Repository navigation
Description
The question. A strict choice reads only the chosen languages' files. Where a word's vocabulary reading is in a chosen file and the name reading that gates it is in another, strict keeps the reading and drops the gate:
>>> ar = Locale(code="ar", lexicon=Lexicon.empty(), languages=("ar",))
>>> parser_for(ar).parse("Cheikh Mansour")
given='Cheikh', family='Mansour'
>>> parser_for(ar, strict=True).parse("Cheikh Mansour")
title='Cheikh', given='Mansour'cheikh's given-name reading is in fr.toml. Sultan Mahmud under strict ms reads title Sultan, family Mahmud the same way, its name readings being in ar.toml and tr.toml. The vocabulary holds 20 such words in 35 word-and-language pairs (STRICT_UNGATED in tests/v2/test_reading_index.py). The RFC left each to be read and, where the language's names carry the word, given a name reading in that language's file.
Measured 2026-11-18 on master 3175314d. Strict against strict with the word's given reading restored (lexicon.add(word, kind="given")), on the ORCID records of the language's countries that hold the word, scored against the filed boxes. That isolates the gate from everything else strict does.
| Strict choice | Country sample | Records holding a listed word | Read differently without the gate |
|---|---|---|---|
ar |
Tunisia, Morocco, Algeria, Egypt | 1,302 | 70 |
ms |
Malaysia | 360 | 3 |
it |
Italy | 123 | 3 |
fil |
Philippines | 105 | 2 |
pt, pt-BR |
Portugal, Brazil | 2,535 | 1 |
nl, fr, ca, ga |
Netherlands, France, Spain, Ireland | 822 | 0 |
The gate acts only where the word opens the name, so da stands in 1,606 Brazilian names and moves none.
62 of the 79 are ben opening a Maghrebi name, where the ungated reading is the right one (Ben Ahmed is the surname; the boxes agree with the gate only where the surname was typed into the given-names box). Two words lose names: cheikh under ar (8 records) and sultan under ms (1).
Proposed.
- Add a
givenreading forcheikhto ar.toml and forsultanto ms.toml, and take both offSTRICT_UNGATED. - Record the other pairs measured as wanted: under strict, the word is the particle.
Open:
gd,hr,sk,slandhiwere not measured, normarquis,marquise,tho,standsushri, which the samples hold too rarely to say anything.- A Chinese given name
Binopening a name in a Malaysian list reads as the family's particle under strictms. Accept, as strict's single-origin premise? - Should
cheikhandsultanbe read from the counts (tools/name_usage.py) before the readings are added?
Source: docs/design/vocabulary-model-rfc.md, "The derived markers under strict".
Activity