Meeting minutes
Wendy Reid: We wanted to discuss extended descriptions
Extended Descriptions
<Avneesh Singh> https://
Avneesh Singh: We have your feedback from extended desc
<Avneesh Singh> https://
Avneesh Singh: We have improved the doc with your feedback
… there is a list of things that came from APA and our response
Matthew Atkinson: We have people that want to make comments that aren't here today
… I see many things are addressed, I am just now seeing the list of closed things, very helpful
… Is it true anything that may affect the layout can't be done, that is why you can't do e.g. details?
Avneesh Singh: Yes. toggling open something completely changes the page layout that you can't really do
… so it is fine for the web, but it is very broken for books
Matthew Atkinson: One big question is plans for future revisions
… is there a plan to change the way details work in reading systems? E.g. by using a popup
… is there a way to use standard elements but with different rendering?
Avneesh Singh: People want to make content that works in most reading systems. So we would need a critical mass of reading systems
Wendy Reid: The best other example is footnotes
… reding systems agreed this was a good idea
… but getting there was fraut
… setting up the logic was hard. Details might be easier
… it would have to be very well defined
… This can also cause a repaint on eink
Brady Duga: To address popups, it's not a silver bullet, it may not work
… most reading systems discard the HTML/CSS of what they popup, since it's not well defined in HTML what to do with a fragment of a document
… most reading systems don't bring up a full new webview in popups, almost all of them discard the styling
… the popups for footnotes and definitions don't often have styling
… reading systems won't want to create a new webview but that's my experience
George Kerscher: If someone wanted to do this as a popup, you would still need to know that this was an extended desc vs some other kind of content
… so you still need the semantics
Avneesh Singh: Agreed
… you still need to be able to identify it
Charles LaPierre: You can also have those details expanded by default which means they may auto pop-up
Matthew Atkinson: We will keep the slot at tpac and provide more feedback there
… In terms of semantics, you would still need it, but maybe there is another way
… You could use aria described-by
… Maybe we can think of some lower level thing. And maybe we can look more at the footnotes implementations
George Kerscher: One thing we want to consider is there are a lot of people with disabilities who do not use screen readers
… Having a visible element that shows this is an extended desc is helpful. Would be great to have an standard icon for extended desc
<Matthew Atkinson> +1 George Kerscher - that's a great point; need to ensure it's accessible for a wide range of people. Interesting suggestion about icons, too.
Charles LaPierre: I am thinking if the ongoing symbols work, maybe there is something in there we could use
Sync Media
<Avneesh Singh> w3c-cg/
Avneesh Singh: We also want to discuss sync media
<Avneesh Singh> https://
Avneesh Singh: we need to align on long term scopes
… some things will be in the 3.5 charter
… Some tools esp on windows loose the alignment of playback
… so the [???] should always be visible to the assistive tech
… We also need different sync levels to current media overlays
… we want sync at various levels
… word, para, etc
… this is the high level view if what want from media CG
… so we want to mention this work has started
<Matthew Atkinson> Digital Accessibility User Requirements: https://
<Matthew Atkinson> Specifically, synchronization: https://
Matthew Atkinson: That sounds great. Some people at APA have suggested similar work
… There are lots of docs on synch (see links)
… this would be of use on the web as well
… there is also a markup language we are considering joint work on
… it would be good to explore what has been done and where we can share work
… I will ask around in that group as well
… aria may also need to be involved
… we can discuss over the next few weeks
Avneesh Singh: Marissa already did some tests with webvtt
… in 3.5 we may need to do it with SMIL
… Marissa may be able to connect at TPAC
Charles LaPierre: We are currently collecting requirements in Marissa's group, so that work is already ongoing
… so these will all become technical requirements eventually
George Kerscher: Just so you know, the text reading level of native sign language users is very low
… that has gotten better, but it is important to understand
… sign language is not a 1:1 to text
… so aligning a video with the text is tricky
Shinya Takami: Is the sync media group for speech or just recorded media?
Avneesh Singh: Recorded media is the current focus
Dale Rogers: Back in 2017 we had a poet and we wanted to approach their work from different points of view
… so we have a web page that had different ways of consuming the poetry
… when I think about how people consume content, from a design viewpoint is it better to specify a different way to display content
<Dale Rogers> https://
Dale Rogers: I will put up a link about what i am talking about
Matthew Atkinson: That is an important point, I will look at the research project
… I think sometimes we want sync, and sometimes we want something totally different
AI Content
Avneesh Singh: last topic - framework triggered
… it should be identifiable in the content and in machine readable MD
… we realized this was a generic problem with a generic solution for epub
… and maybe the same is true for the web
… this is something were we might want to work together
Matthew Atkinson: First, whatever is the law we have to facilitate
… there was a discussion about this precise issue. Should we have an indicator from text that this was generated by a machine
… What is the most important thing about the alt text? Not who made it, but whether it is good
… telling a user that it was AI generated doesn't really help the end use
… r
… There is a genuine problem with e.g. hallucinations, but I am not sure this is the solution
… I will find the issue so you can add comments
… You might suggest a minimum level of conformance is a machine made it, then a level up is a human reviewer, then a human editor, etc
… so it may be useful when making it to know this info
… and if there is a legal requirement then that is another thing
… I will find the original thread and see where it goes
Charles LaPierre: I Knowing that it was AI generated is one thing, but which model may also be a huge difference
… so knowing the model in use can help trustability
JeffAdams-UN: I am thinking about this as an author., and this shows up in books a lot
… so maybe this goes into the metadata. Thinking of info asked about books on upload, Amazon has a list of levels
… so readers could decide to surface this, and it won't interfere with the content itself since it is in the broader book metadata
Matthew Atkinson: This is a great discussion. With no animosity let me flip the question
… you would have to be pretty involved in this stuff. But won't there be an explosion of metadata?
… and for a human, should we ask how busy they were that day? Quality is a general problem
Dale Rogers: The alliance for independent authors discusses this alot, including AI generated vs AI assisted
<Matthew Atkinson> +1 Dale Rogers - using it at authoring time as opposed to runtime changes the game
Dale Rogers: so I may create a description via AI for a comic, but I review it and edit. So that is assisted, and good to know
Matt Garrish: The idea from JeffAdams-UN is interesting, just adding it to metadata, and schema.org may have something to help
… not sure what EAA requires
Janina Sajka: We discussed in the research questions TF a lot
… pushing back on Charles LaPierre I am not convinced all that info will help. The quality depends on how you prompted and followed up
… I think assisted vs generated is really useful
… who cares what tools a human used, as long as it was human generated (e.g. spell checkers)
… So what we want to know was there a human in the loop?
Matthew Atkinson: Of 3 things on APA, there is 3.5 , pronunciations, and symbols stuff. we have 7 minutes left, do you want a symbols demo?
… we can also do a demo at tpac
JeffAdams-UN: One quick though, we discussed EAA. There are also fingerprinting requirements
… a sentence may not get a fingerprint, but longer text would, and we need to consider that
Wendy Reid: There has also been some drama around bad AI claims
Symbols Demo
Matthew Atkinson: Background - this is a long running project
… about 10 years at this point
… some people use AAC symbols to understand text
… symbols help people understand what the text is about
… you can't just add them to the page, as there are a lot of symbols
… We have teemed up with Bliss for their dictionary of symbols (we don't use their full language)
… [demo ongoing]
… can choose which set of symbols to use
… this uses ruby and fonts
… so each symbol is the unicode value for it in the Bliss language
… we are hoping this is a sound approach
Charles LaPierre: That is awesome congrats
… I am happy to see this using ruby and unicode
<Matthew Atkinson> Special acknowedgement to Russell Galvin from Bliss who made the fonts for the demo.
Avneesh Singh: There are no decisions being made at the tpac meeting, it is just for discussion