chaffjs 0.36.0 → 0.37.0

package/src/html/html-site-chrome.ts

@@ -2,16 +2,38 @@
 // any site's wording. A landmark role (banner, search, contentinfo), a <search> element, a <header> that belongs to the
 // page rather than to a section, and a block named by a convention CMSs share: a breadcrumb trail (breadcrumb, pankuzu,
 // topicpath), a search box, the header or footer block (id or class exactly "header" or "footer"), and the U.S.
-// government's ".gov" banner (usa-banner). A block holding the page's <h1> is never chrome. Pure; a regular-expression
-// reading, not a parser.
-import { ATTRIBUTES, elementRanges, isInside, withoutElementsOpening, withoutRanges, type ElementRange } from "./html-elements.ts";
+// government's ".gov" banner (usa-banner). A block holding the page's <h1> is never chrome, nor a header holding a
+// heading followed by prose, nor a search block holding prose (a results page). Pure; a regular-expression reading,
+// not a parser.
+import {
+  ANY_LINK,
+  ATTRIBUTES,
+  CLOSED_SENTENCE,
+  elementRanges,
+  headingRanges,
+  isInside,
+  plainText,
+  withoutElementsOpening,
+  withoutRanges,
+  type ElementRange,
+} from "./html-elements.ts";
 
-const CHROME_ROLES = new Set(["banner", "search", "contentinfo"]);
+const CHROME_ROLES = new Set(["banner", "contentinfo"]);
 
-// A name segment, so that "pankuzu_wrap", "foot_pankuzu" and "cse-search-box" match and "research" does not.
-const CHROME_NAME = /(?:^|[-_])(?:breadcrumbs?|pankuzu|topic-?path|search)(?:[-_\d]|$)/iu;
+// A name segment, so that "pankuzu_wrap" and "foot_pankuzu" match and "research" does not.
+const CHROME_NAME = /(?:^|[-_])(?:breadcrumbs?|pankuzu|topic-?path)(?:[-_\d]|$)/iu;
 
-const REGION_NAME = /^(?:header|footer|usa-banner)$/iu;
+const SEARCH_NAME = /(?:^|[-_])search(?:[-_\d]|$)/iu;
+
+const REGION_NAME = /^(?:footer|usa-banner)$/iu;
+
+const HEADER_NAME = /^header$/iu;
+
+// A form's own controls, also when the form element itself is gone and only its labels remain.
+const FORM_CONTROL = /<(?:input|select|textarea|button|label|fieldset)\b/iu;
+
+/** From this many sentences a block is prose, not a search box: a results page, an article named "search-…". */
+const PROSE_SENTENCES = 2;
 
 const NAMED_OPENING = /<([a-z][a-z0-9-]*)\b[^>]*\s(?:class|id|role)\s*=/giu;
 
@@ -24,24 +46,59 @@ const tokens = (value: string): string[] => value.split(/\s+/u).filter((token) =
 
 const holdsTitle = (range: ElementRange): boolean => /<h1\b/iu.test(range.inner);
 
+const SENTENCE_ENDS = new RegExp(CLOSED_SENTENCE.source, "gu");
+
+/** Each tag read as a space, as hasSentence does, so that "Done.</li><li>Next" ends one. */
+const sentenceCount = (html: string): number => [...plainText(html.replace(/<\/?[a-z!][^>]*>/giu, " ")).matchAll(SENTENCE_ENDS)].length;
+
+/** The block's own running text: what is left without its links (a menu) and its headings (a title, a menu's labels). */
+const ownSentences = (html: string): number => sentenceCount(withoutRanges(html, headingRanges(html)).replace(ANY_LINK, " "));
+
+const isProse = (html: string): boolean => ownSentences(html) >= PROSE_SENTENCES;
+
+const holdsSentence = (html: string): boolean => ownSentences(html) > 0;
+
+/** A heading with a sentence after it: a part of the document, not a site's logo over its menu. */
+const holdsHeadedProse = (range: ElementRange): boolean => headingRanges(range.inner).some((heading) => holdsSentence(range.inner.slice(heading.end)));
+
+/** A search box or its menu: a form, controls, or links with no sentence of their own; never prose such as a results page. */
+const isSearchBox = (range: ElementRange): boolean =>
+  !isProse(range.inner) && (/^<form\b/iu.test(range.openTag) || FORM_CONTROL.test(range.inner) || !holdsSentence(range.inner));
+
+const firstRole = (openTag: string): string => (tokens(attributeValue(openTag, "role"))[0] ?? "").toLowerCase();
+
+const names = (openTag: string): string[] => [...tokens(attributeValue(openTag, "class")), ...tokens(attributeValue(openTag, "id"))];
+
 /** A role is a list of tokens and the first one counts, as for role="main". */
-const hasChromeRole = (openTag: string): boolean => CHROME_ROLES.has((tokens(attributeValue(openTag, "role"))[0] ?? "").toLowerCase());
+const hasChromeRole = (openTag: string): boolean => CHROME_ROLES.has(firstRole(openTag));
+
+const hasChromeName = (openTag: string): boolean => names(openTag).some((name) => CHROME_NAME.test(name) || REGION_NAME.test(name));
+
+/** The search landmark is chrome as marked; a name is only a hint, so the block must also look like a search box. */
+const isSearchBlock = (range: ElementRange): boolean =>
+  firstRole(range.openTag) === "search" ? !isProse(range.inner) : names(range.openTag).some((name) => SEARCH_NAME.test(name)) && isSearchBox(range);
 
-const hasChromeName = (openTag: string): boolean =>
-  [...tokens(attributeValue(openTag, "class")), ...tokens(attributeValue(openTag, "id"))].some((name) => CHROME_NAME.test(name) || REGION_NAME.test(name));
+const isHeaderBlock = (range: ElementRange): boolean => names(range.openTag).some((name) => HEADER_NAME.test(name)) && !holdsHeadedProse(range);
 
 /** A block the markup names or marks as the site's chrome, unless it holds the page's title. */
-export const isSiteChrome = (range: ElementRange): boolean => !holdsTitle(range) && (hasChromeRole(range.openTag) || hasChromeName(range.openTag));
+export const isSiteChrome = (range: ElementRange): boolean =>
+  !holdsTitle(range) && (hasChromeRole(range.openTag) || hasChromeName(range.openTag) || isSearchBlock(range) || isHeaderBlock(range));
 
 const SECTIONING = ["article", "section", "main", "aside", "nav"];
 
 /** A <header> inside an article or a section heads that part; one outside them is the page's banner. */
 const withoutPageHeaders = (html: string): string => {
   const sections = SECTIONING.flatMap((tag) => elementRanges(html, tag));
-  const banners = elementRanges(html, "header").filter((header) => !holdsTitle(header) && !sections.some((section) => isInside(section, header)));
+  const banners = elementRanges(html, "header").filter(
+    (header) => !holdsTitle(header) && !holdsHeadedProse(header) && !sections.some((section) => isInside(section, header)),
+  );
   return withoutRanges(html, banners);
 };
 
-const withoutSearchElements = (html: string): string => withoutRanges(html, elementRanges(html, "search"));
+const withoutSearchElements = (html: string): string =>
+  withoutRanges(
+    html,
+    elementRanges(html, "search").filter((search) => !isProse(search.inner)),
+  );
 
 export const withoutSiteChrome = (html: string): string => withoutPageHeaders(withoutSearchElements(withoutElementsOpening(html, NAMED_OPENING, isSiteChrome)));

package/src/latin-word-boundary.ts

@@ -33,6 +33,24 @@ export const isJoinerSpace = (text: string, boundary: Boundary, joiners: Readonl
   return joiners.has(text.slice(word.start, word.end).toLowerCase()) && standsBetweenJapanese(text, word);
 };
 
+/** チャンネルや宛先の名前（#pj_oss-team、@sato）の頭の印と、名前を作る字。名前は英字で始まる（#1件、@2名は番号と数）。 */
+const HANDLE_MARK = /[#@]/u;
+const HANDLE_CHAR = /[A-Za-z0-9_.-]/u;
+const HANDLE_HEAD = /[A-Za-z]/u;
+/** 印の前の英数字。あれば印は語やメールアドレスの中のもの（C#、sato@example）。 */
+const WORD_CHAR = /\w/u;
+
+/**
+ * チャンネルや宛先の名前（「#pj_oss-team チャンネル」「@sato さん」）の後ろの境目。名前は空白で閉じ、詰めると後ろの字まで名前に
+ * 読まれるので、空けるのは名前を閉じるためで空け方の好みではない。詰めて書いた所（#generalで）も書き手の空け方の票にしない。
+ */
+export const followsHandle = (text: string, boundary: Boundary): boolean => {
+  const before = Array.from(text.slice(0, boundary.offset));
+  const mark = before.findLastIndex((char) => !HANDLE_CHAR.test(char));
+  if (mark === -1 || !HANDLE_MARK.test(before[mark] ?? "") || !HANDLE_HEAD.test(before[mark + 1] ?? "")) return false;
+  return !WORD_CHAR.test(before[mark - 1] ?? "");
+};
+
 /**
  * 英字の語に付けて一つの語にする日本語の接尾の語（PDF版、IT化、Apple製）の前の境目。空けて書くことはほとんど無く、
  * 詰めたのは空け方の好みではない。語の頭の字が同じでも別の語（製品、型番）なら数える。suffixes は語彙表 latin-bound-suffix の語。

package/src/leading-label.ts

@@ -0,0 +1,15 @@
+import type { Span } from "./plugin.ts";
+
+/**
+ * 文の頭で、丸括弧に語を一つ入れた印（問答の（問）（答）、（注）（例））。印は決まった略した形で、本文のその語の書き方
+ * （答え、注意）とは比べない。括弧の中は字と数字だけで、句読点や空白、ほかの括弧があれば印と読まない。
+ */
+const LEADING_LABEL = /^([ \t\u3000]*)([（(][\p{L}\p{N}]{1,4}[）)])/u;
+
+/** text の頭の印の範囲。印が無ければ空。 */
+export const leadingLabelSpans = (text: string): Span[] => {
+  const match = LEADING_LABEL.exec(text);
+  const [lead, label] = [match?.[1], match?.[2]];
+  if (lead === undefined || label === undefined) return [];
+  return [{ start: lead.length, end: lead.length + label.length }];
+};

package/src/link-shape.ts

@@ -71,11 +71,23 @@ const BOLD_LABEL = /^(?:\*\*[^*\n.!?。！？,，、:：]{1,40}\*\*|__[^_\n.!?
 const COLON_LABEL = /^[^.!?。！？,，、:：]{1,40}[:：]$/u;
 
 /**
- * Whether a URL is the value of a field (`**Licence text** https://…`, `Website: https://…`): on its line only a label comes
- * before it and nothing after. There the address is the content, as in a table cell, so writing it bare is right.
+ * A member of a data object written in the text (`"company": "https://…",`), its quotes escaped or not: the line opens with
+ * a quoted key (after the object's brace, if any) and a colon, then the URL inside quotes, then at most a comma, a semicolon
+ * or the object's close. A sentence before the key makes it prose.
  */
-export const isFieldValueUrl = (lineBefore: string, lineAfter: string): boolean => {
+const MEMBER_KEY = /^[\s{]*\\?"[^"\\\n]{1,40}\\?"\s*:\s*\\?"$/u;
+const MEMBER_END = /^\\?"[,;}\]]?$/u;
+
+const isLabelledValue = (lineBefore: string, lineAfter: string): boolean => {
   if (lineAfter.trim() !== "") return false;
   const label = lineBefore.replace(LIST_MARKER, "").trim();
   return BOLD_LABEL.test(label) || COLON_LABEL.test(label);
 };
+
+/**
+ * Whether a URL is the value of a field (`**Licence text** https://…`, `Website: https://…`, `"home": "https://…"`): on its
+ * line only a label comes before it and nothing after, or it is a quoted member of a data object. There the address is the
+ * content, as in a table cell, so writing it bare is right.
+ */
+export const isFieldValueUrl = (lineBefore: string, lineAfter: string): boolean =>
+  isLabelledValue(lineBefore, lineAfter) || (MEMBER_KEY.test(lineBefore) && MEMBER_END.test(lineAfter.replaceAll(/\s/gu, "")));

package/src/name-misspelt-usual.ts

@@ -0,0 +1,25 @@
+// name-variant points at the less frequent form and names the more frequent one as the usual. When the usual form holds a
+// word the dictionary lacks, one letter from the dictionary word the pointed form has in its place ("Release Sechedule" ×3
+// beside "Release Schedule"), the frequent form is a repeated typo, and the pointed form is the right spelling. Pure.
+import { editsOf } from "./spelling.ts";
+
+/** Whether a lower-case word is a dictionary word, with its endings (lords, schedules). */
+export type IsKnown = (word: string) => boolean;
+
+/** A word the edits are made on: ASCII letters only, as editsOf makes them. */
+const PLAIN_WORD = /^[a-z]+$/u;
+
+const wordsOf = (name: string): string[] => name.toLowerCase().trim().split(/\s+/u);
+
+/** The usual form's word is a one-letter slip of the pointed form's dictionary word in the same place. */
+const isSlipOf = (usualWord: string, nameWord: string, isKnown: IsKnown): boolean =>
+  PLAIN_WORD.test(usualWord) && PLAIN_WORD.test(nameWord) && isKnown(nameWord) && !isKnown(usualWord) && editsOf(usualWord).includes(nameWord);
+
+/** Whether the two forms have the same words in the same places but where the usual one has a one-letter slip of a dictionary word. */
+export const usualIsMisspelt = (name: string, usual: string, isKnown: IsKnown): boolean => {
+  const nameWords = wordsOf(name);
+  const usualWords = wordsOf(usual);
+  if (nameWords.length !== usualWords.length) return false;
+  const differing = usualWords.flatMap((word, index) => (word === nameWords[index] ? [] : [index]));
+  return differing.length > 0 && differing.every((index) => isSlipOf(usualWords[index] ?? "", nameWords[index] ?? "", isKnown));
+};

package/src/notice-header.ts

@@ -0,0 +1,57 @@
+// An official notice's head (通達: 課資2－7 / 各国税局長 殿): a document number and an addressee with an honorific of
+// office. Pure. The words come from the language's notice-header lexicon; a notice takes no greeting and no closing.
+import type { Lexicon } from "./plugin.ts";
+
+/** The words of a notice's head, by their group in the notice-header lexicon. */
+export type NoticeWords = {
+  /** What a document number starts with (課資2－7: 課). */
+  readonly numberPrefixes: readonly string[];
+  /** The word before and the word after the number of a document number written 〜第123号. */
+  readonly numberHeads: readonly string[];
+  readonly numberTails: readonly string[];
+  /** The honorific an official notice puts after its addressee (各国税局長 殿). */
+  readonly addressees: readonly string[];
+};
+
+const patternsIn = (lexicon: Lexicon, group: string): string[] => lexicon.filter((entry) => entry.group === group).map((entry) => entry.pattern);
+
+export const noticeWordsOf = (lexicon: Lexicon): NoticeWords => ({
+  numberPrefixes: patternsIn(lexicon, "number-prefix"),
+  numberHeads: patternsIn(lexicon, "number-head"),
+  numberTails: patternsIn(lexicon, "number-tail"),
+  addressees: patternsIn(lexicon, "addressee"),
+});
+
+/** A document number is one short token of a line: 課資2－7, 総務省発第123号. */
+const MAX_NUMBER_LENGTH = 20;
+const DIGIT = /\p{Nd}/u;
+const BLANK = /\s/u;
+/** A document number is written in kanji, digits and dashes; hiragana makes the line words (課題は3つ). */
+const HIRAGANA = /\p{Script=Hiragana}/u;
+
+/** What follows the prefix: the office's kanji, then numbers joined by dashes (課資3-7: 資, 3-7). 課題3 has no dash. */
+const PREFIXED_NUMBER_BODY = /^\p{Script=Han}*\p{Nd}+(?:[-－‐−–―]\p{Nd}+)+$/u;
+
+const isNumberedForm = (line: string, words: NoticeWords): boolean =>
+  words.numberPrefixes.some((prefix) => prefix !== "" && line.startsWith(prefix) && PREFIXED_NUMBER_BODY.test(line.slice(prefix.length))) ||
+  words.numberTails.some(
+    (tail) => tail !== "" && line.endsWith(tail) && words.numberHeads.some((head) => head !== "" && line.slice(0, -tail.length).includes(head)),
+  );
+
+const isNumberShaped = (line: string): boolean => line.length <= MAX_NUMBER_LENGTH && DIGIT.test(line) && !BLANK.test(line) && !HIRAGANA.test(line);
+
+/** Whether a line is a document number alone: short, one run of characters with a digit and no hiragana, in one of the lexicon's forms. */
+export const isDocumentNumber = (line: string, words: NoticeWords): boolean => {
+  const trimmed = line.trim();
+  return isNumberShaped(trimmed) && isNumberedForm(trimmed, words);
+};
+
+/** Whether a line is an addressee: a name, then the notice's honorific at the line's end (各国税局長 殿). */
+export const isNoticeAddressee = (line: string, words: NoticeWords): boolean => {
+  const trimmed = line.trim();
+  return words.addressees.some((word) => trimmed.endsWith(word) && trimmed.length > word.length);
+};
+
+/** Whether head lines (before the first sentence) hold both a document number and an addressee of a notice. */
+export const isNoticeHead = (headLines: readonly string[], words: NoticeWords): boolean =>
+  headLines.some((line) => isDocumentNumber(line, words)) && headLines.some((line) => isNoticeAddressee(line, words));

package/src/sentence-load.ts

@@ -1,6 +1,7 @@
 // How much one sentence asks the reader to hold at once: brackets inside brackets, clauses chained one after another, and how
 // far the topic stands from the end where its predicate is. Pure: a sentence's text and tokens come in, numbers go out.
 import type { Token } from "./plugin.ts";
+import { topicClauseEnd } from "./topic-clause.ts";
 
 /**
  * Opening brackets and the mark that closes each. Quotation marks that open and close with the same character are left out, and
@@ -26,15 +27,24 @@ const OTHER_WIDTH: Readonly<Record<string, string>> = { ")": "）", "）": ")",
 /** A formula written in TeX ($d^k = -\nabla f(x^k)$): its brackets are the formula's, not the sentence's. */
 const FORMULA = /\$\$[^$]*\$\$|\$[^$\n]*\$/gu;
 
+/**
+ * A doubled parenthesis around text with no parenthesis inside (((居住の用に供したとき等)), （（注））): one mark written with two
+ * characters, as a directive sets an article's caption, not a bracket inside a bracket.
+ */
+const DOUBLED_PAREN = /\(\(([^()（）]*)\)\)|（（([^()（）]*)））/gu;
+
+const singled = (text: string): string =>
+  text.replace(DOUBLED_PAREN, (_whole: string, half?: string, full?: string) => (half === undefined ? `（${full ?? ""}）` : `(${half})`));
+
 /**
  * The deepest the brackets of a text nest: 「計画（案）」 is one, 「（注（a）を参照）」 two, plain text none. A closer with no
  * opener (the 1) of a list) is ignored, an opener never closed counts as open until the end, a closer of the other width closes
- * its opener ((1） is one), and formulas in $…$ are skipped.
+ * its opener ((1） is one), a doubled parenthesis ((題)) is one, and formulas in $…$ are skipped.
  */
 export const bracketDepth = (text: string): number => {
   // The closers still awaited, innermost last. Pushed and popped in place: copying it per character is quadratic on deep nesting.
   const awaited: string[] = [];
-  return [...text.replace(FORMULA, " ")].reduce((deepest, char) => {
+  return [...singled(text.replace(FORMULA, " "))].reduce((deepest, char) => {
     const innermost = awaited.at(-1);
     if (innermost !== undefined && (innermost === char || innermost === OTHER_WIDTH[char])) awaited.pop();
     else if (BRACKETS[char] !== undefined) awaited.push(BRACKETS[char]);
@@ -71,6 +81,7 @@ const TOPIC_BEFORE = new Set(["NOUN", "PRON", "PROPN"]);
 /**
  * How far the sentence's first topic (会議は, これは) stands from the end of the sentence, where Japanese puts the predicate the
  * topic belongs to, in characters of the words between them (spaces, and a link's address the document hides, are not read).
+ * When the topic gets its own predicate in an earlier clause (topicClauseEnd), the distance is to that clause's end.
  * undefined when the sentence has no topic. The distance is the reader's wait: everything between has to be held before the
  * topic is said to do anything.
  */
@@ -78,5 +89,5 @@ export const topicDistance = (tokens: readonly Token[]): number | undefined => {
   const topic = tokens.findIndex((token, at) => token.surface === "は" && token.pos === "ADP" && TOPIC_BEFORE.has(tokens[at - 1]?.pos ?? ""));
   const last = tokens.findLastIndex((token) => !TRAILING.has(token.pos));
   if (topic === -1 || last <= topic) return undefined;
-  return tokens.slice(topic + 1, last).reduce((sum, token) => sum + [...token.surface.replace(/\s/gu, "")].length, 0);
+  return tokens.slice(topic + 1, topicClauseEnd(tokens, topic, last) ?? last).reduce((sum, token) => sum + [...token.surface.replace(/\s/gu, "")].length, 0);
 };

package/src/structure/attached-parts.ts

@@ -1,31 +1,46 @@
 // Where an instrument attached to the document starts: a title line naming it (Exhibit A to Merger Agreement, Annex 1).
 // What is attached is a separate document (a restated charter, a form of note), so its definitions are its own. Pure.
 import { escapeRegExp } from "../orthography.ts";
-import { isTitleLine, type TitleNotion } from "../title-line.ts";
+import { isShortTitle, isTitleLine, type TitleNotion } from "../title-line.ts";
 
 /** An instrument's title: any case (EXHIBIT A to the merger agreement), but no `: ; ,` after it, and only blank lines around. */
 export const INSTRUMENT_TITLE: TitleNotion = { sentenceEnd: /[.!?。．？！:;,]$/u, capitalised: false, edge: "blank-line" };
 
-const linesOf = (source: string): { readonly start: number; readonly text: string }[] => {
+/** Bold around a title (## **SCHEDULE A**) is only how it is shown. */
+const BOLD = /\*\*/gu;
+/** A Markdown heading is a block of its own, so it needs no blank line around it (a closing fence may sit just above). */
+const MARKDOWN_HEADING = /^ {0,3}#{1,6}[ \t]/u;
+
+type Line = { readonly start: number; readonly text: string };
+
+const linesOf = (source: string): Line[] => {
   const starts = [0, ...[...source.matchAll(/\n/gu)].map((match) => match.index + 1)];
   return starts.map((start, index) => ({ start, text: source.slice(start, (starts[index + 1] ?? source.length + 1) - 1) }));
 };
 
+const opensInstrument = (source: string, lines: readonly Line[], index: number, title: RegExp): boolean => {
+  const line = lines[index];
+  if (line === undefined) return false;
+  const text = line.text.replace(BOLD, "");
+  const label = title.exec(text);
+  if (label === null) return false;
+  const rest = text.slice(label[0].length);
+  if (MARKDOWN_HEADING.test(text)) return isShortTitle(rest, INSTRUMENT_TITLE);
+  const lineEnd = index === lines.length - 1 ? -1 : line.start + line.text.length;
+  return isTitleLine(source, line.start, lineEnd, rest, INSTRUMENT_TITLE);
+};
+
 /**
  * The offsets of the lines that open an attached instrument: the line holds a label (Exhibit, Annex), its letter or
- * number and at most a short title with no full stop, and a blank line or the document's edge is on both sides.
+ * number and at most a short title with no full stop, and it is a Markdown heading or has a blank line or the
+ * document's edge on both sides. Bold around the title is ignored.
  */
 export const attachedPartStarts = (source: string, labels: readonly string[]): number[] => {
   const words = labels.filter((label) => label !== "").map(escapeRegExp);
   if (words.length === 0) return [];
   const title = new RegExp(String.raw`^[ \t]*(?:#{1,6}[ \t]+)?(?:${words.join("|")})[ \t]+[A-Z0-9](?:[A-Z0-9\-]|\.(?=[A-Z0-9]))*(?=[ \t\r]|$)`, "u");
   const lines = linesOf(source);
-  return lines.flatMap((line, index) => {
-    const label = title.exec(line.text);
-    if (label === null) return [];
-    const lineEnd = index === lines.length - 1 ? -1 : line.start + line.text.length;
-    return isTitleLine(source, line.start, lineEnd, line.text.slice(label[0].length), INSTRUMENT_TITLE) ? [line.start] : [];
-  });
+  return lines.flatMap((line, index) => (opensInstrument(source, lines, index, title) ? [line.start] : []));
 };
 
 /** Which part an offset falls in: 0 for the document itself, n for the n-th attached instrument. */

package/src/structure/definition-use.ts

@@ -2,6 +2,7 @@ import type { Span, StructureNode } from "../plugin.ts";
 import { escapeRegExp } from "../orthography.ts";
 import { inDocumentOrder } from "./issues.ts";
 import { isTitleWords } from "./contents-lines.ts";
+import { verbFormsOf } from "./gerund-phrase.ts";
 import {
   definingNameOf,
   echoesName,
@@ -109,13 +110,23 @@ const groupsOf = (terms: readonly DefinedTerm[]): TermGroup[] => {
   return [...groups.values()];
 };
 
-/** 定義したのに、定義の外で一度も使っていない語。同じ語の定義が二つあれば、最初の一つだけを言う（二つ目は duplicate-definition）。 */
-export const unusedDefinitions = (terms: readonly DefinedTerm[], texts: readonly BodyText[], mentioned: Mentioned = NEVER_MENTIONED): DefinedTerm[] => {
+/** 語か、動名詞で始まる句の動詞の形（Exercising your options の exercise your options）が、定義の外に一度でも現れるか。 */
+const isUsed = (group: TermGroup, texts: readonly BodyText[], determiners: ReadonlySet<string>): boolean =>
+  [group.first.term, ...verbFormsOf(group.first.term, determiners)].some((form) => usesOf(form, texts, group.spans).length > 0);
+
+/**
+ * 定義したのに、定義の外で一度も使っていない語。同じ語の定義が二つあれば、最初の一つだけを言う（二つ目は duplicate-definition）。
+ * determiners: 動名詞の後ろに来て、それが目的語を取る動詞だと分かる語（英語の a, the, your）。
+ */
+export const unusedDefinitions = (
+  terms: readonly DefinedTerm[],
+  texts: readonly BodyText[],
+  mentioned: Mentioned = NEVER_MENTIONED,
+  determiners: ReadonlySet<string> = new Set(),
+): DefinedTerm[] => {
   const groups = groupsOf(terms);
   const examples = quotesExamples(groups, texts, mentioned);
-  return groups
-    .filter((group) => !(examples && group.first.bare === true) && usesOf(group.first.term, texts, group.spans).length === 0)
-    .map((group) => group.first);
+  return groups.filter((group) => !(examples && group.first.bare === true) && !isUsed(group, texts, determiners)).map((group) => group.first);
 };
 
 /**
@@ -144,9 +155,24 @@ const LETTER = /\p{L}/gu;
 /** 札の文が語のほかに持ってよい字の数（"The Agreement." の The）。 */
 const LABEL_SLACK = 3;
 
-/** 語だけの札の文（**12.2.** Taxes.、1. The Agreement.）。条の頭に語を掲げただけで、使用ではない。 */
+/**
+ * 札の終わりに添えた条の番地の注（（第7条関係）、(Section 4)）。同じ形の括弧の一組で、入れ子が無く、中身は番号を含む 20 字まで。
+ * 長い括弧、番号の無い括弧（(if any)）、文の途中の括弧は、文の一部として読む。
+ */
+const SECTION_NOTE = /(?:（[^（）()]{0,20}）|\([^（）()]{0,20}\))[.。．]?\s*$/u;
+const NUMBER = /[\p{N}〇一二三四五六七八九十]/u;
+
+const withoutSectionNote = (text: string): string => {
+  const note = SECTION_NOTE.exec(text);
+  return note === null || !NUMBER.test(note[0]) ? text : text.slice(0, note.index);
+};
+
+/**
+ * 語だけの札の文（**12.2.** Taxes.、1. The Agreement.、（1）文化芸術推進基本計画（第7条関係））。条や節の頭に語を掲げただけで、
+ * 使用ではない。終わりに添えた条の番地の注は札の字に数えない。
+ */
 const isLabel = (body: BodyText | undefined, term: string): boolean =>
-  body !== undefined && (body.text.replaceAll(term, "").match(LETTER) ?? []).length <= LABEL_SLACK;
+  body !== undefined && (withoutSectionNote(body.text.trimEnd()).replaceAll(term, "").match(LETTER) ?? []).length <= LABEL_SLACK;
 
 /** 番地の後ろの、条の題の始まり（Section 1.1   Closing. の Closing）。 */
 const AFTER_ADDRESS = /\d\.?\s+(?=\p{Lu})/gu;

package/src/structure/gerund-phrase.ts

@@ -0,0 +1,31 @@
+// A defined phrase led by an English gerund and its object ("Exercising your options" means ...) names an act, and the
+// document then writes the act as a verb: exercise your options, exercised their options. Pure.
+
+const GERUND = /^(\p{Script=Latin}{2,})ing$/u;
+/** A consonant written twice before -ing (running, setting): the verb has it once. */
+const DOUBLED = /(\p{Script=Latin})\1$/u;
+
+/** The verb forms a stem takes (file, files, filed, filing for "Filing"; exercise for "Exercising"). */
+const formsOf = (stem: string): string[] => {
+  const single = DOUBLED.test(stem) ? [stem.slice(0, -1)] : [];
+  return [...new Set([stem, `${stem}e`, `${stem}es`, `${stem}s`, `${stem}ed`, `${stem}ing`, ...single, ...single.map((verb) => `${verb}s`)])];
+};
+
+const capitalised = (word: string): boolean => word.charAt(0) !== word.charAt(0).toLowerCase();
+
+/**
+ * The same phrase with its leading gerund written as the other forms of its verb ("exercise your options",
+ * "exercised your options" for "Exercising your options"). Only when a determiner follows the gerund (a, the, your):
+ * then it takes an object and is a verb. "Building materials" and "Pricing information" name things, and "Licensing"
+ * alone may be a noun. Empty when the phrase does not take that shape.
+ */
+export const verbFormsOf = (term: string, determiners: ReadonlySet<string>): string[] => {
+  const [first, second, ...rest] = term.split(" ");
+  if (first === undefined || second === undefined || !determiners.has(second.toLowerCase())) return [];
+  const stem = GERUND.exec(first.toLowerCase())?.[1];
+  if (stem === undefined) return [];
+  return formsOf(stem)
+    .map((form) => (capitalised(first) ? `${form.charAt(0).toUpperCase()}${form.slice(1)}` : form))
+    .map((form) => [form, second, ...rest].join(" "))
+    .filter((form) => form !== term);
+};

package/src/structure/universal.ts

@@ -5,7 +5,7 @@ import { numberInSentence } from "./number-in-sentence.ts";
  * 「4.2 設定」「3.1.4 Scope」のような通し番号。数字と点だけで書くので言語を問わない。
  * 各部分を 3 桁・深さを 6 までに抑え、後戻りが爆発しない形にしてある。
  */
-const DOTTED = /^[ \t]{0,3}(?<number>\d{1,3}(?:\.\d{1,3}){0,5})(?<closing>[.．])?[ \t\u3000]+(?<rest>\S.*)$/u;
+const DOTTED = /^[ \t]{0,3}(?<number>\d{1,3}(?:\.\d{1,3}){0,5})(?<closing>[.．])?(?<gap>[ \t\u3000]+)(?<rest>\S.*)$/u;
 /**
  * 見出しで、点で閉じた番号に題を詰めて書いたもの（「7.委託」「7．委託」「2.Overview」）。後ろが数字なら小数（「1.5万人」）、
  * 小文字なら版（「2.x 系」）なので読まない。
@@ -26,8 +26,11 @@ const isTopSection = (line: string, rest: string): boolean => TOP_SECTION.test(l
 const dottedGroups = (line: string, context: NumberingContext): Readonly<Record<string, string | undefined>> | undefined =>
   DOTTED.exec(line)?.groups ?? (context.isHeading ? TIGHT_HEADING.exec(line)?.groups : undefined);
 
-/** closedByDot: 番号を点で閉じたか（「5. 」）。「1.5 万人」の 1.5 は閉じていない。 */
-type DottedLine = NumberedLine & { readonly closedByDot: boolean };
+/**
+ * closedByDot: 番号を点で閉じたか（「5. 」）。「1.5 万人」の 1.5 は閉じていない。wideGap: 番号と題の間に全角の空白を置いたか
+ * （「9（全角空白）年間収入階級」）。どちらも番号の札の書き方で、数量はそう書かない。
+ */
+type DottedLine = NumberedLine & { readonly closedByDot: boolean; readonly wideGap: boolean };
 
 /**
  * 本文の「1. 」は箇条書きで、見出しの「1. 」は章番号。点を含まない番号は、見出しの行か、テキストの章見出しの形でだけ読む。
@@ -41,15 +44,27 @@ export const dottedNumber = (line: string, context: NumberingContext, plainText
   const rest = groups?.["rest"]?.trim() ?? "";
   if (depth < 2 && !context.isHeading && !(plainText && isTopSection(line, rest))) return undefined;
   const closedByDot = groups?.["closing"] !== undefined;
-  return { kind: "article", depth, number, absolute: true, label: number, heading: rest, rest, ordinal: Number(number.split(".").at(-1)), closedByDot };
+  const wideGap = groups?.["gap"]?.includes("\u3000") === true;
+  return {
+    kind: "article",
+    depth,
+    number,
+    absolute: true,
+    label: number,
+    heading: rest,
+    rest,
+    ordinal: Number(number.split(".").at(-1)),
+    closedByDot,
+    wideGap,
+  };
 };
 
 /**
- * 番号の後ろが単位なら数量（「1.5 万人」「2.5 days」）。点で閉じた番号（「5. ページ自身の通信」「2. Days off」）は札で、
- * 数量はそう書かないので、後ろに単位の語が来ても数量にしない。
+ * 番号の後ろが単位なら数量（「1.5 万人」「2.5 days」）。点で閉じた番号（「5. ページ自身の通信」「2. Days off」）と、全角の空白を
+ * 置いた番号（「9（全角空白）年間収入階級」）は札で、数量はそう書かないので、後ろに単位の語が来ても数量にしない。
  */
 export const isAmount = (dotted: DottedLine, countedAfter: StructurePatterns["countedAfter"]): boolean =>
-  !dotted.closedByDot && countedAfter?.(dotted.number, dotted.rest) === true;
+  !dotted.closedByDot && !dotted.wideGap && countedAfter?.(dotted.number, dotted.rest) === true;
 
 /** 年まで書いた日付の値（2022-12-08）。年の無い「2 May」は、番号 2 の題「May」とも読める。 */
 const WITH_YEAR = /^\d{4}-/u;

package/src/title-line.ts

@@ -26,7 +26,8 @@ const ENDS_LINE = /^[ \t\r]*\n/u;
 const BLANK = /^[ \t\r]*$/u;
 const LOWER_START = /^\p{Ll}/u;
 
-const isShortTitle = (rest: string, notion: TitleNotion): boolean => {
+/** rest, the words after a label and its number, are a short title by notion's measure. */
+export const isShortTitle = (rest: string, notion: TitleNotion): boolean => {
   const title = rest.trim();
   const words = title.split(/\s+/u).filter((word) => word !== "");
   return (