RC RANDOM CHAOS

OAuth converts consent into standing permission

A blog that declares lang=en-GB refuses to localize its British references, and the HTML lang tag drives more than spelling.

· 3 min read
OAuth converts consent into standing permission

Every post on shkspr.mobi opens with the same declaration:

<!doctype html>
<html lang=en-GB>

A reader recently left a comment asking the author to make the writing more inclusive, suggesting the British cultural references be swapped for tropes that are more globally known. The author declined and pointed at that lang=en-GB line as the reason. The tag is more than the language, the argument runs: it is the culture, the way of thinking, and the accent behind the text.

That is a stronger claim than most people make for an attribute they copy-paste once and never touch again. It is also mostly right about what the attribute is for.

lang declares a document’s natural language as a BCP 47 tag. en-GB is a primary language subtag, en, plus a region subtag, GB. Conforming software reads it. Screen readers use it to select a voice and pronunciation rules. Browsers use it for hyphenation, quotation marks, and spellcheck dictionaries. Search engines use it for regional relevance. Translation tools treat it as the source language. AI narration that reads a page aloud uses it, or should, to decide how the text sounds. Set en-GB and a text-to-speech engine that honors the tag picks a British voice. The author’s position is that this is the correct output, and that an unfamiliar accent is not a defect to fix.

The content argument sits underneath the markup one. The post uses the US edition of the first Harry Potter book as its example: retitled from “Philosopher’s Stone” to “Sorcerer’s Stone,” with around a dozen other words changed, to the annoyance of the fandom. “Sherbet lemon,” a character’s “fringe,” a “jumper,” “Count Duckula,” “Accrington Stanley,” and the occasional “wanker” stay in deliberately. The claim is that a reader hits an unfamiliar word, uses context, and carries on, the same way the author read a Twinkie for years as “an unappealing foodstuff which, nevertheless, were inexplicably popular” before ever seeing one.

For anyone building systems that consume this text, the practical reading is to honor the tag rather than flatten it.

  • If you generate TTS or AI narration, map the region subtag to a voice. Reading en-GB content in a default US voice is a bug.
  • Don’t rewrite content to a house locale in your pipeline. Localization that strips region-specific vocabulary is the Sorcerer’s Stone edit at scale, and it throws away information the author put there on purpose.
  • Set lang correctly and specifically on your own documents. en alone tells a consumer less than en-GB or en-US, and the region subtag is the part that drives voice and dictionary selection.

There is a limit the author flags in a footnote, and it is real. BCP 47 gives you language and region, so en-GB against en-US, but very little below that. Regional accents within a country have no standard subtag that a TTS engine acts on. The tag can mark text as British English; there is no supported way to say which part of Britain, so an engine cannot voice the difference even where it matters.

Share

Keep Reading

Latest on the Wire

Full wire →

New signal daily · RSS

Stay in the loop

New writing delivered when it's ready. No schedule, no spam.