Pronunciation
Where a word's meaning is ambiguous unless you know how it is said, the reader needs a way to find the pronunciation. The words in context do most of the work here. If the sentence around the word already settles which reading applies, nothing is required, which is why English heteronyms like desert and desert usually need no help at all.
Why it matters
A screen reader has to commit to one reading, and when it picks wrong the sentence stops making sense. W3C puts it well. A word read aloud with the wrong pronunciation can be harder to understand than the same word read on the page. The problem is much larger outside English. A Japanese kanji can carry several unrelated readings, and Hebrew and Arabic leave out marks that would settle the sound, so a run of text can mean different things depending on how it is voiced.
Who this rule protects
Blind readers, readers with low vision, and readers with reading disabilities are the three groups W3C names, along with anybody whose technology reads the words out loud.
How to check it yourself
- Look for words whose meaning changes with the pronunciation, and check whether the sentence already settles which one applies. If it does, nothing is required.
- For Japanese, Chinese or Korean content, check for ruby annotation wherever a reading is genuinely ambiguous.
- For Hebrew or Arabic, check whether diacritical marks are available and whether the reader can switch them off.
- Where the same spelling has two pronunciations on one page, check that every occurrence carries its own guide, since a single note cannot cover both.
Failures we see most often
- Ambiguous kanji readings appear with no ruby annotation to resolve them.
- A Hebrew page carries no diacritical marks and no other route to the reading.
- A pronunciation is given once for a word that is said two different ways on the same page.
Who this one is for
Read from this rule's own note above, so the grouping and the note cannot disagree.
- Blind and screen reader userspeople who cannot see the screen
- Low visionpeople who can see the screen but not easily
- Cognitive and learningpeople for whom the difficulty is understanding, remembering, or staying with it
How to fix it
- Use ruby annotation for East Asian readings. Browsers have supported it widely since 2015, and it carries its own fallback for anything that does not.
- Offer diacritical marks the reader can turn off, which is how W3C's technique for Hebrew and Arabic is written.
- Link the word to a recording of it, which is a documented route and often the clearest one.
- Prefer ruby to a bracketed spelling in Japanese. A screen reader reads a parenthetical kana gloss as well as the kanji, so the listener hears the word twice.
Step-by-step fix guides (5)
Passes vs. fails
Passes
The same page adds ruby annotation to the name, giving the one reading that was meant.
Fails
A Japanese page leaves a name's reading ambiguous, with no ruby annotation attached.
In audits and lawsuits
This is Level AAA and rarely comes up on an English-language site. It becomes real the moment you publish in a language where a written character's reading changes its meaning. The routes differ by writing system, which is the detail most explanations miss. Ruby annotation is the East Asian route. Diacritical marks are the Hebrew and Arabic route, and W3C's technique for those is not simply to add them. It is to let the reader turn them on and off. There is a bonus worth knowing here too. A glossary carrying pronunciation satisfies this criterion, and the same glossary can satisfy Unusual Words and Abbreviations, so one artifact closes three AAA criteria.