p7d-markdown-it-p-captions

p7d-markdown-it-p-captions is a markdown-it plugin. For a paragraph, it determines if it is a caption from the string at the beginning of the paragraph and adds a class attribute value to indicate that it is a caption.

This package uses ESM JSON import attributes, so use a runtime version that supports the with { type: 'json' } import-attributes syntax.

import mdit from 'markdown-it'
import mditPCaption from 'p7d-markdown-it-p-captions'

const md = mdit().use(mditPCaption)

const src = 'Figure 1. A caption.\n';
console.log(md.render(src));
// <p class="caption-img"><span class="caption-img-label">Figure 1<span class="caption-img-label-joint">.</span></span> A caption.</p>

If you want to change the prefix of the class name from ‘caption’, set the option as follows.

const md = mdit().use(mditPCaption, { classPrefix: 'f' })

const src = 'Figure 1. A caption.\n';
console.log(md.render(src));
// <p class="f-img"><span class="f-img-label">Figure 1<span class="f-img-label-joint">.</span></span> A caption.</p>

classPrefix is normalized at plugin setup:

• null / undefined use the default prefix, caption.
• surrounding whitespace is trimmed.
• an empty string is accepted as an explicit no-prefix setting.

const md = mdit().use(mditPCaption, { classPrefix: '' })

console.log(md.render('Figure 1. A caption.\n'));
// <p class="img"><span class="img-label">Figure 1<span class="img-label-joint">.</span></span> A caption.</p>

Install the plugin only once on a markdown-it instance. If .use(mditPCaption) is called more than once on the same instance, later calls are ignored. Use a separate markdown-it instance when you need a different option set.

Helper APIs for Integrators

For integrations such as p7d-markdown-it-figure-with-p-caption, the package also exports helpers:

import mditPCaption, {
  analyzeCaptionParagraph,
  analyzeCaptionStart,
  applyCaptionParagraph,
  buildLabelClassLookup,
  buildLabelPrefixMarkerRegFromMarkers,
  canonicalizeCaptionNumberingMark,
  createCaptionNumberingPolicy,
  createCaptionNumberingRuntime,
  getGeneratedLabelDefaults,
  getFallbackLabelForText,
  isCaptionLabelBoundary,
  isCaptionLabelForMark,
  setCaptionParagraph,
  getMarkRegForLanguages,
  getMarkRegStateForLanguages,
  normalizeAutoLabelNumberSets,
  normalizeLabelPrefixMarkers,
  stripLabelPrefixMarker,
} from 'p7d-markdown-it-p-captions';

• getMarkRegForLanguages(languages) returns the cached matcher map used for caption-start detection. Each matcher exposes exec() / test() like a regex, but mixed-language configs may use multiple compiled regexes internally.
• getMarkRegStateForLanguages(languages) returns the full prebuilt state (languages, markReg, markRegEntries, candidate entry tables, and generatedLabelDefaultsByLang). Valid language order is preserved after filtering duplicates and unsupported languages. If languages is explicitly provided but none of them are supported, it returns an empty state instead of falling back to the default en/ja set.
• analyzeCaptionStart(text, options) performs a pure read-only caption-start analysis without mutating markdown-it tokens.
• analyzeCaptionParagraph(index, state, context, options) applies the same paragraph/list and iframe/social/video guards as setCaptionParagraph, but only returns a frozen decision. It does not mutate tokens, context, or options.
• applyCaptionParagraph(decision, state, context, numberingState, options) applies a decision returned by analyzeCaptionParagraph. It verifies the token array, block/inline token identities and types, inline content, children array, and every child identity/type/content before mutation; stale or copied decisions fail closed and return false.
• createCaptionNumberingPolicy(options) validates and freezes a reusable generic numbering policy. createCaptionNumberingRuntime(policy, renderContext) creates the mutable, render-local counter owner passed in the existing numberingState helper argument.
• isCaptionLabelBoundary(text, offset, { layout, hasNumber }) exposes the same spaced/compact suffix and joint boundary used by caption recognition without embedding chapter, appendix, or other integrator-specific vocabulary.
• isCaptionLabelForMark(labelText, mark, markRegState) tests a source label word against one mark using the same language pattern and case policy as caption detection. It accepts only states returned by getMarkRegStateForLanguages() and returns false for invalid, inactive, copied, or duck-typed state objects.
• canonicalizeCaptionNumberingMark(value) converts only the public code / samp aliases to pre-code / pre-samp and preserves other values; normalizeAutoLabelNumberSets(value) applies that rule while validating the strict standalone/figure-facing allowlist.
• getGeneratedLabelDefaults(mark, text, markRegState, preferredLanguages) resolves locale-aware generated-label metadata such as { label: 'Figure', joint: '.', space: ' ' }. preferredLanguages is optional and is used only as the tie-break order for ambiguous unlabeled fallback text; unsupported or inactive hints fall back to the state language order instead of disabling fallback generation.
• getFallbackLabelForText(mark, text, markRegState, preferredLanguages) remains available as the small helper that returns only the generated label word.
• normalizeLabelPrefixMarkers(value), buildLabelPrefixMarkerRegFromMarkers(markers), and stripLabelPrefixMarker(inlineToken, markerText) expose the same label-prefix handling used internally by setCaptionParagraph.
• buildLabelClassLookup(options) remains available for older integrators that still locate already-created label spans. New numbering integrations should let p-captions apply numbers while it constructs those tokens.
• The returned state is a shared cache object. Treat it as read-only; if you need to modify it, clone it first.

If you call the paragraph helpers directly (outside md.use(mditPCaption, options)), pass languages or a prebuilt markRegState when you use non-default languages. Reusing a prebuilt state avoids repeated caller-side setup:

const opt = {
  languages: ['en'],
  classPrefix: 'caption',
  markRegState: getMarkRegStateForLanguages(['en']),
  // ...other options used by setCaptionParagraph
};

setCaptionParagraph remains the compatibility wrapper around analyze + apply. It mutates the token stream and returns a boolean: true when it detected and transformed a caption paragraph, otherwise false. Direct callers may pass partial options; missing fields use the same defaults as the plugin setup path. Treat the options object as immutable after the first helper call; create a new options object if you need different settings.

The helper's fourth argument accepts either the legacy plain { img, table } counter used with setFigureNumber: true, or a runtime returned by createCaptionNumberingRuntime(). A branded runtime owns its policy independently of setFigureNumber; this lets an integrator enable only the captions that have passed its final structural checks. Direct helpers never create a runtime merely because autoLabelNumberSets appears in their formatting options.

When context is provided to applyCaptionParagraph, or sp is provided to setCaptionParagraph, the helper stores the full decision object as captionDecision:

context.captionDecision = {
  paragraphIndex,
  inlineIndex,
  mark,              // normalized mark key (e.g., 'img', 'table', 'video')
  kind,              // 'caption' or 'label-only'
  matchedText,
  labelText,         // detected label word without number/joint
  number,            // original string, including forms such as '01' or 'A.5'
  joint,
  bodyText,
  hasExplicitNumber, // true only when the source label included a number
  prefixMarker,
};

captionDecision always describes the source. An auto-generated number does not replace number or change hasExplicitNumber; deferred frozen decisions are never mutated.

Pass the frozen decision object returned by analyzeCaptionParagraph directly to applyCaptionParagraph; cloning it intentionally loses the private validation snapshot. Apply decisions before rebuilding or reindexing the block token array. Different paragraph decisions can be analyzed first and then applied because applying a caption only mutates that paragraph and its inline children.

The optional analyzer context uses { captionName, isIframeTypeBlockquote, isVideoIframe }. The apply context is also used for figure-class mirroring (figureClassName) and receives captionDecision after a successful mutation.

Language files under lang/*.json may define generatedLabelDefaults. These are used by getGeneratedLabelDefaults and also exposed on markRegState.generatedLabelDefaultsByLang.

Language metadata under lang/*.json also defines:

• type['label-layout']: caption label syntax shape ('spaced' for forms like Figure 1. Caption, 'compact' for forms like 図1　キャプション)
• markReg entries may be plain strings or { pattern, 'match-case' } objects
• 'match-case': label-word matching strategy ('auto', 'ascii', 'unicode', or 'raw')
• pattern should contain only the label word/phrase portion. The package wraps it internally; you do not need to add your own outer capture group.

If you want to add a new language to the package, see docs/adding-language.md.

When multiple languages are active, generated-label fallback resolves in this order:

1. preferredLanguages reordered onto the active state.languages set (if provided)
2. script-specific detection (for example Japanese text -> ja)
3. the first remaining language in that tie-break order

Because getMarkRegStateForLanguages(languages) preserves the caller's valid language order, that order is the final tie-break when preferredLanguages and script-specific detection do not choose a language. For example, ['ja', 'en'] prefers Japanese generated labels for otherwise ambiguous text, while the default plugin language order remains ['en', 'ja'].

This affects only unlabeled fallback text such as autoAltCaption: true integrations. Normal caption detection still checks every enabled language regex.

{
  "type": {
    "label-layout": "spaced"
  },
  "markReg": {
    "img": {
      "pattern": "(fig(?:ure)?|photo)",
      "match-case": "auto"
    }
  },
  "generatedLabelDefaults": {
    "img": {
      "label": "Figure",
      "joint": ".",
      "space": " "
    }
  }
}

Example pure analyzer usage:

const markRegState = getMarkRegStateForLanguages(['en', 'ja']);

analyzeCaptionStart('Figure 1. A caption.', {
  markRegState,
  preferredMark: 'img',
});
// {
//   mark: 'img',
//   kind: 'caption',
//   matchedText: 'Figure 1.',
//   labelText: 'Figure',
//   number: '1',
//   joint: '.',
//   bodyText: 'A caption.',
//   hasExplicitNumber: true,
//   prefixMarker: ''
// }

Example paragraph analyze/apply usage:

const context = {
  captionName: 'img',
  isIframeTypeBlockquote: false,
  isVideoIframe: false,
};
const decision = analyzeCaptionParagraph(paragraphIndex, state, context, opt);

if (decision) {
  applyCaptionParagraph(decision, state, context, figureNumberState, opt);
}

Shared numbering policy/runtime

The generic numbering API does not know about chapters or document structure. The caller supplies an opaque sequence key and any display context it needs:

const policy = createCaptionNumberingPolicy({
  enabledMarks: ['img', 'pre-code', 'pre-samp', 'video'],
  explicitCounter: 'max', // 'max' | 'replace' | 'ignore'
  getCounterKey({ mark }) {
    return mark === 'img' || mark === 'pre-samp' ? 'figure' : mark;
  },
  getSequenceKey({ captionContext, counterKey }) {
    return captionContext.numbering.sequenceKey;
  },
  parseExplicitNumber({ number, captionContext }) {
    const numbering = captionContext.numbering;
    if (!numbering.scoped) return /^[1-9][0-9]*$/.test(number) ? Number(number) : null;
    const prefix = numbering.displayPrefix + numbering.separator;
    if (!number.startsWith(prefix)) return null;
    const tail = number.slice(prefix.length);
    return /^[1-9][0-9]*$/.test(tail) ? Number(tail) : null;
  },
  formatGeneratedNumber({ sequence, captionContext }) {
    const numbering = captionContext.numbering;
    return numbering.scoped
      ? numbering.displayPrefix + numbering.separator + sequence
      : String(sequence);
  },
});

const numberingRuntime = createCaptionNumberingRuntime(policy, { env: state.env });
setCaptionParagraph(paragraphIndex, state, caption, numberingRuntime, context, opt);

Policy/runtime rules:

• Create a stable policy once during plugin setup and one runtime per Markdown render.
• enabledMarks accepts deduplicated canonical identifiers matching ^[a-z][a-z0-9]*(?:-[a-z0-9]+)*$. It is trusted setup input and has no arbitrary array-length or identifier-length cap. The generic API does not treat code or samp as aliases; user-facing autoLabelNumberSets performs that conversion.
• explicitCounter: 'max' advances only to a greater parsed explicit value, 'replace' assigns the parsed value, and 'ignore' preserves the source number without consulting sequence/parser callbacks or changing counter state.
• getCounterKey() selects the semantic counter series and defaults to the decision mark. It returns a non-empty primitive string of at most 256 UTF-16 code units or a finite number. Equal counter keys let different marks share one sequence.
• getSequenceKey() returns null for the unscoped counter or a primitive string/finite number for a scoped partition, and must not return undefined. The runtime identifies state by counterKey × sequenceKey without string-concatenating the pair.
• getSequenceKey() receives the validated counterKey. parseExplicitNumber() and formatGeneratedNumber() receive both validated keys. The parser returns null or a positive safe integer; the formatter returns a primitive, non-empty string accepted by the caption number grammar. Promises/thenables are unsupported.
• Every callback and return value is evaluated before p-captions removes a prefix marker, adds paragraph classes, or changes inline tokens. If a callback fails, p-captions does not commit its token/counter changes. Callbacks themselves are required to be pure synchronous functions.
• Generated output from the strict policy/runtime API is validated even when the built-in decimal formatter is used. An individual segment longer than six characters (for example 1000000) throws a RangeError before mutation. The legacy setFigureNumber path intentionally keeps its historical overflow behavior.
• generatedNumberHasNumClass: false is available for an integrator that must preserve a legacy pipeline where label-has-num was decided before its later numbering pass. The default is true.

Caption detection rules

First, the strings listed in the table below are required as the first string of a paragraph to be determined as a caption.

| class attribute value | Character string at the beginning of a paragraph (uppercase or lowercase) | | ---- | ---- | | caption-img | fig, figure, illust, photo, 図, 画像, イラスト, 写真 | | caption-video | movie, video, 動画, ビデオ | | caption-table | table, 表 | | caption-pre-code | code, sourcecode[^table-note1], codeblock[^table-note1], program, コード, ソースコード, リスト,プログラム, 算譜 | | caption-pre-samp | console, terminal, prompt, command, commandprompt[^table-note1], 端末, ターミナル, コマンド, コマンドプロンプト, プロンプト, リスト [^table-note2], 図[^table-note2] | | caption-blockquote | source, cited, citation, quote, blockquote[^table-note1], 引用, 引用元, 出典 | | caption-slide| slide, スライド, 発表資料 | | caption-audio| audio, sound, 音, 音声, 音楽, サウンド |

[^table-note1]: 'sourcecode', 'codeblock', 'commandprompt', 'blockquote' allow spaces between words. ex. For example, 'source code' is acceptable. [^table-note2]: 'リスト' and '図' is also applicable only when used via p7d-markdown-it-figure-with-p-caption.

Additionally, a delimiter is required after these strings ([.:．。：　]) as shown below. For half-width character strings, an additional space is required. In Japanese labels, both half-width and full-width spaces can be used as delimiters.

Fig. A caption

Fig: A caption

図．キャプション

図。キャプション

図：キャプション

図　キャプション

図 キャプション

Notice. Even if the label is in English, if the caption body does not start with [0-9a-zA-Z], delimiters are not necessary and only a half-width space is required to convert the caption. ex. Figure 猫

You can also put a serial number, such as 0-9A-Z.-, between the first term and the separator.

Fig 1. A caption

Fig 1.1. A caption

Fig A: A caption

図1.1：キャプション

Only when it has this serial number, it can be identified by omitting the separator and adding only a space. In English, the caption written after a space must begin with an uppercase letter.

Fig 1 A caption.

Fig 1.1 A caption.

Figure A A caption

Also, It identifies the Figure.1 type. This format has a dot immediately after the first term, a serial number and a space after it. In this case, too, the caption written after a space must begin with an uppercase letter.

Figure.1 A caption.

Note: paragraphs immediately following list_item_open are intentionally excluded from caption detection.

Options

Quick reference

| option | purpose | | ---- | ---- | | languages | Restrict caption-label detection to selected languages in lang/*.json. | | classPrefix | Prefix for paragraph/label/body class names (default: caption). | | dquoteFilename / strongFilename | Extract filename token right after a caption label. | | bLabel / strongLabel | Use b/strong tag for the label wrapper instead of span; strongLabel wins if both are true. | | labelPrefixMarker | Allow one/two marker strings before labels (stripped on match). | | hasNumClass | Add label-has-num when label includes a number. | | jointSpaceUseHalfWidth | Convert full-width joint-space labels to half-width body spacing. | | removeUnnumberedLabel | Remove label span when the label has no number. | | removeUnnumberedLabelExceptMarks | Keep unnumbered labels for specific marks. | | removeMarkNameInCaptionClass | Use caption-label/caption-body style classes without mark suffix. | | wrapCaptionBody | Wrap caption body in <span class="...-body">...</span>. | | autoLabelNumberSets | Strict mark allowlist for standalone automatic numbering (img, table, code, samp, video, or canonical code/samp names). | | setFigureNumber | Auto-number img/table captions when number is omitted. | | labelClassFollowsFigure | Reuse integrator-provided figure class base for label/body spans. | | figureToLabelClassMap | Override label/body class bases for specific figure classes. |

Option: Specify file name

Specify the file name before writing the caption. Note that a space is required between the file name and caption.

Use double quote

md.use(mditPCaption, {dquoteFilename: true});

const src = 'Code. "Filename.js" Call a cat.';
console.log(md.render(src));
// <p class="caption-pre-code"><span class="caption-pre-code-label">Code<span class="caption-pre-code-label-joint">.</span></span> <strong class="caption-pre-code-filename">Filename.js</strong> Call a cat.</p>\n

Use strong quote

md.use(mditPCaption, {strongFilename: true});

const src = 'Code. **Filename** A caption.\n';
console.log(md.render(src));
// <p class="caption-pre-code"><span class="caption-pre-code-label">Code<span class="caption-pre-code-label-joint">.</span></span> <strong class="caption-pre-code-filename">Filename</strong> A caption.</p>\n

Option: Set class indicating having label number

If the labels have numbers, add class: label-has-num

md.use(mditPCaption, {hasNumClass: true});

const src = 'Code 1. A caption.\n';
console.log(md.render(src));
//<p class="caption-pre-code"><span class="caption-pre-code-label label-has-num">Code 1<span class="caption-pre-code-label-joint">.</span></span> A caption.</p>\n'

Option: Use b/strong element for label

md.use(mditPCaption, {bLabel: true});

const src = 'Code 1. A caption.\n';
console.log(md.render(src));
//<p class="caption-pre-code"><b class="caption-pre-code-label">Code 1<span class="caption-pre-code-label-joint">.</span></b> A caption.</p>\n'

md.use(mditPCaption, {strongLabel: true});

const src = 'Code 1. A caption.\n';
console.log(md.render(src));
//<p class="caption-pre-code"><strong class="caption-pre-code-label">Code 1<span class="caption-pre-code-label-joint">.</span></strong> A caption.</p>\n'

Option: Use label prefix marker

Allow a marker before the label and strip it when a caption is detected.

md.use(mditPCaption, { labelPrefixMarker: '▼' });

const src = '▼Figure. A caption.\n';
console.log(md.render(src));
// <p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">.</span></span> A caption.</p>\n

labelPrefixMarker also accepts an array (only the first two entries are used):

md.use(mditPCaption, { labelPrefixMarker: ['▼', '▲'] });

Array entries must be non-empty strings. null, undefined, empty strings, and non-string values are ignored. When marker strings overlap, the longest marker is matched first so ['*', '**'] behaves the same as ['**', '*'].

Option: Convert full-width space in label joint to half-width

md.use(mditPCaption, {jointSpaceUseHalfWidth: true});

const src = '図　キャプション\n';
console.log(md.render(src));
//<p class="caption-img"><span class="caption-img-label">図</span> キャプション</p>\n'

Option: Remove unnumbered Label

md.use(mditPCaption, {removeUnnumberedLabel: true});

const src1 = '図　キャプション\n';
console.log(md.render(src1));
//<p class="caption-img">キャプション</p>\n'

const src2 = '図1　キャプション\n';
console.log(md.render(src2));
//<p class="caption-img"><span class="caption-img-label">図1<span class="caption-img-label-joint">　</span></span>キャプション</p>\n'

Option: removeUnnumberedLabelExceptMarks

md.use(mditPCaption, {
  removeUnnumberedLabel: true,
  removeUnnumberedLabelExceptMarks: ["blockquote"],
});

const src1 = '図　キャプション\n';
console.log(md.render(src1));
//<p class="caption-img">キャプション</p>\n'

const src2 = '出典　キャプション\n';
console.log(md.render(src2));
//<p class="caption-blockquote"><span class="caption-blockquote-label">出典<span class="caption-blockquote-label-joint">　</span></span>キャプション</p>\n'

Option: removeMarkNameInCaptionClass

Remove the mark (name) of the class attribute of the span element.

md.use(mditPCaption, {
  removeMarkNameInCaptionClass: true,
});

const src = '図　キャプション\n';
console.log(md.render(src));
//<p class="caption-img"><span class="caption-label">図<span class="caption-label-joint">　</span></span>キャプション</p>\n'

Option: Mirror figure classes for label/body

When labelClassFollowsFigure is enabled and sp.figureClassName is provided by an integrator, label/body span classes also reuse those figure class bases.

md.use(mditPCaption, { labelClassFollowsFigure: true });

You can override bases for specific figure class strings. When figureToLabelClassMap is provided, labelClassFollowsFigure is enabled automatically unless you explicitly set it to false.

md.use(mditPCaption, {
  figureToLabelClassMap: {
    'figure-img': 'media-figure', // => media-figure-label / media-figure-label-joint / media-figure-body
  },
});

Option: Wrap caption body with span

Wrap the caption body (everything after the label and optional filename) in a dedicated span element.

md.use(mditPCaption, { wrapCaptionBody: true });

const src = 'Figure. A cat.\n';
console.log(md.render(src));
//<p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">.</span></span> <span class="caption-img-body">A cat.</span></p>

When removeMarkNameInCaptionClass is enabled, the wrapper class becomes caption-body.

Option: Automatically number selected caption marks

Use autoLabelNumberSets for the strict standalone numbering path. Supported values are img, table, code / pre-code, samp / pre-samp, and video.

md.use(mditPCaption, {
  autoLabelNumberSets: ['img', 'table', 'code', 'samp', 'video'],
});

• code and samp are normalized to the canonical marks pre-code and pre-samp.
• An explicit array takes precedence over legacy setFigureNumber; [] explicitly disables numbering.
• Explicit undefined, null, non-arrays, unsupported marks, and invalid entries throw during the first plugin setup. A duplicate .use() remains first-install-wins and ignores later options.
• Counters are separate by canonical mark in standalone mode. p-captions has no adjacent-block context, so an ambiguous Japanese 図 is classified as img and リスト as pre-code; it does not infer a samp block.
• With removeUnnumberedLabel: true, the strict path numbers only marks also present in removeUnnumberedLabelExceptMarks. The aliases code and samp are accepted there as well.
• Every render creates fresh render-local counters.

Legacy option: Automatically number figures and tables

Turn on automatic numbering for image (caption-img) and table (caption-table) captions that do not already include a number.

md.use(mditPCaption, { setFigureNumber: true });

const src = 'Figure. A cat.\n\nTable. A list.\n\nFigure. Another cat.\n';
console.log(md.render(src));
// <p class="caption-img"><span class="caption-img-label">Figure 1<span class="caption-img-label-joint">.</span></span> A cat.</p>
// <p class="caption-table"><span class="caption-table-label">Table 1<span class="caption-table-label-joint">.</span></span> A list.</p>
// <p class="caption-img"><span class="caption-img-label">Figure 2<span class="caption-img-label-joint">.</span></span> Another cat.</p>

• Counters are maintained separately for figures and tables and reset for every Markdown render.
• Captions that already contain a number keep that number, and the counter is updated so the next auto-generated value continues the sequence.
• This compatibility path intentionally preserves historical 999999 -> 1000000 generation. New autoLabelNumberSets numbering uses strict pre-mutation validation instead.

Detection examples (reference)

  [
    'Figure',
    '<p>Figure</p>\n'
  ], [
    'Figure ',
    '<p>Figure</p>\n'
  ], [
    'Figure.',
    '<p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">.</span></span></p>\n'
  ], [
    'Figure:',
    '<p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">:</span></span></p>\n'
  ], [
    'Figure 1',
    '<p class="caption-img"><span class="caption-img-label">Figure 1</span></p>\n'
  ], [
    'Figure A.1',
    '<p class="caption-img"><span class="caption-img-label">Figure A.1</span></p>\n'
  ], [
    'Figure. A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">.</span></span> A cat.</p>\n'
  ], [
    'Figure: A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure<span class="caption-img-label-joint">:</span></span> A cat.</p>\n'
  ], [
    'Figure is a cat.',
    '<p>Figure is a cat.</p>\n'
  ], [
    'Figure 1. A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure 1<span class="caption-img-label-joint">.</span></span> A cat.</p>\n'
  ], [
    'Figure 1 is a cat.',
    '<p>Figure 1 is a cat.</p>\n'
  ], [
    'Figure A A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure A</span> A cat.</p>\n'
  ], [
    'Figure 1 A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure 1</span> A cat.</p>\n'
  ], [
    'Figure 1 a cat.',
    '<p>Figure 1 a cat.</p>\n'
  ], [
    'Figure 1: A cat.',
    '<p class="caption-img"><span class="caption-img-label">Figure 1<span class="caption-img-label-joint">:</span></span> A cat.</p>\n'
  ], [
    '図',
    '<p>図</p>\n'
  ], [
    '図 ',
    '<p>図</p>\n'
  ], [
    '図.',
    '<p class="caption-img"><span class="caption-img-label">図<span class="caption-img-label-joint">.</span></span></p>\n'
  ], [
    '図1',
    '<p class="caption-img"><span class="caption-img-label">図1</span></p>\n'
  ], [
    '図1.1',
    '<p class="caption-img"><span class="caption-img-label">図1.1</span></p>\n'
  ], [
    '図 猫',
    '<p class="caption-img"><span class="caption-img-label">図</span> 猫</p>\n'
  ], [
    '図1　猫',
    '<p class="caption-img"><span class="caption-img-label">図1<span class="caption-img-label-joint">　</span></span>猫</p>\n'
  ], [
    '図1.1 猫',
    '<p class="caption-img"><span class="caption-img-label">図1.1</span> 猫</p>\n'
  ], [
    '図は猫',
    '<p>図は猫</p>\n'
  ] , [
    '図1は猫',
    '<p>図1は猫</p>\n'
  ], [
    '図1.1は猫',
    '<p>図1.1は猫</p>\n'
  ]