Skip to content

Messages & plurals

Format ICU MessageFormat strings — placeholders, numbers, and especially pluralisation and gender/word selection — without hardcoding any language's grammar rules. A single pattern adapts: English needs one/other, Arabic needs six categories, and message() picks the right branch from CLDR for you.

The pattern language

A pattern is plain text with {…} placeholders. There are four placeholder shapes:

Shape Example Renders
Simple Hello, {name}! substitutes the argument as-is
Number {n, number} · {n, number, integer} · {n, number, percent} locale-formatted number
Plural {n, plural, one {# item} other {# items}} picks a branch by plural category; # is the value
Select {g, select, female {her} male {his} other {their}} picks a branch by exact string match

Inside a plural/selectordinal branch, # is replaced by the (locale-formatted) number. Two refinements:

  • Exact matches — =0 {none} one {# item} other {# items} uses the =0 branch when the value is exactly 0, before falling back to the category.
  • offset:n — {n, plural, offset:1 =0 {nobody} one {you and # other} other {you and # others}} subtracts n from # (for "you and 2 others" style phrasing).

An other branch is required in every plural/selectordinal/select — it's the fallback when no category matches. Escape a literal brace or # by wrapping it in single quotes: '{', '#', and '' for a literal apostrophe.

Messages

c = Cosmo("en")
c.message("{0, plural, one {# file} other {# files}}", [3])
# "3 files"
c.message("{0,number,integer} of {1,number}", [4560, 0.2])
# "4,560 of 0.2"
c.message("{g, select, female {She} male {He} other {They}} replied", {"g": "female"})
# "She replied"

Named vs positional arguments

Pass positional arguments in a list ({0}), or named arguments in a dictionary ({name}). ICU MessageFormat supports number, date, time, and spellout arguments, as well as plural and selection branches.

Ordinals in messages

selectordinal works like plural but uses ordinal categories — the "1st / 2nd / 3rd" wording rules:

c = Cosmo("en")
p = "{n, selectordinal, one {#st} two {#nd} few {#rd} other {#th}} place"
c.message(p, {"n": 1})   # "1st place"
c.message(p, {"n": 22})  # "22nd place"

For ordinal text such as "22nd" without writing branches, use ordinal().

Plural category

Sometimes you want the raw decision rather than a formatted message — e.g. to pick an image or a CSS class. plural_category() returns the LDML category a number falls into for the locale, the same choice message() makes internally.

en = Cosmo("en")
en.plural_category(1)           # "one"
en.plural_category(2)           # "other"
en.plural_category(2, True)     # "two"
Cosmo("ar").plural_category(0)  # "zero"

Returns one of zero, one, two, few, many, other. Pass ordinal = true for ordinal rules instead of cardinal.

Which categories does a locale use?

The set varies wildly: English has one/other (cardinal), Arabic has all six, Japanese has only other. Write every plural branch you reference, but always keep other — it covers any category you didn't spell out.

Practical examples

A notification badge sentence. One pattern handles zero, singular, and plural without branching in your code:

c = Cosmo("en")
p = "{n, plural, =0 {No new messages} one {# new message} other {# new messages}}"
c.message(p, {"n": 0})   # "No new messages"
c.message(p, {"n": 5})   # "5 new messages"

"You and N others" with offset. The offset keeps the displayed # one less than the real count:

c = Cosmo("en")
p = ("{n, plural, offset:1 =1 {You liked this} one {You and # other liked this}"
     " other {You and # others liked this}}")
c.message(p, {"n": 1})   # "You liked this"
c.message(p, {"n": 3})   # "You and 2 others liked this"