週刊OSM 757

3 Replies to “週刊OSM 757”

1. Since the German l10n tools are mentioned here twice, it’s worth noting the inherent limitations to the underlying transliteration software: because OSM does not explicitly indicate the language of the value in the name=* key, the transliterator cannot choose a transliteration scheme appropriate to that original language. Instead, it falls back to a language-agnostic transliteration scheme that’s only designed for collation, not display. Essentially, the transliterator detects the writing system and assumes a particular language, such as Russian for all Cyrillic text, Standard Arabic for all Arabic text, and Mandarin Chinese for all Chinese and Japanese text.

   The upshot is that, if you use this tool for displaying labels on a world map, you can expect almost every label in Ukraine, Iran, Pakistan, or Japan to be wrong or even politically offensive. A developer might decide that showing something misleading is better than showing something completely unreadable, but they should be aware of the implications.

   1. Sorry, I was writing from (false) memory when I first replied; but at the time I’m writing this, that reply has not made it through moderation anyway. (In the preview after submitting, my paragraph breaks looked broken; I don’t know if that caused a rejection.) I have since had the opportunity to quickly check and compare OSM’s German style with OSM’s default style on a more robust system.

      Minh Nguyễn is right: If you use the German style and zoom in on China, Taiwan, or Japan, you see (pseudo*) Hanyu Pinyin everywhere. Which is a nice feature for Chinese places, whereas for Japanese places it’s mostly useless and possibly misleading clutter.

      So, OSM default style displays all “CJK Unified Ideographs” (to use Unicode terminology, however inaccurate the term “ideograph” may be) using Japanese-looking glyphs, even for explicitly Chinese names of places in China.
      The German style does the same thing and additionally displays each ideograph’s most likely Modern Standard Mandarin pronunciation, confusingly even for Japanese names of places in Japan. But this only happens as a fallback wherever OSM hasn’t got a proper romanized form (don’t know if and how it also depends on the stated language of that form), effectively meaning you will only notice if you zoom in.

      We might add the pitfalls of labeling places in Poland (and similarly, in Czechia), where there is often (1) a traditional Polish name if Polish was (also) spoken in or near the place, and/or (2) a traditional German name if German was (also) spoken in or near the place, and/or (3) a non-traditional German name invented by a Nazi bureaucrat after annexation if they dismissed the traditional German name as sounding too Slavic or there was none, and/or (4) a non-traditional Polish name adopted after annexation if the traditional Polish name sounded too German or there was none. Even where (2) existed, it may have seen less currency than (1), even on old German maps. I can’t check now so I don’t remember if OSM German style prefers (2) or (3) for German labels, nor how it tells them apart.

      I still don’t see any potential problems with using ISO 9 for Cyrillic (for example: я→â), which is what German style uses as fallback. I think it’s great to have as a fallback for any Cyrillic.

      As I wrote in my first reply, it helps if you know whether what you’re looking at is just a language-agnostic transliteration which may not convey a good idea of pronunciation; but then, neither does the “proper” transcription for Russian, as it doesn’t indicate stress and neutralization of unstressed vowels.

      *) “Pseudo” because firstly, in the case of Japan, the transliterations are obviously not real-world Chinese names; secondly, the converter cannot identify word boundaries, so converts everything into space-separated single Chinese syllables unlike proper Hanyu Pinyin; thirdly, because it cannot tell where to use uppercase, so it uses lowercase throughout, unlike proper Hanyu Pinyin; and fourthly, there may be some wrong guesses when a character has multiple readings in MSM.

      1. “Almost every” might’ve been hyperbole on my part, but suboptimal transliterations are quite common in non-Russian Cyrillic contexts. There are some examples in https://github.com/onthegomap/planetiler/issues/86#issuecomment-1050039099 and subsequent comments from Planetiler, which uses the same l10n library. (I got a chuckle out of the transliteration of 向 as “xiàng~ke”, which is perfectly reasonable for collation but not so much for display.)