Should a credential, generation or title joined by a hyphen other than the hyphen-minus read as one? (John Smith MD‐PhD gives family MD‐PhD, where MD-PhD gives suffix) #1349

rules/O6 lists eight hyphens that join two names: the hyphen-minus, U+2010, U+2011, the figure dash, the en dash, the minus sign, and the small and fullwidth hyphen-minus. The name match has read all of them as one since #1264. The parser's tests for a joined credential, generation or title still split at the hyphen-minus (and the slash) only. So the same text written with any other hyphen becomes a name word.

Measured on master 970b54e, 2026-11-11 (‐ is U+2010; the en dash reads the same):

Hyphenated names themselves are not affected: Jean‐Paul Smith‐Jones and ANNA VAN‐DER‐BERG (case repair gives Anna van‐der‐Berg) read as their hyphen-minus forms do.

Where the hyphen-minus is asked alone:

• _vocab._JOINERS, _JOINER_SPLIT, the "-" in … tests in _vocab, _comma, _classify and _post_rules.
• N2's particle head (_extract._opens_a_listed_particle, word.partition("-")).
• R4's hyphen clause in _render ("-".join).

_vocab._HYPHENS already holds the full set and _match reads it.

How often (2026-11-11): for every name in the 16 local corpora and the committed corpora, a script parsed the text as written and with each other hyphen mapped to the hyphen-minus.

• 1,716 names hold another hyphen: jrc-names 1,662, orcid-researchers 50, govuk-people 1, corpus_rules 3.
• One moves when mapped: C10's committed boundary row John Smith, MD – PhD (the spaced form, Should a lone dash be a separator, so John Smith - Jr. reads family 'Smith' rather than '-'? #820).
• No real name moves, so every one of them is a hyphenated name, not a credential.
• ctgov-officials holds no other hyphen at all. Its fetch folds none, so this is how the source writes them. Yet 1,670 of its names put a hyphen-minus inside a suffix or title, which is how large the class would be if a source wrote U+2010.

Options:

• (A) Read the whole O6 set wherever a joiner is asked. Credentials, generations and titles; N2's particle head; R4's hyphen clause.
  • When: a word holds one of rules/O6's hyphens between two chunks.
  • Then: it is a joined credential, generation or title, or a particle hyphened to its surname, exactly where the same word written with the hyphen-minus is. The word keeps its own character.
  • Because: the hyphen a writer's keyboard or word processor substitutes says nothing about the name. O6 already reads these hyphens alike between names, and the match reads them alike since Which smaller name-matching gaps should be fixed? (J.P. Smith / John Paul Smith, John Beth Smith / John Anne Beth Smith, and others) #1264.
  • Except: a hyphen standing alone between words stays Should a lone dash be a separator, so John Smith - Jr. reads family 'Smith' rather than '-'? #820's question. The slash is unchanged.
  • Accepted: none found. No measured name moves.
• (B) Fold the O6 hyphens to the hyphen-minus once, for lookup only, and leave the text as written. This is the same reading as (A) with one site instead of several, but every predicate then reads a folded copy.
• (C) Leave it as the recorded limit. No measured name is affected. decisions/name-matching's 2026-11-11 Which smaller name-matching gaps should be fixed? (J.P. Smith / John Paul Smith, John Beth Smith / John Anne Beth Smith, and others) #1264 entry already states the limit, and _PARSER_HYPHEN_LIMITS in tests/v2/test_match.py pins today's readings.

If (A) or (B) ships:

• The _PARSER_HYPHEN_LIMITS rows flip.
• The "Limit" bullet in decisions/name-matching's Which smaller name-matching gaps should be fixed? (J.P. Smith / John Paul Smith, John Beth Smith / John Anne Beth Smith, and others) #1264 entry, and its Standing line's scope, become history.
• The shape and spelling sweeps in nameparser/AGENTS.md gain the class.

Open:

1. A, B or C?
2. Does the snapshot need a recorded control, given that no corpus name moves? A grid of the rows above, under the three orders and the four comma shapes, would carry the change.

🤖 Generated with Claude Code