Repository navigation
Should unspaced 황보라 read 황 + 보라? (황보 and 서문 take the first syllables of names that are 황 or 서 plus a given name) #1213
Description
Rule W1 (unspaced CJK division) divides an unspaced hangul name after the longest listed surname, so a name whose first two syllables spell a two-syllable surname always gets that surname. For 황보 and 서문, the first syllable is a common surname too, and given names that open with the second syllable are common, so one string can be two different names (measured on master 5f22996):
| Input | Person (Wikidata) | family | given | ambiguities |
|---|---|---|---|---|
황보라 |
Hwang Bo-ra (Q6652730) | 황보 |
라 |
segmentation |
황보람 |
Hwang Bo-ram (Q20090966) | 황보 |
람 |
segmentation |
황보승희 |
Hwangbo Seung-hee (Q54802943) | 황보 |
승희 |
segmentation |
황보명 |
Hwangbo Myung (Q134569113) and Bo-Myung Hwang (Q134569097) | 황보 |
명 |
segmentation |
서문경 |
Seo Moon-kyoung (Q12600913) | 서문 |
경 |
segmentation |
서문형원 |
Seomun Hyeong-won (Q113577773) | 서문 |
형원 |
segmentation |
황 보라 |
(spaced) | 황 |
보라 |
none |
The parser has reported this fork as a segmentation ambiguity since #271, but nobody has measured how often the longest match picks the wrong split.
Measured (2026-11-08) on Wikidata's people with a South Korean hangul label opening with a two-syllable surname (102 people). Each split was judged by the person's English label:
황보: 8 people are황+보X(황보라,황보람,황보미,황보름,황보름별,황보운,황보훈, and one황보명), against 5 named황보.서문: 2 against 2.- Others: none of
남궁,선우,사공,제갈,독고or동방reads wrong.
For a census-scale estimate, I multiplied each one-syllable surname's 2015 census bearers (English Wikipedia's List of Korean surnames) by the share of given names opening with the second syllable. That share comes from Wikidata's 34,373 three-syllable South Korean labels.
| Surname pair | Single-surname bearers | Share of given names | Estimated wrong | Compound bearers | Wrong share of unspaced names |
|---|---|---|---|---|---|
황 + 보X vs 황보 |
697,475 | 보 0.614% |
≈4,280 | 10,427 | ≈29% |
서 + 문X vs 서문 |
752,233 | 문 0.544% |
≈4,090 | 2,044 | ≈67% |
선 + 우X vs 선우 |
42,842 | 우 0.975% |
≈420 | 3,648 | ≈10% |
사 + 공X vs 사공 |
10,998 | 공 0.04% |
≈4 | 4,488 | ≈0.1% |
남궁 and 제갈 have none: no label's given name opens with 궁 or 갈. OpenSanctions kr_assembly (299 members) has no two-syllable surname, so it measures nothing here.
Precedent. The Chinese pack's compound test, from decisions/W1 (2026-10-18, #695), turns down a compound where "a Wikidata bearer of its first character alone has a name opening with it". The one exception is a compound whose living bearers outnumber those readings by orders of magnitude, which is how 張簡 was admitted. That test was never applied to the Korean list. Applied now, it turns down both 황보 and 서문, but the census estimate gives opposite answers for the two:
황보: right for about 71% of unspaced황보Xnames.서문: right for about a third.
Proposal, in rule shape (option B):
- When: an unspaced hangul name opens with
서문. - Then: it divides after
서. - Because: about two in three such names are
서+문X. - Accepted:
- bearers of
서문(2,044 in 2015) read서+문X; - with only one listed prefix matching, their names carry no
segmentationreport; 황보stays, and its cost goes on record in decisions/W1 with the recompute above.
- bearers of
Alternatives: A, keep both and record the cost (today's readings); C, drop both.
Questions
- Is the census-times-Wikidata estimate enough to decide on, or should this wait for a source that lists whole names with their surname? The National Assembly's
ALLNAMEMBERAPI has 3,296 members, but it needs a free key. - Should the Korean list carry an admission note for its compounds, like the one above
_COMPOUNDinlocales/zh.py? - If
서문goes, should its bearers keep a report that the division is a guess? That would need W1 to report a split where the dropped compound would have matched.
Activity
Add Korea's and Israel's datasets to opensanctions-parliaments (#1218)
derek73 commentedon Nov 9, 2026
On question 1, about waiting for the National Assembly's
ALLNAMEMBERlist: that list can't settle this issue.Measured (2026-11-08) on Wikidata's holders of "Member of the National Assembly of South Korea" (Q14850694). Wikidata has 3,264 of them; the Assembly API reports 3,296. Of these, 3,166 have a hangul label. The members whose label opens with a two-syllable surname:
황보승희선우련선우is a compound surname)남궁현,남궁진,남궁석No member is
서문X, and none is황+보X. Every parliament in the Assembly's history holds too few of these names to measure the split. (Two other members,어기구and어준선, bear어; they bear on #1215.)So at the scale of whole names, the census estimate above is the only measurement there is. A source that could replace it would have to be a large list of whole names with the surname recorded, such as a sports federation's or a civil-service register. None is in hand. Without a key, the Assembly API returns only 5 sample rows, and a key needs Korean identity verification.
derek73 commentedon Nov 9, 2026
A second measurement, from sources where the person typed the family name in a box of its own, so the box shows where the surname ends.
Measured (2026-11-08) on ORCID's 2025-10 public summaries (about 25M records, family and given names in separate boxes) and PatentsView's inventors (to 2025-12-31, counted by person). "KR" means the record gives South Korea as its country.
황보황/보라,보각,보현HwangboHwang/Bo-Yeon, …Seomun,SeomoonSeo/Moon Jeong,Suh/Moon Suhk, …서문spellingsHwangbo, 3Seomun/Seomoon황보; 4 for서문(Seo/Mun-Jun, …)남궁, 14사공How often the first syllable alone is the surname:
황보서문The controls behave as the census predicts:
남궁and사공almost never split.Limits:
Hwang/Borais missed and the column undercounts.Hwang/Po-Wen).The figures support the census estimate rather than replace it, and they point the same way as the proposal: drop
서문, and keep황보with its cost written down.