Meeting minutes
<PaulG> matak: what is the path forward? Rough plan: get the documents in order (deprecations), snapshot, Irfan and others to invite folks interested in continuing the work (either through TF, or CG), matak to find out if this TF is required
<PaulG> ...This TF was proposed by Mark and is not mandated by funding
<PaulG> ...We can run it, coordinate with a CG, or pause it
<PaulG> ...We need to talk about realistic time scales for getting the docs together
<PaulG> ...For the data- attributes, we could run a TPAC breakout to discuss the experiences and focus on the gap to attract some standards folks to work on it
<PaulG> ...janina can come to one of these calls to discuss how AI might change our path
<PaulG> ...and she will talk to Mark about the data- implementation soon
<PaulG> IrfanA: we need to streamline the process and make some progress. Do we clean up or write a new explainer?
PaulG: When we had the explainer before, we kind of leaked the solution through the explainer.
PaulG: This resulted in some pushback.
PaulG: The idea was to strop out implementation details and bring back a problem-space-oriented explainer.
PaulG: Including priorities, and the technologies this touches (but they don't solve it).
PaulG: We can use that as a means to entice contributions or report status.
PaulG: Maybe we'll get better ideas, or pieces of problems, such as James Craig, and the ARIA diectionary approach.
PaulG: So we could work on problems, hoping for a holistic outcome, but not pushing fro it as much.
PaulG: Re AI, how that affects the work, it's something I've been thinking about, but haven't had conclusions yet. There are local models that could be trained on speech-to-text. If they were trained on pronuniciation that could be built into the authoring experience. But voice packs are limited and hard to make. Maybe we could have AI generate a
voice pack. Could we spec that?
PaulG: Having fresh perspectives, once people are introduced to the problem, and recognize it as one (that was one of our problems - demonstrating the problem to people)...
PaulG: ...having examples 'Here's some HTML, and how it's read by AT' - the output of AT must be deterministic (that's a key word to use).
PaulG: So then we can add the annotations that will allow it to work. At which point, people will ask about <ruby> - so we add that to the examples, but it'll still break - then they'll ask what about training AT on <ruby> but that still has problems. And that's about as far as I get; I thought we gave them our best. We were told 'no' to
first-class objects (SSML).
PaulG: I still think custom elements is a viable solution, because they don't really render. But there's not a current path for standardization of those. Open UI is talking about some of this stuff.
PaulG: aria-* was baked in. But don't think we can standardize custom elements (could we have ssml-* as a reserved space?)
PaulG: Does anyone want that?
<PaulG> matatk: to Irfan, do you know the status of the data- approach and how to find out? Also, Mark will travel for the Summer. What's our timeframe to get the documents in shape?
<PaulG> IrfanA: I'm not sure the status. I was planning to talk to Mark this weekend but it didn't happen. So, I'll reschedule.
<PaulG> ...I know several people who joined our meetings before would still be interested.
<PaulG> ...E/DPUB people
PaulG: This aligns with what I was proposing: paint the picture that there is no standard solution to this - only a custom one.
<PaulG> https://
PaulG: That's what I wrote some time ago in response to the ARIA proposal. The doc is still there.
PaulG: It might be helpful to run this through one of those 'council of experts' models to get all the objections in one place.
PaulG: People wanted a one-pager rather than lots of different documents.
PaulG: I see this as a concern now that we need to address.
<PaulG> matatk: we decided to split the use cases from the technical gap analysis
<PaulG> ...update the pointers to documents (and handle the deprecations)
<PaulG> ...when we get to proposing something, we put together the one-pager
<PaulG> ...for now: 2 documents