Should a conjunction inside a hyphenated surname keep its lowercase? `Ortega-y-Gasset` now capitalizes to `Ortega-Y-Gasset`

[derek73]

capitalize() on a hyphenated compound surname now uppercases a conjunction inside it. Measured on the released wheels and on the 2.2 branch, all with force=True:

input	1.4.0	2.0.0	2.1.0	2.2
jose ortega-y-gasset	Jose Ortega-y-Gasset	Jose Ortega-y-Gasset	Jose Ortega-y-Gasset	Jose Ortega-Y-Gasset
JOSE ORTEGA-Y-GASSET	Jose Ortega-Y-Gasset	Jose Ortega-Y-Gasset	Jose Ortega-Y-Gasset	Jose Ortega-Y-Gasset

The two spellings disagreed on every version through 2.1. #458 made them agree — by taking the uppercase answer, which is a deviation from 1.4.0 for the lowercase spelling. The same applies to Silva-e-Sousa, -et-, -und- and every other member of Lexicon.conjunctions.

Why it happens

#458 moved the conjunction-or-initial decision into the parse, and the decision is the whole token's. Ortega-y-Gasset is one family token carrying no conjunction tag, so repair capitalizes each word of it.

The spaced spelling is unaffected and still reads the convention correctly — there y is a token of its own and is tagged conjunction:

parse("jose ortega y gasset")   # family "ortega y gasset", the `y` tagged conjunction
# capitalize() -> "Jose Ortega y Gasset"   (unchanged on every version)

The case for leaving it

A hyphenated token is one word to the parser, and re-deciding per sub-word inside a token is exactly what #458 removed — it was producing two answers for one name. Reintroducing a sub-word rule for conjunctions only would be a narrower version of the thing that was wrong.

The case for changing it

Ortega-y-Gasset is a real orthography with a real convention, and the hyphenated spelling is common in library catalogues and bibliographies. The library gets the spaced form right, so a user reasonably expects the hyphenated form to match. Today the two spellings of one name repair differently.

Scope, measured

• The parse is unaffected. Ortega-y-Gasset is a single family token before and after; only capitalized() differs.
• 0 of 1,103 differential corpus names have a conjunction inside a hyphenated token, and compare.py never calls capitalized() anyway (decisions.md#R4), so no gate run can see this either way.
• No safe workaround today. A capitalization_exceptions key of ortega-y-gasset never matches — keys are looked up per word. A 'y' -> 'y' entry does restore it, but breaks initials: y. smith repairs to y Smith.

Found while replacing the constructed e-f example in 2.2's release note with a real name (#477).

[derek73]

Related: #479 — the same y/e word, but the parse rather than the rendering, and a different cause. There an all-caps name reads the conjunction as an initial because nothing checks whether the capital carries any signal; here a hyphenated token has no conjunction tag to read at all.