CJK and Latin disagree on baseline and face height
Aliases: ideographic em-box · CJK vertical metrics · x-height mismatch
What it is
Write 「高度 12 mm」 on one line and the square of 「高」 fills the middle of the em, while Latin mm sits on the baseline, belly on that line, only an x-height above it. The two scripts do not share a visual centre: Han looks high and full, Latin low and short. The sizes are not wrong. Two vertical metrics have been forced into one line box.
The gap between scripts is a left–right collision. This is who lines up with whom, vertically. Mixed lines that look as if “the English has sunk” or “the Chinese is top-heavy” are usually this, not a weight problem.
Why it happens
Latin metrics take the baseline as origin; x-height, ascender, descender are all relative to that line. Han metrics take the em square as a box; the face (ideographic em-box) sits centred or slightly high inside it, with no “foot line” that coincides with Latin. The line box packs the larger of the two ascents and descents, so the Latin baseline is pinned somewhere in that box and the Han face is placed by its own em — by default the x-height band falls in the lower half of the Han face, and the English looks sunken.
Vertical numbers in the font files also disagree. Some Chinese cuts inflate winAscent, the line box grows, and Latin looks even shorter in the row. Some ship two sets, hhea and OS/2, and different systems pick different sets, so the relative position of the mix shifts again between macOS and Windows. Families actually designed for mixed setting raise Latin x-height toward the middle of the Han face, or offer a baseline tweak; splicing a Latin face and a Chinese face that were each “nice alone” almost always misses.
Figures are harsher. Oldstyle digits have descenders and some of them fall below the Han face; lining digits sit higher but still on the baseline, and miss the Han head. Units and values are where mixed-script baseline conflict shows first.
Studying it
Measure relative position, not taste. In one line box mark: Latin baseline, x-height line, top and bottom of the Han face. Independents: which Chinese and Latin faces are paired, whether they share a family, which engine. Dependents: pixel gap between the mid-x-height and the mid-Han-face, rated “which side has sunk.”
Compare “two faces spliced” against “a family cut for both scripts” and see whether the gap shrinks. Do not inspect only a one-line title on the artboard — body copy will show extenders, and the conflict is complete only in specimens that include g p y and 「国 量 高」.
Where it stops holding
An all-Chinese or all-English line has no second metric to fight. A short Latin label next to an icon, vertically centred as a graphic, is object alignment, not a type baseline; once that label enters a sentence, the baseline problem returns. Vertical Chinese uses another system (upright Latin, or turned runs); a pairing tuned for horizontal setting does not move there for free. Huge display type can optically align the two scripts as two graphics and ignore the body baseline; back at reading sizes, optical alignment fights the line box, and you are on font metrics again.
Applying it
- For mixed setting, prefer a family whose Latin and Han share one vertical metric, rather than splicing two faces that are each handsome alone.
- Inspect lines that contain figures and units on their own: lining figures usually sit closer to the Han face than oldstyle; a half-step of size on the unit is a last resort after changing the pairing.
- Check the same line on macOS and Windows (or across browsers). Different choices of vertical tables can make the “sinking” appear on only one end.
- Draw two lines: the mid-line of Latin x-height, and the mid-line of the Han face. If they stay apart by more than a stroke width, that pairing is not usable for mixed setting.