- From: Al Gilman <Alfred.S.Gilman@IEEE.org>
- Date: Fri, 24 Feb 2006 15:25:12 -0500
- To: www-voice@w3.org
- Cc: wai-liaison@w3.org
* background: I was discussing with Alistair Miles of the Sematic Web -- the people who bring us SKOS, the potential use of the terms in that lexicon in meeting the objectives of WCAG2, Success Criterion 3.1.3. http://www.w3.org/WAI/GL/WCAG20/guidelines.html#meaning Alistair raised the entirely valid point that a thesaurus in SKOS does not constitute a sufficient knowledge base for auto-translation of free text. I spun out a use case to explain to him that there were niches in the support of people with disabilities where, with a little [e.g. SKOS] help, automatic translation might have a real role to play. Since the example bears on the questions of "PLS and meaning" I wanted to share this here as grist for our discussions on Thursday and through the Last Call. Use case: Let me isolate a disability use case where annotating the intended sense of selected terms (by exception, following the WCAG 2 success critierion -- the sense is not the dominant sense of that term or is exotic, uncommon and likely to be unknown) is likely to make a significant improvement in auto-translation, and the auto-translation is likely to fill a market niche. This has to do with the auto-translation of SRGS grammars that define the catch-phrase structure for voice input in VoiceXML applications, translating these to to a sign-language gesture grammar for use in a gesture-recognition input module. This would be used in an adaptive binding of the VoiceXML application for use by those who are Deaf, communicate in sign language as their first language, and have speech that the speech recognizer does not perform well on. Just as many Deaf people today carry text message enabled mobile devices, and a few are busily engaging in video chat where they can get broadband connections to the Internet, the sign language or culturally-Deaf community will be likely to take to the gesture-recognition-enabled mobile devices emerging on the market. But still sign is a different natural language and requires translation. The [SKOS or similar] markup on terms in the voice grammar is not used to provide all the knowledge used in this translation. It just cues the sense to be translated when the sense is not what the translation software would be likely to assume. Just as the use of SSML in the production of audio books by RNIB needs the Pronunciation Lexicon Specification to work around pronunciation errors in omnibus text-to speech algorithms, so annotations as to sense would give us the means to work around and touch up meaning errors in translations of speech input grammars to sign. The terms that need to be clarified can be identified without having the sign-language translation program to work with, though. Basic semiotic and language-control statistics can tell us when a sense is not obvious and should be made explicit in the markup. The reason that there is a market opportunity for auto-translation here, is that the Deaf, like the Hakka in China and the Romany in Europe, are a minority everywhere. And like users with disabilities everywhere, they have a need for a functional user experience even where a comparable user experience would not be competitive in the market for the attention of the Temporarily Able Bodied. Al
Received on Friday, 24 February 2006 20:25:25 UTC