Repository navigation
Should a lone dash be a separator, so John Smith - Jr. reads family 'Smith' rather than '-'? #820
Description
A dash standing alone between words is read as a name word. Every release since 1.4.0 reads it that way:
| input | reads |
|---|---|
John Smith - Jr. |
middle 'Smith', family '-', suffix 'Jr.' |
Jane Doe - MA |
middle 'Doe', family '-', suffix 'MA' |
Jane Doe - Smith |
middle 'Doe -', family 'Smith' |
Jane Doe nee Smith - MA |
maiden 'Smith -', suffix 'MA' |
Jane Doe nee Smith – MA (en dash) |
maiden 'Smith –', suffix 'MA' |
Declaring the dash with Policy(extra_suffix_delimiters={' - '}) changes none of these. C10 (a declared delimiter parts a trailing suffix part) applies only after a comma, and in the part before the first comma a declared delimiter is a word.
The maiden case was recorded rather than fixed in decisions/M2's 2026-09-19 #533 review entry, "left for its own issue". It is one symptom of the shape here.
Proposal, in rule shape
When: a token consists only of a dash (hyphen-minus, en dash, em dash) standing between words, as in "John Smith - Jr.".
Then, one of:
- (A) It is a separator (T1). It splits like whitespace and is dropped:
John Smith - Jr.→ family 'Smith', suffix 'Jr.'. - (B) It is a suffix delimiter, as a comma would be. It parts the name there:
Jane Doe - MAreads asJane Doe, MA. - (C) It stays a word. The rule states this as an Accepted cost.
Because: a lone dash carries no name content, so it can't be a family name. A dash inside a word (Smith-Jones) stays a hyphen; this question is only about a dash with spaces around it.
Boundaries that stay as they are: Smith-Jones (one word); Smith, RN - CRNA (C10's Accepted bullet keeps the core a word there for v1 parity, and the answer here would have to say whether that changes).
Open questions
- A, B or C?
- Should one answer cover every dash (hyphen-minus, en dash, em dash), and other lone punctuation such as
/or·? - Does the answer change the comma path, where C10 already reads a declared delimiter as a comma?
Activity
Narrow C3's and C10's Accepted bullets; point C9 and M2 at #819 and #820
derek73 commentedon Oct 25, 2026
Corpus survey, 2026-10-24: what a lone dash is in real data
I scanned every committed corpus and the eight cached local corpora for a dash (
-,–,—) standing alone between words. The ClinicalTrials.gov contact notes ("Call 1-877-CTLILLY … Mon - Fri") are left out. Counts are distinct strings.Camellus Law Llp - Trustee - I.O.M. …)Alessandro Armuzzi, MD - PhD), 1 beforeNOEL - BLAGROVE, DIONNE)Kamala Rokaya - Bohara)Chelsea - Everton)Peter - I)Smith, RN - CRNA; CLDR'sфон Брюль — мл.)The shape this issue opens with, a credential after a dash with no comma before it (
John Smith - Jr.), does not appear as a real name in any corpus. What people actually type is a hyphen with spaces around it inside a compound name.PSC shows which box the filer put each part in. Of its 454 names, 364 have the dash in the surname box (
Barnabas - Wood,Al - Kufi), 73 in the middle name and 17 in the forename (Alexandru - Ionut). Over the 364,familymatches the surname box (comparing with dash spacing ignored):Mr Adam John Holland - Edwards→ middle'John Holland -', family'Edwards'The 15 misses under D, all of them:
Wise - As Trustee,Watson - The Estate,Blake - Executor Of …,Cartwright - As Nominee For …).Mrs Annette Erica Hinds Tia - Okwee→'Tia-Okwee', box'Hinds Tia - Okwee'; alsoManzanaro Fernandez - Montes,Olugu Ikechi - Iro,Hameedi Shah - Jan,A A Al - Kuwari,Amobi Kesse - Adu). That is the ordinary multi-word-surname limit, the same without any dash.Kuijpers - Van Den Bos→'Den Bos',Short - Van Der Laan,Dickhoff - D'Eimar De Jabrun,Schwagele - Fan Moniz,Sanchez - Osorio Nicolas).So the data suggests a fourth answer: (D) a lone dash between two name words is a hyphen with spaces around it, and joins them (
Spindler - Vesel→ family'Spindler-Vesel',Doru - Liviu Constantinescu→ given'Doru-Liviu'). Whether the rendering keeps the spaces is a separate question.D cannot be a blanket rewrite. A naive join breaks two shapes the corpora hold:
Alessandro Armuzzi, MD - PhDreads correctly today: suffix'MD - PhD', which is C10's Accepted bullet. Joined intoMD-PhD, the run is no longer suffix vocabulary, and the comma flips to a family comma (family'Alessandro Armuzzi', given'MD-PhD'). The 26 ctgov rows show this is typed. The join must not reach a credential run, which also settles open question 3: the comma path stays as it is.Mr - Ibraheemthe filer typed-as the forename. Joined, it reads given'Mr-Ibraheem'. Today it reads title'Mr', given'-', family'Ibraheem'. A join must not take a title.Names in surname-comma form (
NOEL - BLAGROVE, DIONNE) already get the whole family right; only the spacing would change.On open question 2:
/comes up too (ctgovMD / PhD, 55 strings; nepalLaxmi Sarki / Acchami, 91), but it means "or" or names an alternate surname, which is a different question, so I'd keep it out of this rule.•and·show up only in jrc-names news extraction.Recipe: PSC rows come from
psc-individuals/raw.jsonl, keeping each record whosesurnameholds a whitespace-flanked[-–—]+token. Compareparse(name).familywith thesurnamefield, after folding\s*[-–—]+\s*to-and lowercasing both. D isre.sub(r'\s+([-–—]+)\s+', r'\1', name)applied before the parse.