Should an ALL-CAPS word in a mixed-case name mark the family name? DUPONT Jean gives given DUPONT

French administrative data, and many African and Francophone conventions, write the surname in capitals: Jean DUPONT, DUPONT Jean, Abɖu-Tɔyib GOMINA MUSA (CLDR blo). nameparser reads by position, so a surname-first listing reverses:

Jean DUPONT                     → given Jean, family DUPONT          (right, by position)
DUPONT Jean                     → given 'DUPONT', family 'Jean'
Abɖu-Tɔyib GOMINA MUSA          → middle 'GOMINA', family 'MUSA'     (CLDR: family GOMINA, given incl. MUSA)

The question: is case a signal the parser should read? It's written evidence, not language detection, so it fits the positional doctrine better than most culture-specific rules.

What makes it hard:

A name written wholly in capitals carries no signal, so the rule only applies to mixed-case input.
All-caps words are already claimed by credentials (John Smith MBA, and the CapsSuffixes policy) and by acronym titles.
The capitals can mark a multi-word surname (DE LA FONTAINE).
It would most naturally be an opt-in Policy field, not a default.

Another writing tradition with the same convention: Japanese names written in Latin script often put the surname in capitals, usually surname first: YAMAMOTO Yukio, SATŌ Akari (examples from The Cultural Atlas, Japanese naming). Measured on master 1a4e8a1:

YAMAMOTO Yukio   → given 'YAMAMOTO', family 'Yukio'
SATŌ Akari       → given 'SATŌ', family 'Akari'
Yukio YAMAMOTO   → given 'Yukio', family 'YAMAMOTO'   (right, by position)

This is the surname-first listing that the French examples above also reverse, here from an unrelated tradition. The convention spans languages, so a rule here would be reading how the name is written, not guessing its language. The macron in SATŌ is an uppercase letter, so a test like _vocab.written_as_a_name's (last cased letter) sees it correctly.

Measured 2026-11-07 on France's register of elected officials, mayors and departmental councillors: 36,931 distinct names, family name in capitals as the register writes it, scored against the register's own fields. The given-names field holds every name on the birth record, so the first counts as the given name and the rest as middle names.

Composed as	Right
`Prénom NOM` (default order)	36,458 (98.7%)
`NOM Prénom` under `FAMILY_FIRST`	36,010 (97.5%)

Nearly every miss is a family name of two or more words that the capitals mark and position doesn't:

Given first (455 of the 473 misses):
- 355 two-word family names without a particle (Lucie MARTIN ROUX gives middle 'MARTIN', family 'ROUX')
- 54 with a particle inside (Lucie FABRE DE LA MOTTE gives middle 'FABRE')
- 34 starting with one (EL, BEN, DE X DE Y)
- 12 with a hyphen ending in a particle
Family first (914 of 921):
- 434 starting with a particle: LE GRAND Hugo gives family 'LE', given 'GRAND', because the capitals take away the lowercase that rules/P11 (a lowercase particle under a family-first order) reads
- 359 without a particle
- 54 with a particle inside
- 55 family names that are also titles at the head: DOYEN Hugo gives title 'DOYEN', family 'Hugo'. The titles are BARON ×10, DOYEN ×8, PROVOST ×5, HERZOG, SERVANT, CANON, PASTOR, RITTER, FRAU and others.
- 12 with a hyphen ending in a particle

A collision to settle with rule S6 (the case signal at the suffix slot). S6 reads a credential-or-name word written in capitals, in a mixed-case name, as the credential, even with no word to spare. This register writes every family name that way, so 11 names lose their family name: Hugo LAC gives suffix 'LAC', family '' (Hugo Lac gives family 'Lac'). The words are CARO ×4, LAC ×2, CAP ×2, CAPS, PLA and MOT. A rule reading capitals as the family name would have to decide which reading wins when the capitals are the only name word after the given name.

A third population: Hong Kong. The Legislative Council writes every member's family name in capitals: Hon CHAN Kin-por, GBS, JP, Hon Jeffrey LAM Kin-fung, GBS, JP. Measured 2026-11-08 on LegCo's precedence lists for 1998–2020 (legco.gov.hk, 236 distinct names), with the word in capitals as the family:

Read as	family right
default order	19 of 236
`FAMILY_FIRST`, `two_given_names=True`	107 of 236

All 236 have the family in capitals, and the capitals are the only thing that tells the shapes apart:

CHAN Kin-por: family first (110 names). The family-first setting reads all but 7 of these; the 7 are Should Hong Kong's honours read as post-nominals? (Hon LAU Kwok-fan, MH, JP gives given MH, family LAU Kwok-fan) #1182's post-nominals.
Jeffrey LAM Kin-fung: an English name, then the Chinese name family first (101 names). Wrong under every order; Where should the Chinese given name go when an English name comes first? (Jeffrey LAM Kin-fung gives middle LAM, family Kin-fung) #1181 asks what it should read as.
Margaret NG: given first (the default reads these).
Mrs Rita FAN HSU Lai-tai: a married name, the husband's family then her own, both in capitals.

So for Hong Kong the capitals settle every name in the list, including the shape no name_order can read. Two things the French examples don't show: the capitals word can stand in the middle of the name, and the words around it are in two different orders.

Turkey's parliament writes the surname in capitals too: Mehmet DEMİR, Saliha Sera KADIGİL. Measured 2026-11-08 on the Grand National Assembly's members (OpenSanctions tr_parliament, 592 names), taking the words in capitals as the family, on master ccaf6ad: the default reads 550 right.

Most of the misses are a shape the French data shows less often, two surnames in capitals, nearly all women with a birth surname and a married one: 40 names (Şebnem BURSALI AKSOY gives middle BURSALI, family AKSOY). The capitals mark exactly where the family starts.

Two more lose the family to a credential reading, as CARO and LAC did in the French register: ALA and ALP are listed credentials, and a listed acronym in capitals ending a mixed-case name reads as the suffix (S6, the case signal at the suffix slot).

Input	given	middle	family	suffix
`Efkan ALA`	`Efkan`	`''`	`''`	`ALA`
`İnan Akgün ALP`	`İnan`	`''`	`Akgün`	`ALP`

Both are borne as surnames (ala: US Census 743, Wikidata 14 family; alp: Census 116, Wikidata 55), and in a source that writes surnames in capitals, the capitals say surname, not credential. So if case becomes a signal here, it has to outrank S6 in that source.

Decided 2026-11-11. An opt-in Policy switch, not a default; implementation follows the #1254 PRs, since both touch the suggester's signals and the same stage.

Opt-in. Measured that day on ORCID holders' own boxes (orcid-researchers/raw.jsonl, mixed-case records with exactly one box written wholly in capitals words of two or more letters): the family box is the one in capitals in 41,078 records against 4,500 for the given box where the words have four or more letters (90.1%), and 5,122 against 747 for two or three letters (87.3%); 95,412 records are wholly in capitals and carry no signal. Nine in ten is a prior for a caller who knows the source, not a default: in sources that write credentials instead, an unlisted capitals word is a post-nominal (all 15 such names in govuk-people).
Under the switch, the run of capitals words in a mixed-case name is the family, wherever it stands: DUPONT Jean, Jean DUPONT, and a run in the middle (Jeffrey LAM Kin-fung → family LAM). Particles written in capitals inside the run belong to it (LE GRAND Hugo, FABRE DE LA MOTTE); two surnames stay one family (BURSALI AKSOY).
Capitals outrank the credential and title readings there (Hugo LAC → family LAC; DOYEN Hugo → family DOYEN). A name wholly in capitals is unaffected; dotted or one-letter initials never count as the run; a credential after a comma stays a credential (Hon CHAN Kin-por, GBS, JP → family CHAN, suffix GBS, JP).
A listed credential or generation still peels off the end of the run when a capitals word is left to be the family (GRAJALES III → family GRAJALES, suffix III; Jean DUPONT MBA keeps suffix MBA); a lone capitals word is always the family (Yanjun MA). Accepted cost: a surname whose last word spells a credential (MBALLA MBA), 19 of the 46,200 ORCID records with the family box in capitals.
The other words keep their written order: the first is the given name, the rest middle names (Jeffrey LAM Kin-fung → given Jeffrey, middle Kin-fung, which answers Where should the Chinese given name go when an English name comes first? (Jeffrey LAM Kin-fung gives middle LAM, family Kin-fung) #1181 under the switch).
With unlisted_caps_suffixes: the switch composes with AFTER_COMMA (the default) and OFF, which read the other side of the comma; setting it together with EVERYWHERE raises at construction, the two being opposite statements about an unlisted capitals word with no comma.
The suggester turns the switch on for a list that writes its families in capitals consistently, where today capitals only cast an order vote. Capitals words of two or three letters count as the family under the switch (the caller has declared the convention) and stay out of the suggester's evidence, as today. No locale pack sets the switch.

🤖 Generated with Claude Code

Fix rules/O9's review findings: reports, the given name, conjunctions…

Fix the review of rules/O9's fix commit: mixed case, reports, joining…

Fix the review of 446d992: order-free reports, P6's marks, the peel (#…

Amended 2026-11-12 after measurement. Before any code, the decision above was measured against ORCID holders' own boxes (the full 2025-10 dump, 24,990,909 records), three parliaments that write surnames in capitals (tr_parliament, az_parliament, kz_mazhilis), and the other cached corpora. The switch holds up: taking the capitals words as the family, a prototype reads 591 of 591 Turkish names (the default reads 549), 124 of 124 Azerbaijani (default 0) and 152 of 152 Kazakh. But the wording as written broke on shapes that occur, so it is narrowed in six places.

What counts as a capitals word. Two or more letters written in capitals, O'BRIEN and McDONALD included (a lowercase prefix glued in front does not make the word a name in mixed case). A word holding a full stop or a slash does not count, nor anything in brackets or quotes. Without this, DR. Anna Berg read family DR. (192 ORCID given boxes open with DR.) and Tom N/A Reed read family N/A (a placeholder, in every one of the 128 such names in the UK PSC register).
One run. Runs joined only by particles or conjunctions are one run (Aude BROSSARD de VILLEMONT). A lowercase particle written right before the run joins it: Jean de LA FONTAINE had left de as a middle name (130 ORCID family boxes are written particle + capitals). A run of particles alone is no run, and the name reads by position: Karim AL Haddadi had read family AL.
Two runs. Where a name word separates two runs, the last run is the family and the earlier one keeps its place among the other words: PETER Okello MUWANGA → given PETER, middle Okello, family MUWANGA. Composed given box first, the last run sits in ORCID's family box in 3,113 of 3,971 such names, the first run in 2.
Which names it reads. A name is mixed case only where a word other than a title or a particle holds a lowercase letter; otherwise the switch does nothing. Dr. ANITA ROY had read the whole name as the family (5,878 ORCID records have lowercase letters only in a title), as had MARIA de SOUZA LIMA (about 180, lowercase only in particles).
The peel. It never leaves only particles: Anna DE CARO stays family DE CARO (7 ORCID surnames of this shape). A lone roman numeral, II and up, stays the suffix, and the switch does not act: John Smith III had read family III (2,567 ORCID records hold a numeral beside a name in mixed case).
Commas. The switch reads only the part before the first comma; after a comma the comma rules decide. Every capitals word found after a comma in the local corpora was a credential (Laure MARTEL, MD), and no real name had a capitals name word after a family comma.

Accepted costs, to be stated in the rule:

Joined initials in the middle of a name read as the family: Jan GM Brouwer → family GM. 4,246 ORCID records put such a run in the given box. It is the same shape as Jeffrey LAM Kin-fung, so it cannot be narrowed by length.
A credential after the surname with no comma reads as the family: John Smith MD → family MD. 757 ORCID records, none from a source that writes families in capitals.

Left open: a bare DR in capitals at the head (DR Anna Berg) reads as the family under the switch (204 ORCID given boxes).

Not measured: France's register and the LegCo lists (not cached), and the parliament datasets fetched after 2026-11-08 (Korea, Indonesia, Ethiopia, Israel). A later refresh added Korea's: 293 of its 297 names in Latin script are written wholly in capitals and the other 4 have no word in capitals, so the switch reads none of them.

The review of the implementing PR (#1363) settled some readings this wording does not spell out: a conjunction or suffix written in lowercase does not make a name mixed case; a particle written with a capital outside a run counts as a name word (MORRISON Van is read); a bare one-letter listed conjunction between two capitals words joins them (Ana LOPEZ Y GARCIA → family LOPEZ Y GARCIA). rules/O9 and decisions/O9 in that PR hold the full specification.

🤖 Generated with Claude Code

Measure rules/O9's scaling row on a unit whose default parse is linear (

Navigation Menu

Repository navigation

Should an ALL-CAPS word in a mixed-case name mark the family name? `DUPONT Jean` gives given `DUPONT` #659

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Should an ALL-CAPS word in a mixed-case name mark the family name? DUPONT Jean gives given DUPONT #659

Description

Activity

derek73 commented on Oct 16, 2026

derek73 commented on Nov 8, 2026

derek73 commented on Nov 8, 2026

derek73 commented on Nov 8, 2026

derek73 commented on Nov 11, 2026

derek73 commented on Nov 12, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

Should an ALL-CAPS word in a mixed-case name mark the family name? `DUPONT Jean` gives given `DUPONT` #659