Meeting minutes
Publishing Accessibility
<duga> AvneeshSingh: are you here?
<duga> We are in the same boat
<jasonjgw> I can't join Zoom from here - I get low-level Web protocol errors. I tried on two machines.
<MichaelC> Hi everyone - zoom seems to have a widespread outage, we´re trying to find alternatives
<MURATA> https://
<duga> To join the video meeting, click this link: https://
<MichaelC> https://
Janina: Our first topic is Adapt!
<wolfgang> I'm constantly thrown out of Google meet, due to network errors
Adapt
janina: We will skip the usual preliminaries, say your name when you speak for the scribe
… first topic is updates on Adapt
Publishing Accessibility
Adapt
janina: Taking advantage of a new zakim feature for creating tables of contents in the meetings
… Adapt is not happening, we thought it would be in CR
… long story short, we ran into a buzzsaw called the TAG
… requires architectural due diligence
… a combination of things rthat do not go together
… we have not widely discussed this, we are going to discuss breaking out the part the TAG likes
<wolfgang> And I have got no sound connection.
janina: AAC support, which is well-researched and can become it's own spec
… we will use a W3C registry to capture the identifiers from Blissymbolics
… we'ren ot giving up the rest, but with this we'll have symbols for publications
janina: Are there any questions regarding Adapt?
Spoken Presentation
janina: We are moving forward on spoken presentation
… making good progress
… we'll have a substantive meeting later this afternoon with the spoken presentation TF
… SSML, emotion markup language (EML), and we need to have tagging for pronounciation
… separate from lang packs needing to provide historical pronunciation
… we should have a useful meeting
… to summarize, we had a good meeting with WHATWG on Tuesday
… we have their blessing to move forward
… this will come up with interlinear publications
… they have no procedural concerns
… some of the ARIA developers are concerned about too many edge cases
… we have reasons, but we are negotiating
… how close are we to CR? Still unclear
… 6 months, but at least this time next year
PaulG: From the pronounciation TF
… I need to catch up with my co-chair after yesterday talks
… big implications after speaking with ARIA, it's not clear if they want to abandon their proposal or move on
… without the red light on the possibility of putting SSML into HTML, we now have 4 possible proposals
… not sure how this impacts us
janina: More good news than bad news
Do we need another a11y horizontal review of EPUB 3.3
The change log is at: https://
janina: Any questions about spoken pronounciation?
Spoken Presentation
MURATA: Spoken HTML has two approaches, multi-attribute or single, what are we choosing?
janina: We'll choose whatever approach seems the best for this
… there's possibly 4 approaches
… we've started with 2
… we will sort that out as quickly as we can
… I think we'll end up with one, that works with ruby and any other use case we have
… we need this to be robust
… not just for publishing, which is big enough
<PaulG> This is the link to option 3: https://
MURATA: I'm also involved in QTI and I agree
<PaulG> SSML in HTML is option 4
janina: Mark Hakkonen is also aware f QTI and it factors in to our disucssions
MURATA: Rule of thumb is, don't use ruby for TTS
Do we need another a11y horizontal review of EPUB 3.3
AvneeshSingh: Procedural topic, horizontal a11y review of EPUB 3.3
The change log is at: https://
AvneeshSingh: do we need another one?
… we had it in fall 2021 before we went to CR
<AvneeshSingh> https://
AvneeshSingh: do we need another, or can we send the change log and you can determine
… what is the best way forward?
janina: I think we can rely on the pre-CR review and look at the change log
… but I don't want to speak for the whole group, but publishing understands accessibility in publishing better than anyone else
AvneeshSingh: Great!
… we will send the change logs via email
Updates for accessibility documents being developed in Publishing CG.
AvneeshSingh: We are working on documents in the PCG, want to update you on what we have done and share knowledge
<AvneeshSingh> https://
AvneeshSingh: version 1 of user experience guide for accessibility metadata
… the industry relies on the metadata, it's displayed on retailer websites
… this guide tells retailers how to display the metadata in a human-readable way
… we have received feedback and will be working on a revision
<AvneeshSingh> https://
AvneeshSingh: we have also worked on synchronizing Schema and ONIX metadata standards
… and extended it to MARC
… the library community is participating, which is great to have
<AvneeshSingh> https://
AvneeshSingh: people from the Marrakesh group as well
… we now have a draft crosswalk of Schema, ONIX, and MARC
… we'll be seeking feeback
<AvneeshSingh> https://
AvneeshSingh: lastly, we are working on guidance for accessibility summary
… it's the only human-readable metadata in the standard
… we need a guidance document for helping publishers write good metadata
… should it summarize all of the metadata or just enhance it
<Judy> https://
AvneeshSingh: if the machine readable metadata contains extended descriptions, a11y summary can provide context
… these are the documents we are working on right now
CharlesL: One clarification, in the last few months, we actually made a normative change
… previously the a11y summary was a MUST
… it is now a SHOULD
… this affects how we are approaching the UX Guide
… the summary is not required
… publishers were pushing back on the required field, a lot of copy pasting
… instead of enhancement
… i.e. a hazard, it just says sound hazard
… now you can say "in the video entitled waves, there is a loud crash at 15s"
GeorgeK: We're seeing good uptake of the a11y metadata, repeating that in the summary is no longer needed
… good news
… with this approach of augmenting the a11y metadata, we are more properly aligned with the schema interpretation
wendyreid: AvneeshSingh mentioned we were getting feedback about the UX guide. Does that include the work being done by EDR lab in France?
AvneeshSingh: Yes
janina: This is very exciting
… a few comments
… I'm going to carry this to WCAG 3, we have been promoting machine-readable metadata
… this gives us a good start to what we want
… either writing our own or extending yours
… for movies, amazon prime, etc
… great base for that
<duga> To join the video meeting, click this link: https://
janina: multiple documents, what are normative or notes?
… can you please clarify
AvneeshSingh: They are CG documents, the UX Guide, the A11y Summary guide, and the crosswalk
janina: Intended to stay as CG reports?
AvneeshSingh: This is the plan right now
… the WG is only for EPUB3, the a11y metadata lives outside just EPUB3, it's generic and useful for everyone
janina: The official language is CG report
… I will definitely carry this info into WCAG 3
AvneeshSingh: That would be amazing
… we have been in discussion with WCAG for years, but we're still working on it
janina: I will carry this forward
… i want to promote it for WCAG3
CharlesL: Just wanted to mention that we have changed the conformance statement to make it human-readable
Interlinear e-texts (e.g. Opera parts)
janina: Let's move on to inter-linear
… I've been thinking about it
… adapt code to our needs
… in operas, what happens often is a duet, or quartet
… they will sing in harmony but with different words
… the soprano is singing about murder
… alto is in love
… different words
… but they have ways to keep in sync
… granular level, bar level
… I'm looking at the Music XML spec
… I'm proposing we look there
… if we strip out the musical notation
… are we left with enough structure to adapt the markup for interlinear
… we'd need to built UAs to use it effectively
… that is my proposal
Lars: It would be neat
… I don't think interlinear work
… but it sounds like have synergies with Sync Media
… we've done work based on the SMIL specification
… synchronize multiple medias
Lars: sync and play multiple things in concert
janina: I should clarify, it's used in academic settings
… studying theology, you may read the new testament in greek and english
… medieval texts in latin
… reading Chaucer, spelling and pronounciation
… you might want synchronization
… one top of the other, or keep phrasing in sync
<wolfgang> s/pronouncitaion/pronunciation/
janina: not playing them both at the same time
Lars: We do have other specs for this, like business process modelling language, a markup language that uses these parallel processes
… that would be something to be inspired by
… sync media could be used for this
… it has no preferences
… any media can be synchonized, textual only
… it supports parallel tracks
wendyreid: This topic is incredibly interesting; I have used the print equivalent of what this looks like. I'm happy to reach out to the EPUB WG and CG; we should discuss with our academic colleagues, who
… may've explored this and be working on it.
… I've read medieval and Shakespearian texts that look great in print, but e-texts probably don't handle this well.
AvneeshSingh: On the same lines as Wendy, can Janina please write up an explainer
… not initially clear to me either, but at this point, the goal is to see if there is any traction
… give us some documentation, we can circulate to through the PCG
janina: I will share some documentation
… excited to explore if there is greater interest
matatk: Thank you everyone for bearing with us
<Bill_Kasdorf> I used the link in Wendy's email to Google Meet but I'm still not getting in
<Bill_Kasdorf> I tried the Zoom link in the calendar and that doesn't work either
<Bill_Kasdorf> Can anybody tell me how to join the meeting virtually?
Bill the meeting is over
Started at 8AM PT
<Bill_Kasdorf> Aargh, I was counting on participating. Had it wrong in my calendar, I guess.
Timed Text
Here is our agenda: https://
Matthew Atkinson
<nigel_> Nigel Megitt, BBC. Co-chair TTWG
SAUR: https://
SAUR == Synchronization Accessibility User Requirements
janina: We want to explore if it's valuable to W3C to turn the data that we unearthed in this document (Steve who's on Zoom is the primary reseracher)
… in to a normative document. The researxh goes back to the 1950s. This is about the toleranxes wrt lag between audio and video in media.
… The envelope varies. Is it useful to W3C to have a normative document that specifies that envelope?
… Captions, descriptions of video: APA could do, but we don't want to do this on our own. We found there's very useful data about the 'primary' resource (the thing you're providing alternatives to)
Janina: and what the toleranxes are for when it's out of sync.
janina: We wouldn't want WCAG 3 to come up with one set of values, and other media groups within W3C to come up with another set of values. Can we agree on toleranxes that we can all live with? Nothing's perfect, things happen between server and destination, but these are the tolerances that
… we expect, and preserve comprehension. We have a Note, but this can't be made normative.
… Should we turn it into something normative?
… Steve: any thoughts?
SteveNoble_: The SAUR attempts to serve what's known in the research. Everything we know
… about issues of synchronization that affect accessibility.
… We look at various types of captions, sign language, AD, ...
… We left a marker in there for XR environment synchromizaton - there's not much known yet, but it's there in case someone wants to provide feedback.
janina: That's APA's question - what are your thoughts? - Is this worth doing?
nigel: It's a great document.
janina: Thanks for your help with it, nigel
nigel: What normative statements of requirement would you write down, and how are you going to test them? Would they be on implementations, or services?
… There's multi-layered complexity. The reality of what the user experiences
… can be very different to that which the provider thinks they're having.
… It can be hard for providers _and_ implemenations to meet the requirements.
… Who do you want to impose requirements on, and how would you test them?
janina: I think on the players. I think there's a mechanism for testing: there's a group within Web
… Performance WG that provides reporting facilities
… which can aggregate data, and take a sample to illustrate the performance.
… The reporting API already exists as a framework. The data would need to be collected by the UA somehow. Talking about distribution of content via the web here.
Andreas: My first question was also: is this a requirement for services, or a technology implementation. Agree with Nigel that it's important and should be at the forefront.
… For me, it reads like an explanation for a short normative constraint (WCAG or something else)
… where you specify threshold values for synchronization. But as Nigel said
… where do you apply this constraint?
… You could make it a requirement for certain technologies, i.e. the technologies need to monitor the synchronization.
… Could be part of the player, browser, HTML5. But can
… get complicated - e.g. live streaming.
… But I think it would help that there be a requirement that through the chain, the requirement is honored.
gkatsev: It sounds like there's multiple levels of requirements.
… E.g. on the player: if captions should appear at a certin time, they appear within a tolerance of that time.
<Zakim> gkatsev, you wanted to say maybe there's two requirements? player-level and generation level?
gkatsev: Could have a requirement that when you change a live video to VOD, these issues are corrected. Live video may be more challenging.
gkatsev: Could have a tolerance for those generating captinos for VOD that ensure the captions show up at the right time. (So: requirements on both the provider and the player.)
nigel: wrt Reporting API for implementations to check they did what they were supposed to do, and when. I like the idea of aggregated, anonymized data sets that reflect
… the choices people make around accessibility. This can help content providers
… (ref Wednesday breakout).
… Would be interested to know if the Reporting API is workable there.
… But I think there's some detail that may be hard to ascertain: how does the implementation
… know that it's doing what you asked it to do on time? It's presumably trying
… to do this already, so how does it know any better?
… Expressing the requirement on the UA in a clear technical way that can't get bypassed/skewed by technical reasons would be hard. Especially as
… we're talking about milliseconds (as opposed to seconds).
… There are alternatives, e.g. with a test resource that you can see, and you can time the captions. That might be an easier one-off way, but
… can't take into account all the complexity of real-life situations.
… It's a hard thing to meausre performance at this accuraccy.
… Would be good if we could.
<Zakim> nigel, you wanted to respond to Janina's suggestion of reporting API
Janina: I don't have an answer to your last question. But can we think about which parts are
… acheiveable. Right now we can flag if this is valuable information. We now have
… in W3C a higher level of document called a Statement that could apply across all of our work
… in this area, and be more visible.
… Someone mentioned live environments. WCAG is more tolerant of live environments, but hasn't required that you have to clean up the timings and captions when you post it into archives.
… It seems to me that WCAG 3 needs to aboslutely require something like that. At least at the producer end
… this is where we could tighten the requirements.
… Still looser requirements for live, but archival should have a higher bar.
Andreas: The benefit of this document is to make a clear statement about the quality of the service.
… After making this statement, you need to analyze the situation and find the parts of the chain that are being pushed. Having constraints like these
… will push technology, and it needs that push.
… AI tech for synchronization could be improved. Captions could be produced independently of the video and a big issue is
… to synchronize their arrival time.
<Zakim> nigel, you wanted to mention current BBC practices about making live captioned content available for VOD
nigel: Andreas makes a good point. Sometimes the upstream chain has a difficult time to know what's "right" for synchronization, and pushing that down to the client isn't necessarily going to solve that.
… To Janina saying we should require that live captions are cleaned up for archival.
… BBC does captioning for 100% of broadcast programmes. 40% of that is done live.
… Currently we do a small amount of processing to turn word-by-word to line-by-line subtitles. There is a lot of cost in processes like these. Delays would still be present.
… We may resubtitle if the programme is to be broadcast again.
… It would be great to meet such a requirement, but can be very hard.
nigel: What advantage would there be in turning the note into a statement?
janina: Heightened visibility - statements are from the whole W3C, not just a WG.
nigel: From a tech perspective, a conversation with service providers and browseres is needed to explore the feasibility of measuring inside the client. Dont't think we can have
… a normative requirement until we're sure we can do this.
… There is another layer in that e.g. web standard captions may not be suffixient; providers may have their own way to provide captions.
… This would just look like DOM mutations to any performance tracing tech.
… (i.e. not captions)
matatk: Seems we have a lot of both encouragement, and things to think about.
nigel: Action plan: find out what can be measured. Ensure tests can be written. Scoping it to technical implementations may be best to start with.
nigel: Need tests in order to have a normative requirement.
nigel: Also important to bear in mind. Some time ago we found queue handlers weren't running quick enough. Raised in a WG. Browser vendors found it was not
… behaving as intended, and fixed it. Now queue entry/exit handlers are firing in ~10ms. This gives us a fighting chance of meeting requirements.
… Practically: how to get the desired end result: do we need a normative doc, or are there better ways?
matatk: Think this covers our agenda. Do you have any questions or specs coming for review?
nigel: Couple of pieces of work in TT.
nigel: One is on measuring subtitling and captioning document rendering complexity.
… So overall we can improve the audience experience by verifying that
… subtitling and captioning docs don't place undue burden
… on players. This is already in the existing docs,
… but we are factoring it out so as to de-duplicate and clean up.
… We have an open source implementation that Pierre has prepared, and we think other people will likely use it.
… They may have a lot of content that will pass (we have at least one validating implementation) and may only have one, but think this is an improvement on status quo. This seems related
… to improcving the audience exeprience a la synchronization.
… Another question: if you miss the caption's window, do you try to show it? How long do you show it for? How do you ever catch up?
… Steve, is that covered?
SteveNoble_: Dont't think we covered how to deal with missed data. Like dropped packets perhaps; don't try to catch up, or you'll be out of synch. Not sure if there's been any
… reserach on this; will see if we can find some.
nigel: Some systems have a max word rate, and speed up delivery, up to that rate, if they miss some.
<Zakim> gkatsev, you wanted to say that the HTML spec says to skip cues in that case
gkatsev: HTML spec says if you miss your chance to queue, you should skip it.
nigel: Thanks for the reminder.
nigel: The other piece of work under development is an interchange format for scripts that includes mixing and AD.
… This has both synchronization and general accessibilty implications.
… DAPT profile of TTML2
pal: Want to encourage anyone who has a library of TTML captions to try the open source implementation of the hypothetical render model (HRM) [the first upcoming peice of work discussed above]
… If you know folks who have catalogs, please encourage them to try it.
janina: I think we're done!
Spoken Presentation
Welcome all! Agenda: https://
*Group introductions*
Specifying age and gender in TTS
Dan: SSML is designed to always speak with some voice (rather than throwing an error/not generating at all).
… Voices are typically generated from recorded human speech. You specify charactersistics and the platform selects a suitable voice pack. You can also
… specify orders of characteristics, so fallbacks can be selected.
… Available voice packs are _not_ under the control of the SSML author.
<burn> https://
Dan: ^ describes the <voice> element. This can include desired features for the voice to use, or control behavior.
… You can also ask for a specific voice, but that's processor-specific.
… You can specify the language the voice is to use to speak.
… We found that some people preferred to get a voice that speaks the wrong language (=> accent) but had other attributes that matched. This surprised us.
… There are two important failure modes. We started with lang failure (in 1.0).
… What is the processor supposed to do? Some voices may be able to cope at least partly with some langs but not others.
… setting onlangfailure to change the voice results in the voice changing. Other options available.
… We then added onvoiceailure to choose what to do if a suitable voice doesn't exist.
… This was an overiew of the design choices of SSML.
NeilS: Do you know how well the processors support the SSML fallbacks on language?
Dan: Debbie may have a better idea; it's like with browsers, where you don't control which fallbacks are avaialble.
Debbie: A lot of the big names (browsers and conversational application toolkits) say they support SSML.
… They support some elements; nobody comes close to support everything.
NeilS: I was wondering more about the fallbacks.
Debbie: I don't think they're particularly widely supported: you may get to pick the language but the
… fallbacks are very limited. They won't crash if
… you ask for something unsupported, but you'll just get the default.
Dan: It is a frustration; we built this into the spec but it's difficult to get conformance. You, the users, could drive this. If you say you require it, it's already specified.
Specifying emotion in TTS
janina: Background: the Web Accessibiliy for Children CG and some of our COGA colleagues
… informed us that earlier support, with voices that the child can identify with, can help significantly with communication for the rest of life.
Debbie: We had three use cases for EML: (1) representing the recognition of emotion; (2) rendering emotion in TTS; (3) annotation of emotion for research purposes.
… EML is neutral wrt recognising and rendering emotion.
… From what I see in the research community, recognition is difficult and innacurate. I think we're less interested in recogntion; more rendering.
… Recognition researxch tends to be about 60% accurate only.
<ddahl> https://
Debbie: The <speak> SSML element refs the EML namespace.
… We have an <emotion> element within the SSML.
… The topic of emotion is not cut and dried in the research community. Different vocabularies of emotions, numbers of, and ways of describing emotions.
… We can reference category sets of emotions from the EML.
… a 'worried' value of 0.4, e.g.
… The system responds with 'do you need help?'
… You can also mix emotions.
… The question is: how well will the TTS respect that markup into something you could conceieve as being a mix of those emotions?
<Zakim> burn, you wanted to say something else about platform support for voice selection once Debbie is done
NeilS: This introduces new tags rather than attributes on existing elements; was there a reason for this?
Debbie: We wanted to fit with the existing SSML structure.
NeilS: Why is the "Do you need help?" not a child of the EML tag?
Debbie: The reason for not having another tag around the text is to fit in with SSML structure.
Dan: You could add attributes from another namespace to an SSML element, but if you look at the structure, a variety of things could be placed here from an EML point of view, so an attribute wouldn't cover it.
Debbie: I wasn't involved in these particular decisions, but the overall desire was to be neutral between recognition and rendering, so didn't want to depend on the SSML structure.
Debbie: I'm not aware of implementations of EML outside of research. There's not been a push for it, but recognizing the importance of
… it for some of these use cases, e.g. children
… then this may help.
nigel: Are there any particular emotional vocabularies that have support?
Debbie: Research only. One is "the big six". Some break down emotions into combinations of different characteristics, though they're almost never used.
<Zakim> nigel, you wanted to ask if there are published emotion vocabularies that have widespread support
Debbie: When we want to recognize emotions, we probably only want 4 or 5; 15 is probably getting too fine-grained.
… Same for producing emotions.
… Most of the production of emotions is by actors, who are trained to do this. It's not easy, even for humans, to produce emotions on demand.
NeilS: Any examples of computer voices that produce emotions?
Debbie: Mary, open source TTS by an EU reserach project.
<ddahl> semaine
<ddahl> Mary TTS project
Irfan_Ali: The Children CG sent us several use cases particular to the TTS voice style amd emotions.
<Irfan_Ali> https://
PaulG: Looking at EML within SSML, I see the namespaces. In our talks with WHATWG about allowing SSML in HTML, namespaces won't be allowed. Thinking of options.
… One could be voices with different emotion characteristics.
… Another is bringing in attributes to SSML that convey the info from EML.
… If neither is possible, we would have to do SSML first, then later add EML.
… I don't see how we could blend them at the moment.
janina: WHATWG were willing to discuss the idea of including SSML.
PaulG: We have many different approaches - SSML, multi- and single-attribute approaches and ARIA's approach.
… Our first choice was SSML as a first-class object, but if we did this, we'd be unable to bring in EML right away. If we use a
… different approach, maybe we can do both at the same time,
Dan: When VoiceXML was created, SSML and EML were structured to be embedded within it.
… Instead of changing the SSML standard, you could take the EML elements added in without the namespace.
… Worth checking; would be surprised if that was not doable. Need to check for conflicts.
… Returing to earlier and adoption of voice seletion fallbacks.
… Big names weren't particularly interested in trying to allow someone to write one piece of SSML that'd work everywhere regardless of which voices are installed.
… @name allows you to specify a name, but you don't know which are available.
… If that's acceptable to you based on where you want to use this, then that is an option
Dan: It's not generic, which is not great, but may be an option.
NeilS: So selection across gender and age is not reliable across processors?
Dan: It's not clear how well that's implemented. It's like fonts.
nigel: Is there a font=like equivalent for voices?
Dan: The fallbacks are really depedant on exactly which voices are installed.
Dan: There is no standard format for voices (not like fonts).
Debbie: Thinking about cross-browser respecting of voices. Seems like there'll need to be a lot of thinking on this. Especially if children are the audience, and if the fallbacks are inappropriate or just ignored. That needs a lot of thought.
Lionel: *suggests a different error state for juvenille audiene*
PaulG: *is concerned* (as group)
PaulG: What do we do about sp, mark and sub?
<burn> SSML supports that option of "no voice", if you decide you like it ;)
Historical Pronunciation
<PaulG> potential conflicts: what do we do about s/p/mark/sub?
janina: Much of our consumer base is academia: edu publishing and assessment.
… Languages of the past are spelled and pronounced differently. Chaucer, Shakespeare, Beowulf, Ελληνιστική Κοινή...
… Alos different Latin variants used, e.g. in religious texts.
… Different variants of Greek used in antiquity. We may not even
… have lang packs for them yet, but if poeple study the texts and
… want them pronounced, or to read them in Braille, we need to be
… able to identify those languages.
addison: There are tags for some languages, such as middle/old English, and Greek. There may be a lack of granularity.
… This could be supported by adding sub-tags to the BCP47 registry as-needed.
… For identifying languages, tags are the only game in town.
… If needs arise that go beyond what a language tag can do, BCP47 could be expanded by adding am extension to a language tag. E.g. there's no language tag for historic Rome.
… As you think about other speech reqs, you may build and extension for some of those at the same time, in the same way that Unicode locales has an extension (which has a particular syntax).
… E.g. you can specify language, characters, dates, and sorting in specific detail.
<addison> https://
<PaulG> https://
PaulG: The IANA registry is where these language sub-tags live, and it would require registration of a new language sub tag there?
addison: Yes
<addison> https://
PaulG: ^ link to Old English
PaulG: in HTML the best practice is to include the language and only the language, not the locale, in the @lang attribute.
addison: These values are locales.
PaulG: We're trying to provide author controlled presentation for specific sub-content, and we have @lang from both SSML and HTML, and that's where we can put these additinoal details?
addison: Yes.
addison: One of our goals as a WG is to get those things to stay plumbed everywhere.
… and transported through APIs/JSON/...
… and these could be spoken, because you have the language info for this content, and not just text.
PaulG: How are conflicts resolved (as asked above, across content in HTML, SSML, sub, mark, ...) ?
Mark: How many of the options have working implementations?
<addison> (example of a language tag extension: https://
janina: We asked WHATWG about SSML and (for other work) attributes, and they're open to discussion.
NeilS: But we need to decide one or the other.
PaulG: Yes, need to counter authoring confusion. SVG allows for foreign objects; would that be forced into speech too? What about a <p> tag in the speech content?
Dan: Namespaces by namespaces or prefixes.
Dan: There is something called a fragment, where you want to embed into something else.
… Some of what we designed was designed to be dropable. Where there is bare text, you could have SSML, and you can do something similar with EML.
PaulG: There could be further complications with strikethrough etc.