Some catalogues write "one" in a plural form that also counts twenty-one #250

New issue

Closed

opened 2026-09-16 17:25:27 +00:00 by tiagoagueda · 1 comment

tiagoagueda commented

2026-09-16 17:25:27 +00:00

Owner

The number written out in a form that covers more than one

A Slavic plural rule puts 1, 21, 31, 41 … in the same form. Several catalogues translate that
form with the word one rather than with the placeholder, so at twenty-one applications the
page says "one settled application".

bs, hr, sr, mk, uk: jedna / Една / Једна / Одна in form 0.
sl: ena in form 0, whose rule (n%100==1) also catches 101 and 201.
is: Ein in form 0.

lv, lt, hy and ga get this right already — they carry %(counter)s in that form,
which is what the others need. An English singular with no placeholder is not the cause: a
form may carry fewer placeholders than the source, and these languages need this one.

Slovene uses the plural where it has a dual

sl declares four forms and form 1 is the dual (n%100==2). At least one entry writes the
plural jih there where Slovene wants the dual accusative ju. The strings filled during
the 0.3.0 sweep do use real dual morphology, so this is about what was already in the file.

Irish has five forms doing the work of one

ga declares five, and in several entries forms 0, 1, 2 and 4 are textually identical — so
the table costs five translations and answers like one. Form 1 (n==2) in particular should
read dhá ….

Worth doing at the same time

lv cannot decide what a board is: tāfele in the navigation description, dēlis on
the button. A Latvian reader presses "Skatīt dēli" and then reads about a tāfele.
cs has two words for an application, žádost (canonical, and what the UI strings use)
and přihláška in a handful of statistics strings.
hu mixes registers: roughly seven informal te markers against sixteen formal Ön
forms. es, it and ro have the same split, formal on the Europass import page and
informal everywhere else.
pt-pt mixes them inside one entry: vê-la is the informal imperative where the rest
of that catalogue uses veja-a.
mt's %(counter)s day has ijiem in the form covering 11–19, where Maltese takes the
singular (ħdax-il jum).

How to approach it

These need a speaker, not a sweep: each is a judgement about an existing translation rather
than a gap to fill. A pass per language, with the Plural-Forms expression in front of you
and a list of every entry whose form 0 contains a written-out number, would find the first
group mechanically. Nothing here blocks anything; it is the kind of thing that makes a
catalogue trustworthy rather than merely complete.

Found by the translators who swept the catalogues for 0.3.0. None of it is new to that sweep — these are entries that were already there — and none of it is visible until somebody has twenty-one of something, which is why nobody has seen it. ## The number written out in a form that covers more than one A Slavic plural rule puts 1, 21, 31, 41 … in the same form. Several catalogues translate that form with the word *one* rather than with the placeholder, so at twenty-one applications the page says "one settled application". - `bs`, `hr`, `sr`, `mk`, `uk`: *jedna* / *Една* / *Једна* / *Одна* in form 0. - `sl`: *ena* in form 0, whose rule (`n%100==1`) also catches 101 and 201. - `is`: *Ein* in form 0. `lv`, `lt`, `hy` and `ga` get this right already — they carry `%(counter)s` in that form, which is what the others need. An English singular with no placeholder is not the cause: a form may carry fewer placeholders than the source, and these languages need this one. ## Slovene uses the plural where it has a dual `sl` declares four forms and form 1 is the dual (`n%100==2`). At least one entry writes the plural *jih* there where Slovene wants the dual accusative *ju*. The strings filled during the 0.3.0 sweep do use real dual morphology, so this is about what was already in the file. ## Irish has five forms doing the work of one `ga` declares five, and in several entries forms 0, 1, 2 and 4 are textually identical — so the table costs five translations and answers like one. Form 1 (`n==2`) in particular should read *dhá …*. ## Worth doing at the same time - **`lv` cannot decide what a board is**: *tāfele* in the navigation description, *dēlis* on the button. A Latvian reader presses "Skatīt dēli" and then reads about a *tāfele*. - **`cs` has two words for an application**, *žádost* (canonical, and what the UI strings use) and *přihláška* in a handful of statistics strings. - **`hu` mixes registers**: roughly seven informal *te* markers against sixteen formal *Ön* forms. `es`, `it` and `ro` have the same split, formal on the Europass import page and informal everywhere else. - **`pt-pt` mixes them inside one entry**: `vê-la` is the informal imperative where the rest of that catalogue uses `veja-a`. - **`mt`'s `%(counter)s day`** has *ijiem* in the form covering 11–19, where Maltese takes the singular (*ħdax-il jum*). ## How to approach it These need a speaker, not a sweep: each is a judgement about an existing translation rather than a gap to fill. A pass per language, with the `Plural-Forms` expression in front of you and a list of every entry whose form 0 contains a written-out number, would find the first group mechanically. Nothing here blocks anything; it is the kind of thing that makes a catalogue trustworthy rather than merely complete.

tiagoagueda added this to the 0.4.0 milestone

2026-09-16 17:25:27 +00:00

tiagoagueda added the

bug

internationalisation

labels

2026-09-16 17:25:28 +00:00

tiagoagueda referenced this issue from a commit

2026-09-18 08:58:54 +00:00

Stop nine catalogues saying "one" at twenty-one

tiagoagueda closed this issue

2026-09-18 08:58:54 +00:00

tiagoagueda commented

2026-09-18 08:59:40 +00:00

Author

Owner

Landed on main as 3cfff9291. Every changed entry keeps its draft flag; a speaker still gets the last word. What the pass found beyond the list above, and what it left alone on purpose:

Wider than listed. lt and lv had the same thirteen (eleven) one entries as the Slavic catalogues — only some of their entries carried %(counter)s. The check now evaluates each language's plural rule and refuses a form that drops the count where the rule also selects two or more, so this fails on push from here on (messages_tool.counts_beyond_one).

Maltese 11–19 takes %(counter)s-il jum, not the bare %(counter)s jum suggested above: the -il attaches to a numeral written in figures (the catalogue already has Kull 15-il minuta). 35 entries changed, most only to add the -il; plural nouns in forms 2 and 3 became singular with plural predicates (20 applikazzjoni ntbagħtu).

Hungarian was the other way round. Counting whole entries rather than ^msgstr lines: about 239 informal against about 65 formal, not seven against sixteen. It was unified to the formal register all the same — the Hungarian software convention, and the direction this issue implied — which made it a 239-entry rewrite. If informal was wanted, reversing the ~65 formal entries is the smaller job; say so and it can be redone.

Slovene: four duals fixed (poglejte ju, sta se morda zgodili, spodnji nastavitvi sta shranjeni, sporočili … nista uspeli) and the 3–4 and 5+ forms of two entries that had the wrong pattern; entry %(counter)s earlier moved to masculine to agree with razgovor. Irish: six entries; 34 were already right under the digits convention (lenition after 2–6, eclipsis after 7–10, base noun at 11+). bs/hr/sr: beyond the thirteen, paucal and 5+ agreement fixed in a handful of forms (Svih → Sve at 2–4, osobe → osoba at 5+). Czech: 20 entries to žádost, including two the grep missed (přihlášek).

Left alone, worth knowing:

ro has a formal third outside the Europass page — server, mail and plugin settings, the letter templates, several Choose … first messages (Alegeți beside Alege). es has smaller usted pockets in the same places (Cópielo ahora beside Cópialo ahora). Both are register splits of the kind this issue is about, and neither was in scope here.
cs keeps Uzávěrka přihlášek for the listing's closing date: the idiom, not the noun for an application.
hu has one lowercase ön and uses both irányítópult and vezérlőpult for the dashboard.
lv entry A plugin has suggested… form 2 (zero only) says 0 lietas where strict Latvian says 0 lietu; the banner never renders at zero.
ga has one form for 11 upwards, which cannot serve both 11–19 (na) and the round tens (an) with the article.
pt-pt vê-la → veja-a was the only Portuguese change.

Landed on `main` as `3cfff9291`. Every changed entry keeps its `draft` flag; a speaker still gets the last word. What the pass found beyond the list above, and what it left alone on purpose: **Wider than listed.** `lt` and `lv` had the same thirteen (eleven) *one* entries as the Slavic catalogues — only some of their entries carried `%(counter)s`. The check now evaluates each language's plural rule and refuses a form that drops the count where the rule also selects two or more, so this fails on push from here on (`messages_tool.counts_beyond_one`). **Maltese 11–19** takes `%(counter)s-il jum`, not the bare `%(counter)s jum` suggested above: the `-il` attaches to a numeral written in figures (the catalogue already has `Kull 15-il minuta`). 35 entries changed, most only to add the `-il`; plural nouns in forms 2 and 3 became singular with plural predicates (`20 applikazzjoni ntbagħtu`). **Hungarian was the other way round.** Counting whole entries rather than `^msgstr` lines: about 239 informal against about 65 formal, not seven against sixteen. It was unified to the formal register all the same — the Hungarian software convention, and the direction this issue implied — which made it a 239-entry rewrite. If informal was wanted, reversing the ~65 formal entries is the smaller job; say so and it can be redone. **Slovene**: four duals fixed (`poglejte ju`, `sta se morda zgodili`, `spodnji nastavitvi sta shranjeni`, `sporočili … nista uspeli`) and the 3–4 and 5+ forms of two entries that had the wrong pattern; entry `%(counter)s earlier` moved to masculine to agree with *razgovor*. **Irish**: six entries; 34 were already right under the digits convention (lenition after 2–6, eclipsis after 7–10, base noun at 11+). **bs/hr/sr**: beyond the thirteen, paucal and 5+ agreement fixed in a handful of forms (`Svih` → `Sve` at 2–4, `osobe` → `osoba` at 5+). **Czech**: 20 entries to *žádost*, including two the grep missed (*přihlášek*). **Left alone, worth knowing:** - `ro` has a formal third outside the Europass page — server, mail and plugin settings, the letter templates, several *Choose … first* messages (`Alegeți` beside `Alege`). `es` has smaller *usted* pockets in the same places (`Cópielo ahora` beside `Cópialo ahora`). Both are register splits of the kind this issue is about, and neither was in scope here. - `cs` keeps *Uzávěrka přihlášek* for the listing's closing date: the idiom, not the noun for an application. - `hu` has one lowercase *ön* and uses both *irányítópult* and *vezérlőpult* for the dashboard. - `lv` entry *A plugin has suggested…* form 2 (zero only) says `0 lietas` where strict Latvian says `0 lietu`; the banner never renders at zero. - `ga` has one form for 11 upwards, which cannot serve both 11–19 (*na*) and the round tens (*an*) with the article. - `pt-pt` *vê-la* → *veja-a* was the only Portuguese change.