Add locales.MN, a Mongolian pack reading the father's name as the family (#1237) - #1365

Closes #1237. Refs #1253.

Adds locales.MN, a Mongolian pack, the first leaf of #1253's decision (its first statement: family is the name a convention files a person under).

from nameparser import parser_for, locales
sukhee = parser_for(locales.MN).parse("Батын Сүхээ")
sukhee.family, sukhee.given   # ('Батын', 'Сүхээ')

1.4.0 through 3.0.1, and the default still, read given Батын, family Сүхээ.

What the pack is

• A plain choice of mn and mn.toml's new [rules] table, name_order = "family-first-given-last", so the father's name, which a Mongolian name writes first and passports file as the surname, is the family.
• Given name last, not second, so a word written before the father's name (a clan name, Боржигин Батын Сүхээ) leaves the given name in place. The clan name itself is not read as one.
• The choice of mn changes no reading (mn.toml's lama and an exclusion are on by default); it is recorded so a strict parser reads mn.toml. Two choice tests assumed a non-empty surface, so _INERT_CHOICES exempts MN and a test holds its surface empty.
• The genitive is not read into the patronymic view (see the decision below). The glued form Ц.Сандаг-Очир (Should a Mongolian choice read a glued father's initial as the family? (Ц.Сандаг-Очир gives given 'Ц.', family 'Сандаг-Очир') #1306) reads family Ц., given Сандаг-Очир, by tokenize's full stop; not changed.
• Traditional Mongolian script reads by the declared order like any script rules/W4 does not settle, so under the pack family-first (ᠪᠠᠲᠤ ᠶᠢᠨ ᠰᠦᠬᠡ → family the joined genitive, rules/T6). The default's reading of it is decisions/T6's open question, unchanged.
• The default does not move.

Gate: measured before the code (2026-11-12, master 2b39f23)

Default against FAMILY_FIRST, right where family is the father's name:

• Cyrillic: every cached writing is father-first.
• Romanized writing splits by who writes. News romanizations are father-first, but Mongolians writing their own names in English mostly write the given name first: 65 of 87 two-word credits, which the pack misreads. Of the 150 credits, 11 write an initial first (9 glued, Should a Mongolian choice read a glued father's initial as the family? (Ц.Сандаг-Очир gives given 'Ц.', family 'Сандаг-Очир') #1306) and 13 the given name and an initial (8 spaced).
• CLDR's mn sample renderings (fictional) render every order: the per-pack view reads 19 of 24 by default and 15 of 24 under the pack, losing the given-first renderings.
• Should a Mongolian choice read the father's name first? (Ухнаагийн Хүрэлсүх gives given 'Ухнаагийн', family 'Хүрэлсүх') #1237's Wikidata label counts were not re-taken (the tool cache holds no Mongolian citizen).

Recipes, the comparators and the collisions are in decisions/vocabulary-model's 2026-11-12 #1237 entry.

Decisions for the maintainer

1. Not done: a rule reading the genitive into the patronymic view. Where does a father's name go when the name has no family name? (Abiy Ahmed Ali, Ухнаагийн Хүрэлсүх, Бекмырза Чолпонбай уулу) #1253 allows one. I recommend against it for now, for three reasons:
   • Under this pack's order the view would only repeat family. The one reading it adds alone, a romanized given-first name with the genitive last, appears in no cached source.
   • The romanized endings collide with given names written first: JRC's two-word spellings opening with -yn include Mervyn King, Carolyn Porco and Kathryn Gustafson, and its -ийн ones include the Bulgarian spellings of Madeleine, Gene and Dean.
   • A rule kind is a new PatronymicRule member and a Policy change, in a file another worker is editing.
2. Not taken: an order for Cyrillic only. The evidence argues for reading Cyrillic father-first and leaving romanized names to the declared order. Policy.script_orders could say that if Script had a Cyrillic member, but it has none (_policy.py again), and rules/W4's licence is a script that settles the order, which Cyrillic, shared with Russian, does not. The pack ships the order Should a Mongolian choice read the father's name first? (Ухнаагийн Хүрэлсүх gives given 'Ухнаагийн', family 'Хүрэлсүх') #1237's option 1 and Where does a father's name go when the name has no family name? (Abiy Ahmed Ali, Ухнаагийн Хүрэлсүх, Бекмырза Чолпонбай уулу) #1253's decision name.
3. Accepted costs (each pinned by a test):
   • A romanized name written given-first reads backwards (Suhee Bataagiin → family Suhee).
   • So does the given name followed by the father's initial (Uyanga B. → family Uyanga), which the default reads right and reports as an order ambiguity: 8 spaced among ORCID's 150, against 2 with the initial first.
   • Russian names in a Mongolian list (raised in review): the given-last order reads Сидоров Иван Петрович given Петрович, where FAMILY_FIRST reads given Иван; and under parser_for(locales.MN, locales.RU) the patronymic lands in given, where R9's East Slavic rule (which reads a middle word) does not see it, so no patronymic is read. Neither the clan-name shape the order was chosen for nor this one is measured; your call whether FAMILY_FIRST is the better default.
   • Traditional script written with a plain space before the genitive suffix (ᠪᠠᠲᠤ ᠶᠢᠨ ᠰᠦᠬᠡ) reads the suffix as a middle name, as the default does; with U+202F it joins (rules/T6).
   • The CLDR sample's own title Ноён (Mr.) is no listed title: a vocabulary candidate, not filed.

Sweeps under the pack

• Comma shapes: a family comma keeps the default's reading (Батын, Сүхээ, Сүхээ, Батын, , Сүхээ).
• Titles and suffixes: Др. Батын Сүхээ, Dr. Bataagiin Suhee, Батын Сүхээ Jr. and Bataagiin Suhee, PhD read the father's name as the family.
• One word: a lone word reads given; Policy.lone_name_word=Role.FAMILY on the base reads it family.
• Word counts and spellings: three and four words, a hyphenated given name, one case, NFD (Дорждэрэмийн, Khürelsükh).
• Parser.revise keeps the family. No view fills.
• Pack unions:
  • parser_for(locales.MN, locales.RU) declares the order, so Сидоров Иван Петрович reads given Петрович (O1's scope: a declared family-first order stands).
  • MN+TR_AZ and MN+VI read by the order too.
  • strict=True reads the same names alike; Lama stays a title.
• Pickling: the pack and a parser round-trip. Under the 3.0.1 wheel both are refused, but for 3.1's own new fields (PolicyPatch.initials_as_family, a military reading's same_as), as every 3.1 pickle is. A hand-built Locale of the same data parses alike there.
• Matching:
  • Батын Сүхээ / Сүхээ Батын conflicts under the pack as it does by default; across-roles does not reach a two-word swap.
  • Б. Сүхээ is consistent with the full name only with family-initial accepted.
  • Батын / Бат (genitive against nominative) conflict: a candidate, not filed.
• Default:
  • tools/differential/snapshot.py → no reading moves; the snapshot gains one row, O4's new example read by default.
  • snapshot.py --local <each> --base <master 2b39f23d checkout> → matches the base for all 16 local corpora (ctgov-officials, cz-candidates, dof-signatures, fec-individuals, govuk-people, jrc-names, nepal-candidates, opensanctions-parliaments (re-run on the refreshed cache), orcid-researchers, patentsview-inventors, psc-individuals, tse-donors, tw-donors, uit-vinames, ukec-individuals, vn-assembly).

Also in this PR

• tools/differential/cldr_agreement_locales.json gains MN's record, and nothing else moved in the re-record.
• docs/contributing.rst and decisions/deviates-registry record the inert choice. docs/policy.rst still names only vi as setting FAMILY_FIRST_GIVEN_LAST; that file belongs to another in-flight PR, so it is left alone.
• Three feat(#1237) case rows. Each was checked against the 1.4.0 and 3.0.1 wheels, and each fails on master (no locales.MN).
• rules/O4 gains an [mn] example.
• With the maintainer's approval, the shared ~/.cache/nameparser-tools/local/opensanctions-parliaments was refreshed: it held 17 datasets, which the tools now refuse, and now holds 23 (24,066 names).

Checks

• uv run pytest with the ja extra: 46,040 passed. Without it: 46,008 passed. Python 3.15 in its own environment: all passed apart from the corpus test the rules-corpus rebuild then fixed.
• uv run pytest -n0 -m serial: 35 passed.
• uv run mypy, uv run ruff check, tools/vocab/build.py --check: clean.
• sphinx-build -b doctest docs, sphinx-build -W -b html docs, python -m doctest README.rst: clean.

Reviews

Each review ran as a subagent in its own detached worktree on the pushed head (b4aee8e), with a read-only master checkout beside it.

🤖 Generated with Claude Code