone, which is not what a
+ * complement leaves it doing.
+ *
+ * Both sides are needed and neither is enough. A verb before the 得 rules out
+ * 只得上山 and 不得下车, where the 得 belongs to an adverb and the character
+ * after it starts a verb phrase of its own; the complement after it rules out
+ * 取得上级批准, where 上 is the first character of 上级 rather than a complement.
+ * That second test is why the complement has to be the whole of the word
+ * starting there.
+ *
+ * Over the 88,866 lines it changes 5 conversions and all 5 are 算得上, which is
+ * the only word of this shape the corpus holds. CPP does not move.
+ */
+export const POTENTIAL_DE = {
+    name: "potential-de",
+    verdictFor: (context) => {
+        const { characters, edge } = context;
+        if (edge.to - edge.from !== 2 ||
+            edge.text.at(-1) !== "得" ||
+            edge.reading.length !== 2 ||
+            !isDe(edge.reading[1])) {
+            return "keep";
+        }
+        if (!tagOf(context, characters[edge.from]).startsWith(VERB_TAG)) {
+            return "keep";
+        }
+        return COMPLEMENTS.has(characters[edge.to] ?? "") &&
+            !isSwallowed(context, edge.to)
+            ? "forbid"
+            : "keep";
+    },
+};
+/**
+ * The tags a word carries when the 的 after it is the structural particle.
+ *
+ * A pronoun, a noun, an adjective, a verb or a distinguishing word: 他的,
+ * 图书馆的, 伟大的, 见过面的, 男的. What the particle marks is a modifier, and
+ * these are what a modifier is.
+ */
+const MODIFIED_TAGS = /^(?:r|n|a|v|b|z)/u;
+/**
+ * A word beginning with 的 does not start where the particle 的 does.
+ *
+ * The structural particle attaches to the modifier in front of it and the head
+ * follows, so a dictionary key spanning that 的 and the head's first character
+ * is describing some other sentence: 没有人知道他的真名字 came out
+ * `tā dí zhēn míngzi` on the strength of 的真, 图书馆的卡 on 的卡, and 你說的對
+ * came out as one word, `deduì`.
+ *
+ * Only a key **no source has tagged** is forbidden, which is the same line
+ * {@link TEACHING_JIAO} draws for a word ending in 教. 的确, 的士 and 的哥 are
+ * words jieba counted and tagged, and every one of them can genuinely begin
+ * where this fires: 我的确知道, 我要一辆的士, 那个的哥. That leaves 我的哥哥们
+ * reading `wǒ dī gēgē men`, since 的哥 is a taxi driver and jieba has counted
+ * them. What this does reach is the untagged tail — a reading somebody asserted
+ * rather than a word anybody counted — where the 的 is the particle every time.
+ *
+ * Measured over the 88,866 lines of Tatoeba and zh.wikipedia it forbids edges
+ * in 40 runs: 的筆 13, 的這 11, 的真 6, 的對 4, 的卡 4 and 的歷 once. Every one
+ * is a correction, and half of them are corrections to the spacing as much as
+ * to the reading — 你說的對 was one word `deduì` and is now two.
+ */
+export const PARTICLE_DE = {
+    name: "particle-de",
+    verdictFor: (context) => {
+        const { edge } = context;
+        if (edge.to - edge.from < 2 ||
+            edge.partOfSpeech !== "" ||
+            !edge.text.startsWith("的")) {
+            return "keep";
+        }
+        const before = wordEndingAt(context, edge.from);
+        return MODIFIED_TAGS.test(tagOf(context, before)) ? "forbid" : "keep";
+    },
+};
+//# sourceMappingURL=de-rule.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"de-rule.js","sourceRoot":"","sources":["../../src/decode/de-rule.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAEvD,OAAO,EAGL,KAAK,EACL,YAAY,EACZ,cAAc,EACd,eAAe,GAChB,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;;;;;;;;;;;;GAYG;AACH,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAElD;;GAEG;AACH,MAAM,QAAQ,GAAG,GAAG,CAAC;AAErB;;GAEG;AACH,SAAS,IAAI,CAAC,QAA8B;IAC1C,OAAO,CACL,QAAQ,EAAE,OAAO,KAAK,GAAG,IAAI,QAAQ,CAAC,KAAK,KAAK,GAAG,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,CAC3E,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,WAAW,CAAC,OAAoB,EAAE,EAAU;IACnD,MAAM,MAAM,GAAG,CAAC,IAAY,EAAW,EAAE,CACvC,eAAe,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,IAAI,CACjC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,CACxC,CAAC;IACJ,OAAO,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;AACvC,CAAC;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,CAAC,MAAM,YAAY,GAAa;IACpC,IAAI,EAAE,cAAc;IACpB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACrC,IACE,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,KAAK,CAAC;YACzB,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG;YACxB,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC;YACzB,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EACtB,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;YAChE,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,OAAO,WAAW,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;YAC/C,CAAC,WAAW,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC;YAC9B,CAAC,CAAC,QAAQ;YACV,CAAC,CAAC,MAAM,CAAC;IACb,CAAC;CACF,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,aAAa,GAAG,mBAAmB,CAAC;AAE1C;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,CAAC,MAAM,WAAW,GAAa;IACnC,IAAI,EAAE,aAAa;IACnB,UAAU,EAAE,CAAC,OAAoB,EAAE,EAAE;QACnC,MAAM,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC;QACzB,IACE,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,GAAG,CAAC;YACvB,IAAI,CAAC,YAAY,KAAK,EAAE;YACxB,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAC1B,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QACD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAChD,OAAO,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC;IACxE,CAAC;CACF,CAAC"}

package/dist/decode/decode.d.ts

@@ -1,28 +1,15 @@
+/**
+ * Decoding a Han run into words: the two entry points a caller reaches for.
+ *
+ * The sweeps over the lattice itself are in `lattice-decode.ts`, and are
+ * re-exported here so that the decoder's parts stay one import away.
+ */
 import type { Dictionary } from "../dictionary/dictionary.js";
 import type { ResolvedHints } from "./hints.js";
-import { type Lattice } from "./lattice.js";
-import { type ReadingProjection, type ReadingUnit } from "./locking.js";
 import type { EdgeRule } from "./rules.js";
 import type { DecodedWord, ScoredWord } from "./word.js";
-/**
- * Decode the run's readings, scoring only where the readings are still open.
- *
- * The asymmetry ALGORITHM.md calls for. A stretch between two cut points whose
- * positions all locked reads the same way on every path, so its readings are
- * read off the locks and no shortest path is run over it at all. Only the
- * stretches with a genuine choice in them are decoded.
- */
-export declare function decodeReadings(lattice: Lattice, projection: ReadingProjection): readonly ReadingUnit[];
-/**
- * Where a decode would put the word boundaries, as character positions.
- *
- * Run over the whole lattice rather than the open stretches, because spacing
- * stays ambiguous where readings do not: 研究生命起源 reads identically whether
- * it splits 研究生/命 or 研究/生命, so every position in it locks, and the
- * spacing question is still open.
- */
-export declare function decodeSpacing(lattice: Lattice): readonly number[];
-export declare function decodeRun(dictionary: Dictionary, run: string, rules?: readonly EdgeRule[], before?: string, hints?: ResolvedHints): readonly DecodedWord[];
+export { decodeReadings, decodeSpacing } from "./lattice-decode.js";
+export declare function decodeRun(dictionary: Dictionary, run: string, rules?: readonly EdgeRule[], before?: string, hints?: ResolvedHints, after?: string): readonly DecodedWord[];
 /**
  * Decode a Han run, keeping what each reading was chosen over.
  *
@@ -31,9 +18,9 @@ export declare function decodeRun(dictionary: Dictionary, run: string, rules?: r
  * only worth running for a caller that will use the answer — rendering
  * uncertain readings differently, or reporting them.
  *
- * `before` is the context {@link decodeRun} takes, and means the same here: it
- * is decoded with the run and reported with neither words nor confidence of its
- * own.
+ * `before` and `after` are the context {@link decodeRun} takes, and mean the
+ * same here: they are decoded with the run and reported with neither words nor
+ * confidence of their own.
  */
-export declare function decodeRunScored(dictionary: Dictionary, run: string, rules?: readonly EdgeRule[], before?: string, hints?: ResolvedHints): readonly ScoredWord[];
+export declare function decodeRunScored(dictionary: Dictionary, run: string, rules?: readonly EdgeRule[], before?: string, hints?: ResolvedHints, after?: string): readonly ScoredWord[];
 //# sourceMappingURL=decode.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"decode.d.ts","sourceRoot":"","sources":["../../src/decode/decode.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAE9D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,EAAa,KAAK,OAAO,EAAE,MAAM,cAAc,CAAC;AACvD,OAAO,EAGL,KAAK,iBAAiB,EACtB,KAAK,WAAW,EAGjB,MAAM,cAAc,CAAC;AAEtB,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAE3C,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAEzD;;;;;;;GAOG;AACH,wBAAgB,cAAc,CAC5B,OAAO,EAAE,OAAO,EAChB,UAAU,EAAE,iBAAiB,GAC5B,SAAS,WAAW,EAAE,CAmBxB;AAED;;;;;;;GAOG;AACH,wBAAgB,aAAa,CAAC,OAAO,EAAE,OAAO,GAAG,SAAS,MAAM,EAAE,CAIjE;AAGD,wBAAgB,SAAS,CACvB,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,KAAK,GAAE,SAAS,QAAQ,EAAkB,EAC1C,MAAM,SAAK,EACX,KAAK,CAAC,EAAE,aAAa,GACpB,SAAS,WAAW,EAAE,CAOxB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAC7B,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,KAAK,GAAE,SAAS,QAAQ,EAAkB,EAC1C,MAAM,SAAK,EACX,KAAK,CAAC,EAAE,aAAa,GACpB,SAAS,UAAU,EAAE,CAcvB"}
+{"version":3,"file":"decode.d.ts","sourceRoot":"","sources":["../../src/decode/decode.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAE9D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAIhD,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAEzD,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAGpE,wBAAgB,SAAS,CACvB,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,KAAK,GAAE,SAAS,QAAQ,EAAkB,EAC1C,MAAM,SAAK,EACX,KAAK,CAAC,EAAE,aAAa,EACrB,KAAK,SAAK,GACT,SAAS,WAAW,EAAE,CAOxB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAC7B,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,KAAK,GAAE,SAAS,QAAQ,EAAkB,EAC1C,MAAM,SAAK,EACX,KAAK,CAAC,EAAE,aAAa,EACrB,KAAK,SAAK,GACT,SAAS,UAAU,EAAE,CAcvB"}

package/dist/decode/decode.js

@@ -1,48 +1,11 @@
 import { scoreReadings } from "./confidence.js";
-import { cutPoints } from "./lattice.js";
-import { isSettled, projectReadings, settledUnits, unitsOf, } from "./locking.js";
+import { decodeReadings } from "./lattice-decode.js";
+import { projectReadings } from "./locking.js";
 import { READING_RULES } from "./reading-rules.js";
-import { readingCost, shortestPath, spacingCost } from "./viterbi.js";
-/**
- * Decode the run's readings, scoring only where the readings are still open.
- *
- * The asymmetry ALGORITHM.md calls for. A stretch between two cut points whose
- * positions all locked reads the same way on every path, so its readings are
- * read off the locks and no shortest path is run over it at all. Only the
- * stretches with a genuine choice in them are decoded.
- */
-export function decodeReadings(lattice, projection) {
-    const cuts = cutPoints(lattice);
-    const units = [];
-    for (const [index, from] of cuts.entries()) {
-        const to = cuts[index + 1];
-        if (to === undefined) {
-            break;
-        }
-        if (isSettled(projection, from, to)) {
-            units.push(...settledUnits(projection, from, to));
-            continue;
-        }
-        for (const edge of shortestPath(lattice, from, to, readingCost)) {
-            units.push(...unitsOf(edge));
-        }
-    }
-    return units;
-}
-/**
- * Where a decode would put the word boundaries, as character positions.
- *
- * Run over the whole lattice rather than the open stretches, because spacing
- * stays ambiguous where readings do not: 研究生命起源 reads identically whether
- * it splits 研究生/命 or 研究/生命, so every position in it locks, and the
- * spacing question is still open.
- */
-export function decodeSpacing(lattice) {
-    return shortestPath(lattice, 0, lattice.characters.length, spacingCost).map((edge) => edge.from);
-}
+export { decodeReadings, decodeSpacing } from "./lattice-decode.js";
 import { runGroups, runLattice, wordFrom } from "./run-lattice.js";
-export function decodeRun(dictionary, run, rules = READING_RULES, before = "", hints) {
-    const held = runLattice(dictionary, run, rules, before, hints);
+export function decodeRun(dictionary, run, rules = READING_RULES, before = "", hints, after = "") {
+    const held = runLattice(dictionary, run, rules, { before, after }, hints);
     const { lattice } = held;
     const units = decodeReadings(lattice, projectReadings(lattice));
     return runGroups(held, units).map((group) => wordFrom(dictionary, lattice, group));
@@ -55,12 +18,12 @@ export function decodeRun(dictionary, run, rules = READING_RULES, before = "", h
  * only worth running for a caller that will use the answer — rendering
  * uncertain readings differently, or reporting them.
  *
- * `before` is the context {@link decodeRun} takes, and means the same here: it
- * is decoded with the run and reported with neither words nor confidence of its
- * own.
+ * `before` and `after` are the context {@link decodeRun} takes, and mean the
+ * same here: they are decoded with the run and reported with neither words nor
+ * confidence of their own.
  */
-export function decodeRunScored(dictionary, run, rules = READING_RULES, before = "", hints) {
-    const held = runLattice(dictionary, run, rules, before, hints);
+export function decodeRunScored(dictionary, run, rules = READING_RULES, before = "", hints, after = "") {
+    const held = runLattice(dictionary, run, rules, { before, after }, hints);
     const { lattice } = held;
     const units = scoreReadings(lattice, projectReadings(lattice));
     return runGroups(held, units).map((group) => ({

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;AAEjE,OAAO,EAAE,SAAS,EAAgB,MAAM,cAAc,CAAC;AACvD,OAAO,EACL,SAAS,EACT,eAAe,EAGf,YAAY,EACZ,OAAO,GACR,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAEnD,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,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AACnE,MAAM,UAAU,SAAS,CACvB,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa,EAC1C,MAAM,GAAG,EAAE,EACX,KAAqB;IAErB,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC;IAC/D,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IACzB,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAChE,OAAO,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAC1C,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC,CACrC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,eAAe,CAC7B,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa,EAC1C,MAAM,GAAG,EAAE,EACX,KAAqB;IAErB,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC;IAC/D,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IACzB,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAE/D,OAAO,SAAS,CAAa,IAAI,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QACxD,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":"AAOA,OAAO,EAAE,aAAa,EAAmB,MAAM,iBAAiB,CAAC;AAEjE,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAC/C,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAInD,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAEpE,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AACnE,MAAM,UAAU,SAAS,CACvB,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa,EAC1C,MAAM,GAAG,EAAE,EACX,KAAqB,EACrB,KAAK,GAAG,EAAE;IAEV,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,KAAK,CAAC,CAAC;IAC1E,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IACzB,MAAM,KAAK,GAAG,cAAc,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAChE,OAAO,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAC1C,QAAQ,CAAC,UAAU,EAAE,OAAO,EAAE,KAAK,CAAC,CACrC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,eAAe,CAC7B,UAAsB,EACtB,GAAW,EACX,KAAK,GAAwB,aAAa,EAC1C,MAAM,GAAG,EAAE,EACX,KAAqB,EACrB,KAAK,GAAG,EAAE;IAEV,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,KAAK,CAAC,CAAC;IAC1E,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IACzB,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,EAAE,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IAE/D,OAAO,SAAS,CAAa,IAAI,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QACxD,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/decoders.d.ts

@@ -7,12 +7,17 @@
  */
 import type { Dictionary } from "../dictionary/dictionary.js";
 import type { ResolvedHints } from "./hints.js";
+import type { DecodeContext } from "./lattice-slice.js";
 import type { ScoredWord } from "./word.js";
 /**
  * How a Han run is turned into words, with whatever the decoder can say about
  * how settled they were.
+ *
+ * The context is one argument here where the exported decoders keep `before`
+ * and `after` apart, since nothing outside this package calls it and the two
+ * are one claim about the text around the run.
  */
-export type Decode = (dictionary: Dictionary, run: string, before: string, hints: ResolvedHints | undefined) => readonly ScoredWord[];
+export type Decode = (dictionary: Dictionary, run: string, context: DecodeContext, hints: ResolvedHints | undefined) => readonly ScoredWord[];
 /**
  * The lattice decoder, reporting nothing about its own confidence.
  */
@@ -24,9 +29,9 @@ export declare const SCORED: Decode;
 /**
  * The greedy baseline, which has nothing to report either way.
  *
- * The context in front of a run goes unused: longest-match has no way to weigh
- * one segmentation against another, which is the whole reason it is the
- * baseline, and a 汉字 it cannot report would only be another thing to trim.
+ * The context around a run goes unused: longest-match has no way to weigh one
+ * segmentation against another, which is the whole reason it is the baseline,
+ * and 汉字 it cannot report would only be another thing to trim.
  */
 export declare const GREEDY: Decode;
 //# sourceMappingURL=decoders.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"decoders.d.ts","sourceRoot":"","sources":["../../src/decode/decoders.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAG9D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAEhD,OAAO,KAAK,EAAe,UAAU,EAAE,MAAM,WAAW,CAAC;AAEzD;;;GAGG;AACH,MAAM,MAAM,MAAM,GAAG,CACnB,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,aAAa,GAAG,SAAS,KAC7B,SAAS,UAAU,EAAE,CAAC;AAoB3B;;GAEG;AACH,eAAO,MAAM,OAAO,EAAE,MAErB,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,MAAM,EAAE,MAC2C,CAAC;AAEjE;;;;;;GAMG;AACH,eAAO,MAAM,MAAM,EAAE,MAAiC,CAAC"}
+{"version":3,"file":"decoders.d.ts","sourceRoot":"","sources":["../../src/decode/decoders.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AAG9D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAExD,OAAO,KAAK,EAAe,UAAU,EAAE,MAAM,WAAW,CAAC;AAEzD;;;;;;;GAOG;AACH,MAAM,MAAM,MAAM,GAAG,CACnB,UAAU,EAAE,UAAU,EACtB,GAAG,EAAE,MAAM,EACX,OAAO,EAAE,aAAa,EACtB,KAAK,EAAE,aAAa,GAAG,SAAS,KAC7B,SAAS,UAAU,EAAE,CAAC;AAoB3B;;GAEG;AACH,eAAO,MAAM,OAAO,EAAE,MASrB,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,MAAM,EAAE,MAQlB,CAAC;AAEJ;;;;;;GAMG;AACH,eAAO,MAAM,MAAM,EAAE,MAEpB,CAAC"}

package/dist/decode/decoders.js

@@ -5,7 +5,7 @@ import { READING_RULES } from "./reading-rules.js";
  * A decoder that reports no confidence at all, which is every decoder but one.
  */
 function unscored(decode) {
-    return (dictionary, run, before, hints) => decode(dictionary, run, before, hints).map((word) => ({
+    return (dictionary, run, context, hints) => decode(dictionary, run, context, hints).map((word) => ({
         word,
         confidence: [],
     }));
@@ -13,17 +13,17 @@ function unscored(decode) {
 /**
  * The lattice decoder, reporting nothing about its own confidence.
  */
-export const LATTICE = unscored((dictionary, run, before, hints) => decodeRun(dictionary, run, READING_RULES, before, hints));
+export const LATTICE = unscored((dictionary, run, context, hints) => decodeRun(dictionary, run, READING_RULES, context.before, hints, context.after));
 /**
  * The lattice decoder, with what each reading was chosen over.
  */
-export const SCORED = (dictionary, run, before, hints) => decodeRunScored(dictionary, run, READING_RULES, before, hints);
+export const SCORED = (dictionary, run, context, hints) => decodeRunScored(dictionary, run, READING_RULES, context.before, hints, context.after);
 /**
  * The greedy baseline, which has nothing to report either way.
  *
- * The context in front of a run goes unused: longest-match has no way to weigh
- * one segmentation against another, which is the whole reason it is the
- * baseline, and a 汉字 it cannot report would only be another thing to trim.
+ * The context around a run goes unused: longest-match has no way to weigh one
+ * segmentation against another, which is the whole reason it is the baseline,
+ * and 汉字 it cannot report would only be another thing to trim.
  */
-export const GREEDY = unscored(decodeGreedily);
+export const GREEDY = unscored((dictionary, run) => decodeGreedily(dictionary, run));
 //# sourceMappingURL=decoders.js.map

package/dist/decode/decoders.js.map

@@ -1 +1 @@
-{"version":3,"file":"decoders.js","sourceRoot":"","sources":["../../src/decode/decoders.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE7C,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAcnD;;GAEG;AACH,SAAS,QAAQ,CACf,MAK2B;IAE3B,OAAO,CAAC,UAAU,EAAE,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,CACxC,MAAM,CAAC,UAAU,EAAE,GAAG,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACpD,IAAI;QACJ,UAAU,EAAE,EAAE;KACf,CAAC,CAAC,CAAC;AACR,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,OAAO,GAAW,QAAQ,CAAC,CAAC,UAAU,EAAE,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,CACzE,SAAS,CAAC,UAAU,EAAE,GAAG,EAAE,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,CACzD,CAAC;AAEF;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAAW,CAAC,UAAU,EAAE,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,CAC/D,eAAe,CAAC,UAAU,EAAE,GAAG,EAAE,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC;AAEjE;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,MAAM,GAAW,QAAQ,CAAC,cAAc,CAAC,CAAC"}
+{"version":3,"file":"decoders.js","sourceRoot":"","sources":["../../src/decode/decoders.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAG7C,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAkBnD;;GAEG;AACH,SAAS,QAAQ,CACf,MAK2B;IAE3B,OAAO,CAAC,UAAU,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,CACzC,MAAM,CAAC,UAAU,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QACrD,IAAI;QACJ,UAAU,EAAE,EAAE;KACf,CAAC,CAAC,CAAC;AACR,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,OAAO,GAAW,QAAQ,CAAC,CAAC,UAAU,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,CAC1E,SAAS,CACP,UAAU,EACV,GAAG,EACH,aAAa,EACb,OAAO,CAAC,MAAM,EACd,KAAK,EACL,OAAO,CAAC,KAAK,CACd,CACF,CAAC;AAEF;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAAW,CAAC,UAAU,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,CAChE,eAAe,CACb,UAAU,EACV,GAAG,EACH,aAAa,EACb,OAAO,CAAC,MAAM,EACd,KAAK,EACL,OAAO,CAAC,KAAK,CACd,CAAC;AAEJ;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,MAAM,GAAW,QAAQ,CAAC,CAAC,UAAU,EAAE,GAAG,EAAE,EAAE,CACzD,cAAc,CAAC,UAAU,EAAE,GAAG,CAAC,CAChC,CAAC"}

package/dist/decode/edge-context.d.ts

@@ -14,10 +14,29 @@ import type { EdgeContext } from "./rules.js";
  * usually choose, since the reading charge prefers fewer words.
  */
 export declare function wordEndingAt(context: EdgeContext, at: number, longest?: number): string | undefined;
+/**
+ * Every word the dictionary has ending at a position, longest first.
+ *
+ * {@link wordEndingAt} answers with the longest, which is what a rule asking
+ * "what word is this" wants. A rule asking whether a *particular* word stands
+ * here wants all of them, because the longest match is often a word the rule
+ * has no opinion about wrapped around the one it does: 很 is a degree adverb
+ * and 得很 is the longest word ending at the same position, so 留得很长 hid its
+ * 很 and read `zhǎng`. 一个 hides 个 the same way.
+ */
+export declare function wordsEndingAt(context: EdgeContext, at: number, longest?: number): readonly string[];
 /**
  * The word starting at a position, longest first, or undefined.
  */
 export declare function wordStartingAt(context: EdgeContext, at: number, longest?: number): string | undefined;
+/**
+ * Every word the dictionary has starting at a position, longest first.
+ *
+ * The mirror of {@link wordsEndingAt}, and needed for the same reason: 五公 is a
+ * word and hides the 五 inside it, so a rule asking whether a numeral starts
+ * here has to be shown all of them.
+ */
+export declare function wordsStartingAt(context: EdgeContext, at: number, longest?: number): readonly string[];
 /**
  * The part of speech the dictionary gives a word, or the empty string.
  */

package/dist/decode/edge-context.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"edge-context.d.ts","sourceRoot":"","sources":["../../src/decode/edge-context.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C;;;;;;;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;AAED;;;;;;;GAOG;AACH,eAAO,MAAM,MAAM,aAA0C,CAAC;AAE9D;;GAEG;AACH,eAAO,MAAM,YAAY,MAAM,CAAC"}
+{"version":3,"file":"edge-context.d.ts","sourceRoot":"","sources":["../../src/decode/edge-context.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAC1B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,MAAM,GAAG,SAAS,CAYpB;AAED;;;;;;;;;GASG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,SAAS,MAAM,EAAE,CAanB;AAED;;GAEG;AACH,wBAAgB,cAAc,CAC5B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,MAAM,GAAG,SAAS,CAUpB;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAC7B,OAAO,EAAE,WAAW,EACpB,EAAE,EAAE,MAAM,EACV,OAAO,SAAI,GACV,SAAS,MAAM,EAAE,CASnB;AAED;;GAEG;AACH,wBAAgB,KAAK,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CAI5E;AAED;;;;;;;GAOG;AACH,eAAO,MAAM,MAAM,aAA0C,CAAC;AAE9D;;GAEG;AACH,eAAO,MAAM,YAAY,MAAM,CAAC"}