Vertical Typography & Punctuation Rules

The rules that govern how CJK text is set vertically (top→bottom) and right-to-left in Vertically Works. This is the canonical, implementation-neutral reference: the rules follow the authorities cited at the bottom (KLREQ / JLREQ / UAX #50), not any one codebase, and apply equally to a CSS writing-mode: vertical-rl renderer and to the Swift text engine in Vertically Verse (the reference implementation these rules were extracted from — see Provenance).

Vertical CJK setting is not rotated horizontal text — each glyph kind needs its own orientation, punctuation is repositioned or swapped for a vertical form, and line breaking obeys 금칙/kinsoku (禁則) rules.

The identifier names in the tables below (kind(of:), verticalizePunctuation, PunctuationCell, composeColumns, groupDigits, spaceWeight, …) label the functions in the reference implementation. Treat them as names for each rule, not as an API you must reproduce — a CSS/JS renderer expresses the same rules differently.

---

1. Glyph kinds

Every grapheme cluster is tokenized into a cell tagged with a GlyphKind (kind(of:)):

Classification order (kind(of scalar:)): space → digit (0x30–0x39) → latin (A–Z/a–z) → punctuation (if in the punctuation sets) → otherwise cjk.

---

2. Latin & word grouping

• Consecutive Latin letters are grouped into a single latin cell (a word) and rotated 90° CW as a unit, so "King" reads as one rotated token rather than four stacked letters.
• The rotated word reserves an approximate footprint (size * 0.60 * count) so it doesn't overlap its neighbors.

---

3. Digits

• Body text (reader): one upright cell per digit — the KLREQ rule for vertical numerals. Multi-digit numbers stack top→bottom.
• Chrome labels (groupDigits: true): a short run of consecutive digits (≤ 3, e.g. a chapter number) folds into one tate-chu-yoko cell (縦中横) — the digits sit upright, side-by-side, on a single line within the column. Runs of 4+ fall back to one-per-cell.
• A digit run is unbreakable: column composition never splits it across two columns (§5).

---

4. Punctuation

Two stages: first the source text is normalized (§4.1), then each punctuation cell is rendered with the correct orientation (§4.2).

4.1 Verticalization of the source (verticalizePunctuation)

Traditional 세로쓰기 substitutions applied before tokenizing:

Decimal points and thousands separators between two digits are deliberately left intact.

4.2 Orientation / rendering (PunctuationCell)

The punctuation sets that drive classification:

• verticalPunctuation — CJK marks: 。 、 ， ． ： ； ！ ？ 「 」 『 』 （ ） ［ ］ 【 】 〈 〉 《 》 ・ ー ― … 〜 ～
• asciiPunctuation — halfwidth marks that occur in Korean text: . , ! ? : ; ( ) [ ]

---

5. Line composition — 금칙 / kinsoku (禁則)

composeColumns(_:perColumn:cellOf:keepFromColumnEnd:spaceWeight:) packs cells into columns of a target height, applying the forbidden-position rules so a mark never lands illegally at a column edge.

5.1 行頭금칙 — may not begin a column (KLREQ §4.3.3)

Closing brackets, stops, commas, sentence-final marks, middle dot, ellipsis, prolonged/wave marks, and the dash (JLREQ treats ― as inseparable — keeping it off a column head keeps a ―― pair together):

If a column would start with one of these, it is pulled up onto the previous column (small overflow allowed).

5.2 行末금칙 — may not end a column (KLREQ §4.3.4)

Opening brackets:

If a column would end with one of these (or an orphaned verse-number marker passed via keepFromColumnEnd), it is pushed down to the next column.

5.3 Other constraints

• Digit runs are not split across columns — a run straddling the boundary is pushed down.
• No leading space: a column never begins with a .space cell.
• spaceWeight: a .space cell renders shorter than a glyph box, so it counts as less than a full slot toward the perColumn height budget. Passing the space cell's real height ratio lets Korean 띄어쓰기-heavy columns pack enough glyphs to fill a tier uniformly with the space-less JP/CN. Default 1 = pure cell count (unchanged for callers without word spaces). See the algorithm sketch in §5.4.

5.4 Algorithm sketch

1. Skip a leading space; greedily fill the column until the accumulated (weighted) height reaches perColumn.
2. Pull up any 行頭금칙 marks that fell at the next column's head.
3. Push a split digit run down.
4. Push down a trailing 行末금칙 opener / kept cell.
5. Emit the column; repeat.

---

Provenance

These rules were extracted from the Vertically Verse vertical text engine (Swift), which remains the reference implementation. The concern → source map there:

On the web, the same rules ride on writing-mode: vertical-rl + text-orientation: mixed. The Vertically Works design tokens and vertical components implement the interface layer:

---

References

• KLREQ — Requirements for Hangul Text Layout and Typography (W3C). §4.3.3 line-start, §4.3.4 line-end forbidden rules.
• JLREQ — Requirements for Japanese Text Layout (W3C) — kinsoku, tate-chu-yoko.
• UAX #50 — Unicode Vertical Text Layout — per-character upright/rotated orientation.
• CSS Writing Modes Level 4 — writing-mode: vertical-rl, text-orientation.
• CSS Text Level 4 — tate-chu-yoko (縦中横).