- From: Jason White <jasonw@ariel.its.unimelb.edu.au>
- Date: Fri, 31 Dec 2004 19:00:52 +1100
- To: Web Content Guidelines <w3c-wai-gl@w3.org>
Gregg Vanderheiden writes: > > 1) Yes - this whole guideline (and all of the programmatically located > provisions) is based on the cascading dictionary concept (or a comprehensive > dictionary). When we put these in we said that they could stay only as > long as we were able to solve that. but if we arent - then we need to > remove them -- not change them to programmatically determined. That has > not been discussed at all. To amplify the latter point, there are good reasons for this: until and unless software techniques are developed that enable natural language to be understood, the best a computer can do is to locate a definition, or set of definitions, associated with a given word. It can't reliably ascertain which definition is applicable in context. A user agent, gateway, proxy server or whatever could consult a standard set of dictionaries selected by the developer, the user, or a combination of both. However, if different "assistive technologies" consult different sets of dictionaries, how is the content developer supposed to know whether the requirement is met. A tool analogous to a spelling checker could in principle identify words in a document that don't appear in a certain set of dictionaries, but if this set is different from that on which the user's hypothetical assistive technology relies, the latter may not be able to locate the required definitions. It might be suggested that the author could nominate dictionaries in metadata attached to the content, but this wouldn't address the problem as the author's chosen dictionaries might be of a different kind than the user requires. Suppose for example the user needs dictionaries that give translations into another language (for example a sign language or a symbol set), but the author nominates dictionaries offering definitions in the primary language of the content. The latter wouldn't help this particular user at all, yet the requirement would be satisfied, and a more appropriate dictionary might nonetheless exist. For words that aren't found in any of the dictionaries searched by the hypothetical checking software, what is the author to do? Is providing a written definition in a glossary sufficient? Exception would have to be made of course for proper nouns, which are words, but for which definitions can't be provided (the name refers to the entity named). > RE User agent issue. There is a user agent component to this in that > the user agent would have to do something with the information provided. > > 3) RE comment regarding pronunciation being in the SC. -- the question > wasnt whether pronunciation was in the SC. You are correct - it was. but > it was only programmatically located - not programmatically determined. > that was the problem. So, if I have a text to speech algorithm for English that converts words into phonemes (this is essentially the first step in speech synthesis) then does this mean that the pronunciation requirement is redundant for all English content? Sometimes the results of the text to speech algorithm will be wrong. In those instances, does the requirement fail to be satisfied? Bear in mind that different text to speech systems make different errors and have different dictionaries of exceptions. What happens to the requirement if the content is written in a language (e.g., an indigenous language) for which there are no text to speech systems? Also, what counts as "programmatically locating" the pronunciation of proper nouns? Does a plausible pronunciation in the natural language of the content suffice, or must it be possible to find a pronunciation in the language from which the name originated? In the latter case how can this be practically achieved? There are serious ambiguities and problems here that demand further attention.
Received on Friday, 31 December 2004 08:01:03 UTC