@kensio/pinyinjs 1.7.4 → 1.7.6

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

@@ -0,0 +1,54 @@
+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 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":"AASA,OAAO,EAEL,KAAK,QAAQ,EAId,MAAM,YAAY,CAAC;AAgCpB;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AACH,eAAO,MAAM,QAAQ,EAAE,QActB,CAAC;AAWF;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,eAAO,MAAM,WAAW,EAAE,QAczB,CAAC"}

package/dist/decode/de-rule.js

@@ -0,0 +1,113 @@
+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";
+    },
+};
+/**
+ * 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":"AASA,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;;;;;;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

@@ -1,8 +1,8 @@
 import type { Dictionary } from "../dictionary/dictionary.js";
-import { type ResolvedHints } from "./hints.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 { EdgeRule } from "./rules.js";
 import type { DecodedWord, ScoredWord } from "./word.js";
 /**
  * Decode the run's readings, scoring only where the readings are still open.
@@ -22,19 +22,6 @@ export declare function decodeReadings(lattice: Lattice, projection: ReadingProj
  * spacing question is still open.
  */
 export declare function decodeSpacing(lattice: Lattice): readonly number[];
-/**
- * Decode a Han run with the lattice: the recommended path.
- *
- * Stages 1 to 3 of ALGORITHM.md end to end — build every candidate reading,
- * lock the positions that cannot vary, score the rest — where the greedy
- * baseline commits to the longest match at each position and never revisits it.
- *
- * `before` is 汉字 standing in front of the run that the decode should see and
- * not report: the 汉字 a number written in digits would have been written with.
- * A run is decoded on its own without it, and a run reached that way has no idea
- * what preceded it — 个人 alone is the word `gèrén`, while the 两 of 2个人 makes
- * 两个 the word and leaves 人 to itself.
- */
 export declare function decodeRun(dictionary: Dictionary, run: string, rules?: readonly EdgeRule[], before?: string, hints?: ResolvedHints): readonly DecodedWord[];
 /**
  * Decode a Han run, keeping what each reading was chosen over.

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;AAG9D,OAAO,EAAE,KAAK,aAAa,EAAc,MAAM,YAAY,CAAC;AAC5D,OAAO,EAAqC,KAAK,OAAO,EAAE,MAAM,cAAc,CAAC;AAC/E,OAAO,EAGL,KAAK,iBAAiB,EACtB,KAAK,WAAW,EAGjB,MAAM,cAAc,CAAC;AAEtB,OAAO,EAAkB,KAAK,QAAQ,EAAE,MAAM,YAAY,CAAC;AAE3D,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;AA+JD;;;;;;;;;;;;GAYG;AACH,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,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"}

package/dist/decode/decode.js

@@ -1,10 +1,7 @@
-import { toCharacters } from "../script/characters.js";
 import { scoreReadings } from "./confidence.js";
-import { shiftHints } from "./hints.js";
-import { allEdges, buildLattice, cutPoints } from "./lattice.js";
+import { cutPoints } from "./lattice.js";
 import { isSettled, projectReadings, settledUnits, unitsOf, } from "./locking.js";
 import { READING_RULES } from "./reading-rules.js";
-import { applyEdgeRules } from "./rules.js";
 import { readingCost, shortestPath, spacingCost } from "./viterbi.js";
 /**
  * Decode the run's readings, scoring only where the readings are still open.
@@ -43,123 +40,7 @@ export function decodeReadings(lattice, projection) {
 export function decodeSpacing(lattice) {
     return shortestPath(lattice, 0, lattice.characters.length, spacingCost).map((edge) => edge.from);
 }
-/**
- * Gather the reading units into the words the boundaries make of them.
- *
- * Where the two decodes disagree the readings win, which is what "advisory"
- * means for the spacing decode: a boundary falling inside a reading unit is
- * dropped rather than allowed to break the unit up. 玩儿 read as `wánr` is one
- * syllable over two characters, and no spacing can be put between them.
- */
-function groupUnits(units, boundaries) {
-    const starts = new Set(units.map((unit) => unit.from));
-    const kept = new Set(boundaries.filter((at) => starts.has(at)));
-    const groups = [];
-    let held = [];
-    for (const unit of units) {
-        if (kept.has(unit.from) && held.length > 0) {
-            groups.push(held);
-            held = [];
-        }
-        held.push(unit);
-    }
-    if (held.length > 0) {
-        groups.push(held);
-    }
-    return groups;
-}
-/**
- * Assemble one word from the units the boundaries kept together.
- */
-function wordFrom(dictionary, lattice, group) {
-    const first = group[0];
-    const last = group.at(-1);
-    /* c8 ignore next 3 -- a group is only made by pushing a unit into it */
-    if (first === undefined || last === undefined) {
-        throw new Error("a word must cover at least one reading unit");
-    }
-    const text = lattice.characters.slice(first.from, last.to).join("");
-    const entry = dictionary.lookup(text);
-    return {
-        text,
-        reading: group.flatMap((unit) => [...unit.reading]),
-        isProperNoun: entry?.isProperNoun ?? false,
-        partOfSpeech: entry?.partOfSpeech ?? "",
-        isKnown: entry !== undefined,
-    };
-}
-/**
- * The lattice a run decodes over: stage 1 built, then stage 4 applied.
- *
- * The rules run between building and decoding rather than over the output,
- * which is the whole point of them — they change what the decode gets to choose
- * between, so everything downstream sees one lattice rather than a decode and
- * a list of corrections to it.
- */
-function ruledLattice(dictionary, run, rules, hints) {
-    return applyEdgeRules(buildLattice(dictionary, run, hints), dictionary, rules);
-}
-/**
- * Whether any reading in the lattice would hold the two sides of a join
- * together.
- *
- * A reading of one syllable per character can be cut anywhere; anything else —
- * 儿化, a character with no reading — is a single claim about its whole span,
- * so a claim spanning the join cannot be reported for the run alone. 1点儿事
- * is that case: 一点儿 is one `yìdiǎnr` over the 一 the context supplied and
- * the 点儿 the run did.
- */
-function isJoinedAt(lattice, at) {
-    return allEdges(lattice).some((edge) => edge.from < at &&
-        edge.to > at &&
-        edge.reading.length !== edge.to - edge.from);
-}
-/**
- * The lattice a run decodes over, with whatever context stands in front of it.
- *
- * The context is dropped where a reading would hold it to the run, leaving the
- * run decoded on its own, which is what it would have been before there was any
- * context to give.
- */
-function runLattice(dictionary, run, rules, before, hints) {
-    const alone = () => ({
-        lattice: ruledLattice(dictionary, run, rules, hints),
-        at: 0,
-    });
-    if (before === "") {
-        return alone();
-    }
-    const held = toCharacters(before).length;
-    // The context is decoded with the run, so every hint position moves along
-    // with it. Shifting a copy keeps the caller's positions relative to the run
-    // they were given for.
-    const lattice = ruledLattice(dictionary, before + run, rules, hints === undefined ? undefined : shiftHints(hints, held));
-    const at = held;
-    return isJoinedAt(lattice, at) ? alone() : { lattice, at };
-}
-/**
- * Cut a decode of the whole lattice down to the run at the end of it.
- *
- * Everything before `at` was context rather than run — 汉字 a caller put in
- * front so that the decode could see them — and is dropped once it has done its
- * work.
- */
-function runGroups({ lattice, at }, units) {
-    return groupUnits(units.filter((unit) => unit.from >= at), decodeSpacing(lattice).filter((from) => from >= at));
-}
-/**
- * Decode a Han run with the lattice: the recommended path.
- *
- * Stages 1 to 3 of ALGORITHM.md end to end — build every candidate reading,
- * lock the positions that cannot vary, score the rest — where the greedy
- * baseline commits to the longest match at each position and never revisits it.
- *
- * `before` is 汉字 standing in front of the run that the decode should see and
- * not report: the 汉字 a number written in digits would have been written with.
- * A run is decoded on its own without it, and a run reached that way has no idea
- * what preceded it — 个人 alone is the word `gèrén`, while the 两 of 2个人 makes
- * 两个 the word and leaves 人 to itself.
- */
+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);
     const { lattice } = held;

package/dist/decode/decode.js.map

@@ -1 +1 @@
-{"version":3,"file":"decode.js","sourceRoot":"","sources":["../../src/decode/decode.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAE,aAAa,EAAmB,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAsB,UAAU,EAAE,MAAM,YAAY,CAAC;AAC5D,OAAO,EAAE,QAAQ,EAAE,YAAY,EAAE,SAAS,EAAgB,MAAM,cAAc,CAAC;AAC/E,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,EAC1B,KAAgC;IAEhC,OAAO,cAAc,CACnB,YAAY,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,CAAC,EACpC,UAAU,EACV,KAAK,CACN,CAAC;AACJ,CAAC;AAUD;;;;;;;;;GASG;AACH,SAAS,UAAU,CAAC,OAAgB,EAAE,EAAU;IAC9C,OAAO,QAAQ,CAAC,OAAO,CAAC,CAAC,IAAI,CAC3B,CAAC,IAAI,EAAE,EAAE,CACP,IAAI,CAAC,IAAI,GAAG,EAAE;QACd,IAAI,CAAC,EAAE,GAAG,EAAE;QACZ,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,IAAI,CAAC,EAAE,GAAG,IAAI,CAAC,IAAI,CAC9C,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,SAAS,UAAU,CACjB,UAAsB,EACtB,GAAW,EACX,KAA0B,EAC1B,MAAc,EACd,KAAgC;IAEhC,MAAM,KAAK,GAAG,GAAe,EAAE,CAAC,CAAC;QAC/B,OAAO,EAAE,YAAY,CAAC,UAAU,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,CAAC;QACpD,EAAE,EAAE,CAAC;KACN,CAAC,CAAC;IACH,IAAI,MAAM,KAAK,EAAE,EAAE,CAAC;QAClB,OAAO,KAAK,EAAE,CAAC;IACjB,CAAC;IACD,MAAM,IAAI,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IACzC,0EAA0E;IAC1E,4EAA4E;IAC5E,uBAAuB;IACvB,MAAM,OAAO,GAAG,YAAY,CAC1B,UAAU,EACV,MAAM,GAAG,GAAG,EACZ,KAAK,EACL,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,EAAE,IAAI,CAAC,CAC1D,CAAC;IACF,MAAM,EAAE,GAAG,IAAI,CAAC;IAChB,OAAO,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;AAC7D,CAAC;AAED;;;;;;GAMG;AACH,SAAS,SAAS,CAChB,EAAE,OAAO,EAAE,EAAE,EAAc,EAC3B,KAAsB;IAEtB,OAAO,UAAU,CACf,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC,EACvC,aAAa,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CACpD,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,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":"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"}

package/dist/decode/decoders.d.ts

@@ -0,0 +1,32 @@
+/**
+ * The three decoders a conversion can be run with.
+ *
+ * Each takes a Han run and hands back words; only one of them has anything to
+ * say about how settled it was, so the other two are wrapped to report
+ * nothing rather than each caller having to allow for the difference.
+ */
+import type { Dictionary } from "../dictionary/dictionary.js";
+import type { ResolvedHints } from "./hints.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.
+ */
+export type Decode = (dictionary: Dictionary, run: string, before: string, hints: ResolvedHints | undefined) => readonly ScoredWord[];
+/**
+ * The lattice decoder, reporting nothing about its own confidence.
+ */
+export declare const LATTICE: Decode;
+/**
+ * The lattice decoder, with what each reading was chosen over.
+ */
+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.
+ */
+export declare const GREEDY: Decode;
+//# sourceMappingURL=decoders.d.ts.map

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

@@ -0,0 +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"}

package/dist/decode/decoders.js

@@ -0,0 +1,29 @@
+import { decodeRun, decodeRunScored } from "./decode.js";
+import { decodeGreedily } from "./greedy.js";
+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) => ({
+        word,
+        confidence: [],
+    }));
+}
+/**
+ * The lattice decoder, reporting nothing about its own confidence.
+ */
+export const LATTICE = unscored((dictionary, run, before, hints) => decodeRun(dictionary, run, READING_RULES, before, hints));
+/**
+ * The lattice decoder, with what each reading was chosen over.
+ */
+export const SCORED = (dictionary, run, before, hints) => decodeRunScored(dictionary, run, READING_RULES, before, hints);
+/**
+ * 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.
+ */
+export const GREEDY = unscored(decodeGreedily);
+//# sourceMappingURL=decoders.js.map

package/dist/decode/decoders.js.map

@@ -0,0 +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"}

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

@@ -0,0 +1,38 @@
+/**
+ * What a rule can ask about the words around an edge.
+ *
+ * The lookups every edge rule shares: which word ends here, which starts here,
+ * and what part of speech the dictionary gives it.
+ */
+import type { EdgeContext } from "./rules.js";
+/**
+ * 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;
+/**
+ * What follows a 教 that is teaching where no object does: aspect and the
+ * complement marker, which only a verb takes.
+ *
+ * 他教了三年书, 他教过我英语, 她教书教得很快乐. Matched as characters rather
+ * than through their tags, because what the dictionary has starting at a 得 is
+ * as likely to be 得很 as the particle itself.
+ */
+export declare const ASPECT: Set<string>;
+/**
+ * The tag prefix jieba gives a particle, which nothing teaching can follow.
+ */
+export declare const PARTICLE_TAG = "u";
+//# sourceMappingURL=edge-context.d.ts.map

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

@@ -0,0 +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"}

package/dist/decode/edge-context.js

@@ -0,0 +1,57 @@
+/**
+ * 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 ?? "");
+}
+/**
+ * What follows a 教 that is teaching where no object does: aspect and the
+ * complement marker, which only a verb takes.
+ *
+ * 他教了三年书, 他教过我英语, 她教书教得很快乐. Matched as characters rather
+ * than through their tags, because what the dictionary has starting at a 得 is
+ * as likely to be 得很 as the particle itself.
+ */
+export const ASPECT = new Set(["了", "过", "過", "着", "著", "得"]);
+/**
+ * The tag prefix jieba gives a particle, which nothing teaching can follow.
+ */
+export const PARTICLE_TAG = "u";
+//# sourceMappingURL=edge-context.js.map

package/dist/decode/edge-context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"edge-context.js","sourceRoot":"","sources":["../../src/decode/edge-context.ts"],"names":[],"mappings":"AAQA;;;;;;;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,MAAM,CAAC,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;AAE9D;;GAEG;AACH,MAAM,CAAC,MAAM,YAAY,GAAG,GAAG,CAAC"}

package/dist/decode/hint-lattice.d.ts

@@ -0,0 +1,45 @@
+import { type Lattice } from "./lattice.js";
+import type { ResolvedHints } from "./hints.js";
+/**
+ * The same hints with every position moved along by an offset.
+ *
+ * Word hints match on what is written and so are carried over untouched; only
+ * the positional ones are counted from somewhere. Used where a run is decoded
+ * with context in front of it, which moves the run's own characters up the
+ * lattice by the length of that context.
+ */
+export declare function shiftHints(hints: ResolvedHints, by: number): ResolvedHints;
+/**
+ * The hints that fall inside a run, with their positions made relative to it.
+ *
+ * A caller counts positions from the start of the whole text, while a run is
+ * decoded on its own, so the positions have to be moved to where the run put
+ * them — and the ones naming a character in some other run dropped, rather than
+ * landing on whatever happens to sit at that index here.
+ */
+export declare function hintsWithin(hints: ResolvedHints, from: number, length: number): ResolvedHints;
+/**
+ * A caller's readings, put into a lattice that was built without them.
+ *
+ * Two kinds of hint with deliberately different reach, which is the whole of
+ * the semantics:
+ *
+ * A **word** hint is an assertion about the text it names, so it rewrites the
+ * edge spanning exactly those characters and supplies the single-character
+ * edges underneath it. It says nothing about a *longer* word that happens to
+ * contain it — a bare 长 hint leaves 校长 as `xiàozhǎng`, because the dictionary
+ * knowing 校长 is better evidence about that stretch than a caller's remark
+ * about one of its characters. That is what makes a corrections table safe to
+ * accumulate: entries do not reach into words nobody was thinking about.
+ *
+ * A **positional** hint is an assertion about one character of one text, and it
+ * is the escape hatch precisely because nothing outranks it: it rewrites every
+ * edge covering that position, the enclosing word included. Only a reading of
+ * one syllable per character can be rewritten this way; a 儿化 edge is a single
+ * claim about its whole span and there is no one syllable in it to replace.
+ *
+ * Nothing is invented that a caller did not write, and nothing is dropped: an
+ * edge whose reading no hint touches comes through as it was.
+ */
+export declare function applyHints(lattice: Lattice, hints: ResolvedHints): Lattice;
+//# sourceMappingURL=hint-lattice.d.ts.map

package/dist/decode/hint-lattice.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"hint-lattice.d.ts","sourceRoot":"","sources":["../../src/decode/hint-lattice.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,KAAK,OAAO,EAAoC,MAAM,cAAc,CAAC;AAC9E,OAAO,KAAK,EAAE,aAAa,EAAY,MAAM,YAAY,CAAC;AAiB1D;;;;;;;GAOG;AACH,wBAAgB,UAAU,CAAC,KAAK,EAAE,aAAa,EAAE,EAAE,EAAE,MAAM,GAAG,aAAa,CAK1E;AAED;;;;;;;GAOG;AACH,wBAAgB,WAAW,CACzB,KAAK,EAAE,aAAa,EACpB,IAAI,EAAE,MAAM,EACZ,MAAM,EAAE,MAAM,GACb,aAAa,CAQf;AAeD;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,UAAU,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,aAAa,GAAG,OAAO,CAyF1E"}