10

There's a strait which is called Øresund in Danish and Öresund in Swedish.

Looking at Latin Capital Letter O with Stroke, it has no decomposition rules.

Looking at Latin Capital Letter O with Diaeresis, it decomposes into Latin Capital Letter O and Combining Diaeresis.

Why is Stroke not a Combining Stroke? I can't find any such Combining Mark at all. In other words, why is "Ø" a base letter (sorry, it seems there's no better term based on this answer) when "Ö" is not?

3 Answers 3

20

Contrary to what the other answers have stated, the Unicode Standard does not actually care whether Ø or Ð or any other character is “fundamentally a different letter” or not, whatever that may even mean. Nor does it matter what purpose certain diacritical marks are used for. It’s all about typography and the capabilities of fonts and text engines.

From section 2.12 of The Unicode Standard, Version 15.0.0:

Most characters that one thinks of as being a letter “plus accent” have formal decompositions in the Unicode Standard. [...] There are, however, exceptions involving certain types of diacritics and other marks.

Based on the pattern for accented letters, implementers often also expect to encounter formal decompositions for characters which use various overlaid diacritics such as slashes and bars to form new Latin (or Cyrillic) letters. For example, one might expect a decomposition for U+00D8 LATIN CAPITAL LETTER O WITH STROKE involving U+0338 COMBINING LONG SOLIDUS OVERLAY. However, such decompositions involving overlaid diacritics are not formally defined in the Unicode Standard.

For historical and implementation reasons, there are no decompositions for characters with overlaid diacritics such as slashes and bars, nor for most diacritic hooks, swashes, tails, and other similar modifications to the graphic form of a base character. In such cases, the generic identification of the overlaid element is not specific enough to identify which part of the base glyph is to be overlaid.

Essentially, overlaid diacritics are too ambiguous in most situations to be of practical use. A sequence “capital D + overlaid short stroke” could result in Ð, with the stroke crossing the left side of the D, but the stroke could just as easily go over the right side, or the middle (like Ꟈ), or lie somewhere in the upper or lower parts of the glyph. And there’d be no way to differentiate any of these options.

There is also the problem that diacritics like this often need to be contorted to fit the shape of the base character. Consider all the different letters with an overlaid tilde for instance: ᵬ ᵭ ᵮ ᵯ ᵰ ᵱ ᵲ ᵳ ᵴ ᵵ ᵶ. Modern fonts can certainly deal with this, but any font that didn’t specifically account for all possible combinations would look awful, potentially even leaving some letters unidentifiable.

A similar issue exists for the palatalized hook and retroflex hook diacritics, which is why these are also not used for decompositions. For example, an s with a palatal hook looks like ᶊ which is easy enough, but an ʃ with a palatal hook actually has it attached to its side by an extra “stem” like this: ᶋ

There are very few overlaid diacritics in Unicode and those that are actively used mostly belong to writing systems where complex shaping rules and glyph substitutions are a prerequisite for rendering any text at all.

5
  • 1
    “but the stroke could just as easily go over the right side, or the middle (like Ꟈ), or lie somewhere in the upper or lower parts of the glyph. And there’d be no way to differentiate any of these options” — Would that be a problem, as long as the decomposition were compatible rather than canonical? Decomposition isn’t bijective, anyway. The same issue exists with ogoneks (the placement in ę is different from ą and ų – in the latter two, the placement of the ogonek differs from, say, a tilde below), and ogonek combinations do decompose. Jul 19, 2023 at 15:21
  • Ogonek, cedilla, and the Vietnamese horn are the three major exceptions to the “no attached diacritics” rule. I suppose their behaviour was predictable enough that canonical decompositions weren’t considered too much of an issue, though they do sometimes still cause problems in implementations. I don’t think compat decompositions would have been a good solution. They kinda defeat the purpose of combining diacritics since the decomposed sequences wouldn’t be equivalent to the precomposed char, so you would always need to use the precomposed chars anyway. You wouldn’t really gain anything. Jul 19, 2023 at 15:46
  • 1
    I don’t really agree with that last part – you gain nothing composition-wise, but for things like sorting strings alphabetically, being able to actively normalise đ or ᵰ would be very useful in contexts where you don’t have the advantage of database collation algorithms doing the heavy lifting for you. Jul 19, 2023 at 16:25
  • 1
    I suppose you’re right, but maybe it would have sent the signal that you can use these diacritics productively which is definitely not recommended. For a lot of compat decompositions (though certainly not all), the decomposed sequence is generally the recommended representation and the precomposed char has little to no use outside of mappings to legacy encodings. Jul 19, 2023 at 18:04
  • 1
    Fun fact, you can do crossword puzzles in Arabic, and when doing so, each letter stands alone.
    – Joshua
    Jul 19, 2023 at 19:58
6

For the same reason that eth (Ðð) doesn't compose into Dd + COMBINING SHORT STROKE OVERLAY. Namely, it looks like a D with a stroke through it, but it's fundamentally a different letter, and there's no systematicity to how you would add bars to other letters to change their meaning.

The umlaut/diaeresis on the other hand is used on a bunch of different vowels, generally in a systematic way—you put an umlaut on a back vowel to make it front, producing ü ö ä from u o a, or you put a diaeresis on any vowel to indicate it's not a diphthong, producing ë ï and so on.

Though a COMBINING SHORT/LONG SOLIDUS OVERLAY (U+0337 and U+0338) was eventually added, Ø is fundamentally considered its own thing rather than an accented form of O, much like Đ and G are (despite originating as modifications of D and C with an extra stroke).

(It's worth noting that this argument was originally applied in general rather than for the purposes of any specific language. In Spanish, Ññ is treated as its own letter, and there's no system for applying a tilde to anything else. But in Portuguese, it's applied systematically to vowels. So the tilde gets treated as a diacritic rather than an inseparable part of the letter Ñ.)

7
  • 3
    Letters with umlauts/diaresises, at least ä, ö, ü, are also fundamentally considered to be their own things rather than accented forms forms of a, o, u. But only in some of the languages that uses them. That's the case in the Scandianavian languages, but not in German. As for the tilde, it is at least also applied to m: m̃ occurs in a few languages. So I'd say there's more to it. Such as whether the changed letters are only in one language, how important that/those language(s) are, and whether the same or similar symbol is used in different ways in different languages. Jul 19, 2023 at 8:54
  • Eth is a somewhat special case, because it contrasts with the ‘regular’ stroked d ⟨đ⟩ as used for example in Serbian, and having two different characters decompose to the same sequence is not tenable. But aside from that, there is certainly a systematicity to adding bars to other letters (the meaning of those letters is unrelated to their composition). For the purpose of alphabetising and diacritic-free writing, ⟨ø⟩ does decompose to ⟨oe⟩ just like ⟨ö⟩ does, and there really is no reason why ⟨ø⟩ and ⟨Ø⟩ shouldn’t decompose to o + U+0337 and O + U+0338, respectively. Jul 19, 2023 at 9:30
  • Compare how the similarly limited letter ⟨å⟩ (also primarily used in Scandinavia, and also nearly unique – apart from Czech ů – in using a ring modifier) does decompose into a + U+030A. Meanwhile, of the three very transparent ligatures ⟨æ⟩, ⟨œ⟩ and ⟨ffi⟩, the first two do not decompose at all, while the third decomposes to f + f + i (with compatibility decomposition). There’s no reason why ⟨æ⟩ and ⟨œ⟩ should not be considered compatible with a/o + e. Jul 19, 2023 at 9:54
  • Okay, brain still waking up. Ignore the first part of my first comment above: eth isn’t necessarily that much of a special case. There’s no real reason why ⟨ð⟩ couldn’t decompose (at least as compatible) to d + U+0337 (combining solidus) and ⟨đ⟩ to d + U+0335 (short stroke). There’s even less reason why characters like ⟨ƀ ǥ ᵽ ħ⟩ do not decompose to b + U+0335 – but they don’t. Jul 19, 2023 at 10:03
  • 1
    I had always thought eth was combined with a stroke rather than d combined with a stroke
    – Henry
    Jul 19, 2023 at 10:09
1

Diaeresis was historically used in a number of letter combinations (a,e,i,o,u and sometimes y, even ι and υ) in a large number of languages, but ø / Ø are used only in Danish, Norwegian, Faroese and S. Saami, and slash (distinct from bar) is not historically widely used. The precomposed letters are preserved in Unicode for legacy reasons, and having a dedicated Ø key was a reasonable solution in Scandinavia, whereas five or so precomposed umlauted keys would be impractical.

3
  • 2
    Input methods are separate from encodings. Unicode made legacy precomposed letters for compatibility with legacy encodings, not for compatibility with legacy keyboards or input methods. Jul 19, 2023 at 8:56
  • German keyboards have separate precomposed keys for Ä. Ö and Ü, and a ß key as well. Jul 19, 2023 at 12:47
  • @SirCornflakes Older Mac OS (I think they dropped this around the Catalina release) always composed German umlauts, while Linux and Windows always decomposed them. When several people who had different OS were working on a common SVN repository, after some check-ins and check-outs there were two same-looking filenames within the same directory (one with composed and one with decomposed Umlauts). Jul 21, 2023 at 17:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.