Docs / Reference

Languages

One generated page per language, read from its pack: languages/index.md. Each page lists the quote marks and nesting, the sentence dash, orphan letters, short words, abbreviations, % and currency spacing, number format, and the sources the pack was written from.

The engine has 26 language packs, src/core/lang/<code>.ts: the 24 official languages of the EU (bg cs da de el en es et fi fr ga hr hu it lt lv mt nl pl pt ro sk sl sv), Ukrainian (uk) and Swiss usage (ch). Lang is derived from the packs, and en is the default when none is given. American English is not a pack: it is the option quoteStyle: 'us' (“…” with ‘…’ inside) on any language.

Read this first

• No pack has been read by a native speaker. Every pack has reviewed: false. The five oldest (en uk de pl fr) and ch came from general knowledge; the other twenty were written from published style guides (the EU Interinstitutional Style Guide first, then the national guide), and each page names its sources. Some of those sources could only be read in part, so a pack is a defensible style, not a certified one. Have a native reader check a language before you rely on it, then set reviewed: true.
• Which decisions have a source read, and which are still general knowledge, is recorded per rule in sources.md. Quote marks are disputed there: CLDR 46 differs from the engine for English, German, French level two and de-CH.
• The abbreviation lists are short on purpose: a list is only as long as its source. Add to a list only from a style guide.
• The short-word lists (shortWords, for the opt-in rule space.short-word) are the exception, and they say so: five packs have one (cs en pl sk uk), the words are this package’s own enumeration of two- and three-letter prepositions, conjunctions and the English article, and the rule is off by default. What the sources do back is where such binding belongs: Wolański (PWN) and the ÚJČ advice service both say a short word may stay at a line end in body text and must travel with what it belongs to in a title. Use the rule for titles and headings, and read sources.md before trusting a list.
• Dash profiles: twelve packs use a spaced en dash for a break in the sentence (cs de et fi hr hu lt lv ro sk sl sv), as the Interinstitutional Style Guide and their national guides set it, and so does English in the literary register only; every other pack uses the spaced em dash. Dutch defaults to the spaced em dash, but usage there often prefers a spaced en dash; this was not verified.
• The package’s benchmark has a few cases per language and mostly tests English. They check that the output is a defensible style, not that a native reader approves it.

Quote marks

A quotation nested deeper than the levels a pack lists starts over from the first. Source marks may be curly, low, straight or guillemets; any mix is read, and the marks of the language and level are set. The literary register uses the pack’s literary quotes where it has them (British English ‘…’ with “…” inside) and the display quotes otherwise.

quoteStyle: 'us' sets “…” with ‘…’ inside for any language and takes precedence over both the language’s marks and the literary register. The default, 'language', keeps each language’s own marks. English’s own default is «…», which is this package’s house style; the benchmark counts it as a failure against its own check (see apps/bench/RESULTS.md).

German nested quotes are »outer ›inner‹ outer«. The other common German style, „outer ‚inner‘ outer“, is not implemented.

Dialogue dash

Where a language writes a line of dialogue with a dash instead of quotation marks, a paragraph that opens with a hyphen, en dash or em dash and a space is such a line (dash.dialogue), in both registers: the dash becomes the language’s own, followed by a no-break space (—~Привіт), or attached with no space in Spanish (—¿Esperamos?). The pack field is dialogueDash; each language page shows it.

• Em dash: uk pl fr it pt bg da (Polish, Italian, Bulgarian and Danish use the dash the rest of the text uses); es attached.
• En dash: ro hr sl hu fi sv cs sk lt.
• Not set: et (dialogue is quoted), el and lv (not confirmed by a source read), and en de ch nl mt ga.

Only the start of the text (blockStart), and not when a later line of the same paragraph also opens with a dash: a dash after a line break is as likely a list marker, and a paragraph of such lines is a list. A single line that opens with - in one of these languages is read as dialogue, so a one-item list written that way gets the dash too. Sources per language are in sources.md.

Short words and one-letter words

Two things keep a short word off the end of a line, and only the first runs by default:

• space.orphan binds the one-letter words a pack lists (cs en pl sk uk), in any text. The letters are in each language page.
• space.short-word binds the two- and three-letter words a pack lists, and is off until named in enable. It never binds more than two words in a row.

Not implemented

• A literary profile for any language but English: every other pack (Ukrainian included) gives the same output in both registers. German books use „…“ with ‚…‘ inside.
• A quotation that continues across paragraphs (books reopen the quote at each paragraph start). Pairing is per paragraph: a lone straight opener is converted by where it stands, but the quotation is not followed into the next paragraph.
• A lone curly quotation mark (“, „) with no partner is left as written when its side cannot be told from where it stands.
• Per-level quote shapes that differ in kind (Latvian and Lithuanian single quotes), a third Swedish level, dotted time ranges (9.00-18.30), and abbreviations written without a full stop (Estonian nr, Finnish klo): listed in backlog.md.
• Italian and Maltese truncation apostrophes (po', va') inside straight single quotes can close the quotation early.
• A lang outside Lang is a type error in TypeScript, throws a TypeError at run time on any text, even an empty one, and is refused by the command line.

Something wrong or unclear here? Report an issue.