Luc Pardon lucp at
Fri Jan 6 11:23:18 CET 2017

On 06-01-17 06:19, Martin J. Dürst wrote:
> Wouldn't it be possible to create a text-to-speech engine that e.g. uses
> an English dictionary for pronouncing words that occur in English, and
> then a Spanish dictionary for words that aren't found in the English
> dictionary? I don't think such a thing exists, but it wouldn't be too
> hard to create if the pieces are available. Of course, there are some
> words (spellings) that appear in both languages, so they might still
> need fine-grained tagging.

  Sure, that would be possible, and, being a programmer myself, I'd even
say it should not be too difficult. Same for spell checkers and several
other types of language processors.

  However, it will be guesswork at best. It's not only the correctly
spelled "common" words that may cause the engine to guess wrongly. These
ambiguities could be flagged by pre-processing software, such as a
spell-checker that is adapted to do the same guesswork. But what if a
mis-spelled word in one language happens to be identical to a word in
the other? Your spell checker won't be able to help you out here, as it
won't see the error either.

  The question is also whether the screen reader vendors could be
convinced to adapt their software. They too make cost-benefit analyses.

  Also, they may decide, as a matter of principle, that it's not up to
them to step in for authors that can't be convinced to adhere to
standards and tag properly. They may not want to spend good money to
save those authors some relatively small effort, and I would agree.

  Even so, until text-to-speech engines are adapted to cope with
multiple languages in the way you describe, you'd still need full
fine-grained tagging in situations where your document must meet the
WCAG accessibility criteria.

  After that, you'd need to convince the WCAG group to relax its 3.1.2
criterion. You're not likely to succeed, if only because it addresses
broader needs than "just" screen readers. And they may also decide that
guesswork by an engine is not a valid substitute for proper tagging by a
human, and I would agree.


More information about the Ietf-languages mailing list