Meeting minutes
<alastairc> Revisit Assertions https://
Introductions and Announcements
AC: Any new members?
RM: Just a reminder that next week we are holding a 2 day session, and it will replace all ag related meetings.
WCAG 2 proposed changes https://lists.w3.org/Archives/Public/w3c-wai-gl/2026AprJun/0012.html
AC: the meetings have been added to the calandar.
PL: you still have until 11th of May to on the proposed changes
… Two errata, one very, very minor, one quite substantive.
<Patrick_H_Lauke> The emailthat went out https://
<alastairc> Consistency: use "touchscreen" rather than "touch screen" (normative)
<alastairc> #5038 w3c/
PL: which are the ones that we have been working on putting them aside for a potential WCAG 2.3, but both of those have kind of.
… kicked off some discussion in the pull requests themselves, which I thought might be more appropriate to have with a wider group about what is or isn't appropriate.
<alastairc> Update normative wording of 4.1.3 and definition of "status message"
<alastairc> #4952 w3c/
AC: one character change, adding a space in touchscreen.
AC: And the second one was more substantive.
… Whenever the task force comes up with a normative update e can consider it for ertata
… any things like that will come through a CFC process.
WCAG 3 introduction text https://www.w3.org/wbs/35422/about_26/?login=
awk: is this a fyi or a question?
AC: a FYI
<Patrick_H_Lauke> it was intended as an answer to "why are we doing this? is this even an erratum?" questions in the PRs
<Rachael> About proposal https://
RM: so I went through all the comments, and I've pulled them in to a revision that you have the link for.
<Rachael> The WCAG 3 guidelines address the accessibility of user facing digital content on a wide variety of devices such as desktops, laptops, tablets, phones, and wearables.
<nat_tarnoff> Does this language cover things like audio only devices?
<alastairc> "such as" it not limiting, it's just trying to give examples
<Patrick_H_Lauke> +1 for that change
<Charles> user facing should be user-facing
RM: big concern people had with the word devices as written. Other suggestions was just adding the word other before devices.
<nat_tarnoff> +1 to alastairc as a response to my comment.
(RM goes though document)
<AWK> "for example" sounds good to me
Gregg: English language is such that not everybody knows what such as means.
<LoriO> +1 for example
<kirkwood> rather than “devices” say “other devices”
<Patrick_H_Lauke> maybe even add an em or en dash in front of "including"
<bbailey> +1 for "including but not limited to"
Gregg: could say a wide variety of devices, including but not limited to,
<alastairc> +1 for example, simpler language
<nat_tarnoff> +1 to not limited to
<BrianE> +1 "including but not limited to"
<joryc> +1 for "including but not limited to"
<Heather> +1 including but not limited to
<Charles> +1 to not limited
<Francis_Storr> +1 for example
<AWK> +1 for example, simpler
<GreggVan> +1 not limited to
<Patrick_H_Lauke> 0
<Ben_Tillyer> 0
<scott> 0
<ShawnT> +1 for "including but not limited to"
<Jennie_Delisi> 0
<jtoles> 0
<maryjom> 0
<Adam> +1 for example, plainer
<stevef> +1 for example
<LenB> 0
<Makoto_U> 0
<CClaire> 0
<Eloisa> 0
<Francis_Storr> nope
<Patrick_H_Lauke> (it will look a bit heavy-handed having that construct in both sentences perhaps)
Steve: I would object, for example is simpler
Gregg: It has caused problems in the past.
<nat_tarnoff> So is this text meant to be normative?
Steve: Is this a normative statement?
<bbailey> +1 to gregg's experience that "for example" is not explicit enough in practice
<ShawnT> +1 to Julie's comment
Julie: could be agree to this for now and come back to it later?
JK: Just quickly, I think you said it correctly, where it said, and other devices, but...
It reads just as devices, and
AC: it is informative.
<alastairc> https://
<nat_tarnoff> +1 to stevef comment as informative.
Steve: It shouldn't be taken by...
by anybody as... as being, you know, normative in the case of it.
<AWK> Just asked google and apparently using "for example" has caused problems - not all judges are well-versed in the English language apparently.
Ben: I think I personally like Julie, was it Julie's comment of pick one and then revisit later.
<AWK> But still +1 for casual language for informative content. We've already read this paragraph more than anyone outside the group ever will
Gregg: plain language always does not mean fewer words. It means easier to understand and harder to misunderstand.
RM: l've put in 3 different versions that I believe now expand it and make it clear.
<Charles> opinion: not a fan of “etc.” even in informative text. it has several adopted meanings.
<nat_tarnoff> Number three is not clear due to use of "etc."
<Rachael> The WCAG 3 guidelines address the accessibility of user- facing digital content on a wide variety of devices including, but not limited to, desktops, laptops, tablets, phones, and wearables.
<stevef> +1 to 2nd
<Rachael> 2. The WCAG 3 guidelines address the accessibility of user- facing digital content on a wide variety of devices, for example desktops, laptops, tablets, phones, and other devices.
<alastairc> no objection, prefer 2nd as simpler.
<GreggVan> 0
<nat_tarnoff> no objection. As it is informational I lean 2.
Kevin: In the new style guide that WAI is preparing, we discourage "etc." and the suggested substitute for it _is_ "such as"
<maryjom> 0
<julierawe> +1
<jkatherman> 2
<ShawnT> 2
<GN015> 1
<kirkwood> 2
<Heather> 1
<julierawe> 1
<Eloisa> +1
<erinevans> 1
<BrianE> 1
<joryc> 1
<Makoto_U> 2
<bbailey> 1
<Charles> 1
<Detlev> don't care
<Jennie_Delisi> 1
<Francis_Storr> 2
<Azlan> 0
<atya> 1
<scott> 0
<LauraM> 0
<Monica> 0
<Ben_Tillyer> 0
<LenB> 1
<AWK> 1
<Adam> 2
<Gez> 0
<jtoles> 1
<filippo-zorzi> 1
<bbailey> +1 to avoiding "guidelines" because WCAG2 using "guidelines" three different ways has sometimes caused mild confusion
<GN015> The 'G' in "WCAG" stands for "Guidelines", doesn't it?
CH: Just a quick thought to Greg's comment about reusing the word guidelines. The second instance could say guidance.
<bbailey> +1 for "guidance" rather than "guidelines"
<Adam> “maximally accessible”?
CH: where it says: as applicable as possible. I'm curious, to whom is it possible, and do we use that language anywhere else?
<Adam> er, “maximally applicable”
CH: what is the limitation of what's possible?
<kevin> "as applicable as possible" is not in use in WCAG 3.0
RM: Mary Jo raised the point that provisions is a legal language, and maybe we don't want to put that here, nor is it consistent, so l've just replaced provisions with guidelines, so everything is consistent throughout.
<Detlev> I disagree with Gregg's take as to the meaning
<kirkwood> including but not limited to ?
Gregg: making it as wide as possible. If you take off the as possible, and then you put including those two things, it means that it's only those two things and absolutely nothing else, because it says not It just says there's a wide range, including these two things, period, and there's nothing there telling you that it's anything more than that.
<Heather> suggestion: WCAG 3.0 provides a broadened, comprehensive a set of provisions for web technologies.
<Charles> note: “as applicable as possible” occurs in the prior section for types of content and then again as web and non web technologies.
Ken: On the topic of the WCAG 3 guidelines wondering whether we even need the word. I mean, the suggestion of guidance rather than guidelines also works
<kevin> +1 to Rachael
<Zakim> GreggVan, you wanted to say "agree not
<bbailey> one distinction I have seen is "where practical" or "where feasible" (with the later being the higher bar)
LO: RE: applicable as possible is not in use in WCAG
3.0. But since we're wordsmithing, does this mean, in general, that if it's not already in use in 3.0, we can't use it? Or what... what exactly does that mean?
RM: I don't think it actually is relevant. We are writing this document and revising it, so. If we wanted to compare it and see how we had used it elsewhere as part of the conversation, that's helpful, but I don't think it stops us from using a phrase if it's not elsewhere.
<julierawe> Suggested rewrite to avoid "as possible" in either sentence: "WCAG 3.0 aims to cover web technologies thoroughly. It’s also written so it can apply to many non-web technologies—including but not limited to non-web documents native desktop and mobile software—but it doesn’t fully cover those areas yet."
Gregg: I still think a set of provisions works.
<Ben_Tillyer> My understanding is that comprehensive means "all, or almost all, the items". A quick check of Oxford English Dictionary backs that up
<nat_tarnoff> "Provides provisions" sounds awkward.
<kevin> https://
<AWK> WCAG 3.0 provides guidance for web technologies that is as comprehensive as possible.
<GreggVan> +1
RM: I believe we have a definition, at least working somewhere for provisions that states that it is what we're using when we talk about requirements and assertions.
<AWK> either way
<alastairc> no objection to either
RM: we could link to that.
<GreggVan> +1 to rachael
<AWK> ready to let the editors edit
<GreggVan> grin
RM: next change that was suggested was getting rid of the word platforms, because some people are going to find that complex.
<Charles> +1 to oxford comma
LO: shouldn't that be...
Documents, comma, remove the AND.
Revisit Assertions https://docs.google.com/presentation/d/1qWuFM3fFgC_e1Jik05Os11O0Rl86HLDXu9dolwyWWtc/edit?slide=id.g3d460567fce_0_6#slide=id.g3d460567fce_0_6
RM: does anyone have any concerns with us moving forward with the content we have as written?
(None)
AC: Coming back to assertions.
… assertions is a new thing in WCAG 3.
Where we're asking people to say they have done a thing, usually a process,
<Zakim> bbailey, you wanted to ask for clarification on first bullet
AC: We have had some responses (goes though slide 24)
… like a success criteria or a requirement where you're saying, on this page, it passes this requirement, so that would be an outcome. whereas if you're saying, we have done this process, that process doesn't necessarily result in any particular outcome.
AC: Goes over slide 25
<kirkwood> assertions speak to LOE rather than compliance: for legal protection
<alastairc> LOE?
AC: we need to make sure we're differentiating these sufficiently. they're not intended to be compliance guarantees.
<AWK> (level of effort)
<alastairc> thanks
<kirkwood> it’s a legal backstop (as Gregg is saying) as well
<kirkwood> +1 to Gregg
gregg: it's easier to make process than outcome, correct?
On the flip side, We were asked by the consumer side, to not make process. It is a make-outcome, I should say, because There's a long history of process.
<AWK> On the flip side the flood of nuisance lawsuits would likely be reduced if it required discovery (work) to prove non-conformance.
Gregg: The point is, that are we making guidelines? that some companies cannot meet. Not, you know, that they're hard to meet, or whatever else. But they legally are prevented.
AC: We're not requiring anyone to make a conformance statement.
<shadi> [for the record, the feedback from an e-commerce company was not from Amazon]
<kirkwood> it depends on the assertion. and the LOE (“burden”)
<bbailey> +1 to write what we can to reduce nuisance lawsuits
AC: the scoping bit, I find really interesting and important.
really interesting and important.
… I don't think they should or have to apply to whole organisations.
<kirkwood> where would a training assertion sit then?
SO: at Microsoft, there are some product teams that are basically teams within teams.
<GN015> Please note that for trainings, numbers of completion may not be tracked in some countries (like Germany).
AC: guess my question then would be in order to make the assertion, would you know Would whoever's trying to make a claim be able to find that out?
<Eloisa> Wouldn't it fall on you, the business, to require an ACR from the third party vendor and assess risk yourself?
GP: How do I know, is an assertion made by someone else outside the company valid for my current company?
Because I don't have control on the third party that is creating the website for me.
<kirkwood> is “we only use WCAG 2.1 compliant software?” an assetion emample?
<LoriO> giacomo +1
<Eloisa> I believe it's the company that requires it
GP: who is making the assertion? Is it the third party?
Is it the company that received the product from the third party?
AC: how is that different to now? So if they provide a product, are they providing WCAG conformance information about that product?
Scribe change?
LO: having this assertion puts undue
pressure on a company.
GreggVan: Giacomo and Lori have good points — we've run into instances where someone contracted someone else to deliver the website, once the company accepts it, it loses all ability to take action against their contractor.
GreggVan: If I were the lawyer, I wouldn't want to make assertions that happened in that company because I can make assertions about the provenance of what I have, check it for accessibility, but any processes that went into it at another company, I have no recourse against it.
<kirkwood> “process assertion” “training assertion” “complinace assertion”
GreggVan: Making an assertion about anything is something companies will be able or not want to do, and that's a problem — you can't possibly pass something you cannot do
<Zakim> alastairc, you wanted to comment on fluid teams
<kevin> Health and safety, lifting stuff!
alastairc: Lori's comment about fluid teams — some organisations have to track certain trainings like security training, that's monitored even with contractors. There are things we have to track, regardless. We keep those constrained to things that people should be tracking anyway.
<Zakim> Rachael, you wanted to say chair hat off can we divide types of assertions into things that affect content and things that affect people
alastairc: Some of these points come back to being careful about which ones to include and how they are scoped. Going back to scoping a conformance claim, it will help with that but keep an eye out on things where that won't work.
Rachael: There's a difference bet. assertions that affect an org or the people within an organisation like training, and assertions that are directly around things done to content then delivered like plain language review was conducted.
Rachael: I still hope to have assertions that affect content directly at a core level, but recognise the challenges there.
kirkwood: Concerned we're thinking of assertions a different way. You can make assertion that you meet the guideline, done training, a lot different ways — going above and beyond just meeting the guidelines is asserting a process. I think we should be specific about what these assertions are talking about.
alastairc: The assertions aren't on the topic of asserting that you're meeting something. If you look at an ACR, it's an assertion whether you've met guidelines or not on a certain scope. But what we're talking about is process-based, best example is Rachael's on a plain language review.
kirkwood: What about training?
<GreggVan> I think alastair that what you are referring to are training that everyone has to do (e.g. safety, harassement, security) ALL employees need these trainings so not a problem if they move in an out of tasks that surround accessible design or design. But we are talkng about task specific training that you only need for this specific task in the company - where people fly in and out without special training for the task.
alastairc: That has floated as an idea, along the lines of whether authors have training on alt text as an example. They were topic-specific.
kirkwood: From a legal perspective, that's how we meet legal requirements by asserting.
<AWK> Assert that you provide a way for user-generated content to include alternative text for images. Doesn't mean the alt IS provided or that it is any good, but may be a reasonable approach.
GN: In some companies it might be possible to track who are involved in a specific product or had set of trainings, some countries aren't even allowed to track.
<bbailey> "we have a style guide (which includes accessibility)" is an assertion
<kirkwood> interesting point!
GN: It's not possible in countries like Germany, we may not push companies into being able to state something that could violate local laws.
alastairc: Some highlight training as part of internal document but not part of assertion. The only one that explicitly mentions is if author training is provided, it must provide guidance on these aspects of clear structure. Part of the clear structure review.
alastairc: It's not tracking whether people have done it or not, if you provide it, it should include this content.
GreggVan: Any requirement that requires clear structure — there are no clear definitions — requiring people to be clear is interesting.
GreggVan: We need to differentiate bet. mandatory trainings and training you have to have if you do design, and when you're doing electrical design, it has to be signed off by a professional engineer but nobody else needs to be trained, but making assertions about what teams do where there are no sign offs or required training to do them, companies slide people in and out of these things all the time and don't track the training.
<ShawnT> Example of training: the Accessible Canada Regulations on Digital Accessibility states "Federal public sector organizations, large and medium-sized businesses must train their employees on digital accessibility by December 5, 2027." (cite:
alastairc: Let's not ask for anything that would require people to track where training has taken place.
GreggVan: If you are saying that something was done, that people have to be trained what those words mean — all assertions require training.
GreggVan: You have to explain, there's implied training.
Detlev: One of the critical questions we're discussing is whether assertions can be placed at the core level to meet some requirement, to make it more concrete we could take an example, say, large e-commerce company with lots of images that claim to have a process whenever uploading an image, there is a process we check whether it has alt text.
Detlev: If it doesn't have one, we generate one through AI, this might be a way of meeting the graphics part which is now 1.1.1 Alternative Text for Images
Detlev: Imagine many companies would like that, they'd have automated processes to do it. We can show we have the processes to do it. Would it be the point where you imagine that WCAG 3 would include assertions or these cases?
alastairc: We use assertions to fill gaps where you can't have more outcome-based things. We're taking the approach of test the alt text, not what generates it, it is a testable thing.
Detlev: You have a human element checking whether the alt text is correct. You could say it can't be done, is that the point where an automated way generates it — cannot r eally be done on an outcome-based basis at scale can it?
alastairc: We're tending not to use assertions, that would be more of WCAG-EM type of thing for compliance, not conformance.
<Zakim> joryc, you wanted to say training is not really any guarantee of a11y and maybe doesn't matter in the world of LLMs
<alastairc> Personally disagree with the training bit, it's never there from the start which is the issue.
joryc: Detlev is bringing up an important point — I'm hesistant to add required training and proof of this — proces will change so quickly, training does not work, if we spent the last 10 years to train everyone and it didn't work, I'd rather have a solid review system than have every single person trained. Then you'd know exactly what things are going to happen.
joryc: Speaks more to the outcome, assert that they have systems in place, without having to demonstrate a particular process — it won't age well and even if it's a difficult process at different sizes of companies, it will have differing effects.
<kirkwood> Could it be that assertions have a danger of a “get out of jail free” card?
alastairc: If you have an outcome, why would you need an assertion?
joryc: You can assert, if you have the outcome, but also assert that you have a process to generate this outcome. Assert that there is a process that leads to that outcome rather than explicitly trained.
<kirkwood> +1 to Jory
From the state of the web right now not everyone will be able to make every assertion — the existence of WCAG 3 won't force your hand into making an assertion to get points towards the next level of conformance. Having assertions with these examples is great, it's good to have the framework there.
Ben_Tillyer: If legislators did want to assert things, it becomes their problem to balance that with other laws that exist in their market, we are not forcing anyone. I don't know whether we should be ensuring that all assertions could be answered without breaking laws.
julierawe: There's at least one assertion that talks about, if training is provided — comments about training, if we just said that the organisation must have a style guide that takes the place of training, it is a tricky one. I think John Kirk mentioned that training has been used elsewhere in assertions but we have to figure out what this means.
julierawe: It is impossibel to make a testable requirement that's why we're in the zone of assertions, looks like we're l eaning towards reviews being okay and training is not okay — but not sure how much clarity I am getting from it.
alastairc: There's no training-centred assertion. IF you have training, THEN include this topic — that's the limit of it.
shadi: These comments are not f rom shadi or that organisation.
alastairc: Tech company slide — they are concerned about maintenance due to assertions.
alastairc: Broad assertions would likely be too generic to be truly helpful. That would support guidance on best practices, but not as normative requirements to disclose internal processes or policies.
alastairc: There's a misunderstanding that we're asking for any public internal policies — would just be doing a statement.
alastairc: Similar issues in terms of scoping that we discussed last feedback.
alastairc: We can come up with guardrails for assertions on what we can include. Question about whether there are particular types of assertion that need to get more specific with more specific examples.
scott: Even through reading people's feedback, the examples in the presentation last time that made it unclear to my group when we were reviewing this what was required or not. I don't know how to talk to other people at Microsoft about what is being proposed.
scott: I think it's also being stated that it would be an optional thing, so I don't understand, could we have examples? That worked in the presentation before.
<shadi> +1 to scott
<kirkwood> +1 specific examples would be helpful
alastairc: Will try and come back to that.
<SydneyColeman> +1 i do not know what anyone is talking about
<shadi> [it's a lot of working gathering feedback at a large company]
Rachael: Chairs will do a reset and come back and restart this.
<SydneyColeman> or how to represent this discussion to google in a tl;dr type way
Good enough" conversation https://docs.google.com/presentation/d/1-124T8uEc9H6xAEB-ezcanXb34TNjemHcWWr7T5p1MU/edit?slide=id.p#slide=id.p
alastairc: Thank you for those who provided feedback — maybe we will come back to the group with another version of this to improve on that.
Rachael: This wasn't clear enough in the agenda, today's conversation is around when is AT or user agent or additional support — the tech stack support — good enough to step back?
Rachael: Guardrails for this conservation: one of the hopes of WCAG 3 is to be future proof.
Rachael: When we put it on the author, it's harder to get accessible sites back and we want to figure out ways forward, especially as technology speedily changes.
Rachael: Using technical infrastructure, in situations where the author can focus on avoiding/breaking that support, instead of actively providing support shifts the burden a little bit off the author.
Rachael: Question around the tech infrastructure itself — when is it good enough to allow us to step off? How much support is available, in a language set, the number of browsers or screen readers able to support, etc. the cost of it and technical complexity.
Rachael: Is it something to add as an extension in browser? Other cutoff points.
Rachael: This is not what we're talking about today — today is about the provision-specific quality measures.
Rachael: We have guidance around when the technical infrastructure is good enough but when is the actual support itself good enough for each provision? There's a new provision not in WCAG 2 that gives a baseline to explain this.
Rachael: A proposed baseline is to use what a human expert can do, the cutoff point. That's what the author is.
Rachael: There are other measures, presence or absence, error frequency, misrepresentation, quality checkpoints of different types.
Rachael: (slide 4) Captions — rapidly becoming true, we have a core requirement for captions being available for prerecorded content. 98-99% accuracy is being captioned with a trained human expert captioner but 95% accuracy for technical jargon.
Rachael: What would be the accuracy needed for automated captions?
<Detlev> are you sharing??
(slide 5) Text detectable — if it is detectable, the requirement is all visible text has programmatically determinable equivalent.
Rachael: Assuming all of the text and images provide correct alternatives, can the requirements shift where the authors is not breaking things.
<Detlev> alastairc ah wasN#t aware of having several tabs now...
Rachael: (slide 6)
Rachael: (slide 7) In WCAG 2 when something was met by a tech stack, we didn't write it in WCAG 3.
Rachael: (Reading slide 7)
<Zakim> GN, you wanted to say the accuracy percentage might depend on further conditions
Rachael: Key question: is this a section we want to add and discussion around what that would look like
GN: The first example with automated caption, it might be more complex and depend on more conditions. People who rely on captions see a big difference in the need for a live meeting where they could immediately ask if something is not understand. Important information like news or TV, or enjoyable content, they'll need higher currency and higher quality.
GN: When it comes to technical jargon and non-technical jargon, its target is user-based. It needs to be accurate for non-technical users. Technical users might be able to guess the term.
GN: As for how far it helps, I'm open to learn.
Charles: I have the same question and concern as GN on accuracy point, but if context/percentage of accuracy is dependent on context, it gets more complicated. It could include ability to compare the caption to a transcript where one of those methods says one thing and other says other things.
Charles: The content vs. the context — example we have the concept of a financial obligation being made by an interaction — this could be captions on a medical procedure or financial transaction/instruction where the consequences of the accuracy are higher.
Rachael: One thought is we could simply set it to the highest context. Simply say, the best a human can do is 98%, automated captionos 98%, that would be an acceptable place to make that replacement.
Rachael: Is the value of creating something like this at this time in the next 3-4 years — is that worthwhile? If it helps us refine the requirement text. Or is it something to postpone after we get the requirement text?
giacomo-petri: Is the way captions are provided relevant? If 98% is what we want to validate the captions requirement, is it relevant if it is provided automatically or by a human?
<Ben_Tillyer> +1 to giacomo-petri
<Ben_Tillyer> This is about the user need surely
<GN015> +1 o Giacomo
<Zakim> Rachael, you wanted to answer giacomo
giacomo-petri: If I am accessing the video, I need it to be accurate, if it is above 98, it passes, and below, it fails. Not sure the distinction bet. automation/AI — what the user perceives must be accurate.
<julierawe> +1 to Giacomo
<joryc> Strong +1 Giacomo
<Charles> percent of accuracy would also vary by volume of content. 98% of 500 words is different than 98% of 50000 words.
Rachael: The use case is, for example, a video with no captions — I will put it on YouTube to provide automated captions. As author, I am required to fix them or create my own or can I simply say YouTube captions are enough and not required to do anything beyond that?
Rachael: The question is if we need these kinds of measures to make those decisions?
alastairc: If we had 2 methods, you apply captions or you rely on an automated tool, when can we say that that second method is good enough, is that right?
<Zakim> alastairc, you wanted to comment on available vs quality level provisions
alastairc: Chair hat off, is this always going to be built in to base level 1? Depending on how captions are defined, does it need a quality metric?
shadi: If I understand correctly, there are criteria where quality measure would be very useful, the difficulty is to define the quality measure, e.g. caption quality — different types of methods, accuracy of words themselves, punctuation and grammar, identification of people, presentation/identification of speakers, is it accurate in the positioning?
<kirkwood> This is a legal question the answer is in the definition in our definition of alt text. the legal issues is that the author needs to attest that it’s correct
shadi: There are different kinds of metrics, I think it will come down to how we define these metrics. I don't think it makes a difference if there's a quality threshold?
<Zakim> GN, you wanted to say that human accuracy to my experience is higher than 98% on prerecorded media
shadi: Does it need to uphold a certain threshold?
GN: Humans can reach accuracy 98% , I doubt it and would not be content with it. If we match it to written text that has 100 letters and 2 lines, and if a secretary does a typo every two lines, they would not pass an exam. Captions on prerecorded videos, and strive for an accuracy of 100%, I claim we reach 99%, and nothing lower, there would be too many errors.
<kirkwood> alt text “ conveys the purpose and meaning of an image”
julierawe: Interesting conversation, noting that adding percentage accuracy — this shortname is Captions available, but we're not talking about available only, shifts to accuracy — is it available AND accurate? Additional thing that is not mentioned until you get to standards section.
julierawe: How many provisions would need to be adjusted in this way?
<kirkwood> how can you be accurate about the purpose? of the image etc. that is up to the author.
<Jon_Avila> What if we had an automatic caption option for user generated content that the site owner needs to meet - in that case some limited level could be required - but it would then be on the user who provided the content to provide captions that meet the full requirement.
Rachael: That's a great point, it highlights a problem in this text — we don't have the word equivalent, and there are additional requirements that clarify, but the missing term equivalent gives it a gap — the process of thinking through this may have great value.
alastairc: Definition of captions implies accuracy.
<kirkwood> author purpose can be lost through ai
janina: We're trying to think of AI in APA Research Questions Task Force — sometimes context can determine whether auto-generated captions/descriptions/content might just be sufficient because it's close enough and just there for entertainment or information, but in other circumstances, absolutely not acceptable esp legally held to or educational settings.
janina: We're probably not going to get it anytime soon from an AI, but we're going to be inundated. The comprehensive quality matters, as well as the context.
Jon_Avila: If we had an authoring tool requirement, require them, and platforms, to provide some level of captioning, but since it's the platform doing the captioning, we can't have the same requirement for content that the author supplies themselves.
Jon_Avila: We might need to have requirements that allow for different possible levels.
Jon_Avila: To make sure we're not preventing tools from being used.
<Zakim> alastairc, you wanted to comment on having a higher level regs for certain sectors
<kevin> There is a separate piece of work looking at authoring tools!
alastairc: Chair hat off, building on julierawe and Jon_Avila 's points, we have a fairly basic, base level 1 Captions are available and anything goes, then medium level, human accuracy, and higher level where we wouldn't say that this is necessarily for education/legal, but could be part of a higher level regulators could look into and say that it is necessary if it is legal rather than entertainment.
<kirkwood> the PURPOSE needs to be conveyed (the author most accurately provide purpose) one should know if approved by author or not
alastairc: AI might soon beat us and we don't have an opinion on how captions were generated but we need a mechanism to hand off — if we have a human or method that adds captions and check accuracy, if we could have another method under the same requirements, if platform meets this level of quality, use it and don't worry about it.
LoriO: Concern about the automatic captions, has to do with places where captions must be 100% correct, medical content — a doctor looking up in a medical journal and your caption is not quite right but if you go by what the caption is, you can cause harm and one of our tenets is do no harm.
LoriO: There are times when accuracy MUST be 100% or else fails.
<bbailey> Just noting that caption quality, in addition to accuracy, includes synchronicity and placement. FCC caption quality standards: https://
<Zakim> joryc, you wanted to say human accuracy should probably be the standard so we don't force AI use, but AI will soon beat us, we should not have an opinion on how they were generated.
alastairc: Similar to what I said a while ago, to have a potentially higher level.
<kirkwood> we should provide a framwork to let the reader/viewer know who the author of the captions are
joryc: We should get away from any preconception on how captions are created and just go with standard. The human standard is the way to go, don't want to force people to use AI, but by the time WCAG comes out, AI may be more accurate.
<Jon_Avila> If you look at your medical record - i can almost guarantee it's not 100% accurate from your doctor's office already. So, 100% is not likely.
joryc: I don't think we need to talk about anything other than accuracy that needs to be reviewed, not that a human needs to look at it — need to be agnostic about what creates the captions and stick to accuracy — human standard is the right standard.
joryc: There are cases to have 100% accurate that is valid.
<Ben_Tillyer> Verbal contracts are legally binding in the UK.
GreggVan: Getting into realm of making up problems, anything can be conveyed verbally and not in writing — can be conveying legal information, it would have been done in text. Medical information thought important could just be said to somebody and not provided to them in writing — high accuracy, none of them would be presented orally but in writing.
GreggVan: All the places we need high accuracy shouldn't rely on caption.
<kirkwood> wwe should have an attestation statement on caption by author ;)
<Ben_Tillyer> +1
<Zakim> Rachael, you wanted to close
<bbailey> giacomo-petri: I don't think we shouldn't rely on automation, at the end it could be a platform that says their tool is 98% accurate in providing captions but at the very end you don't know until you test it. At the end if we have a method, it might be invalid and still needs to be tested. It could still fail regardless of the statement of the
<bbailey> platform.
<GreggVan> -1
<Rachael> Should we continue this conversation?
<bbailey> Rachael: Should we continue this conversation or just put this into the methods process?
<Ben_Tillyer> +1 to continue
<stevef> +1
<ShawnT> +1 to continue
<Rachael> +1 if its useful -1 if this should remain in subgroups
<Francis_Storr> +1
<Detlev> +1
<julierawe> +1 (helpful to see examples from several groups)
<CClaire> +1
<scott> +1
<Makoto_U> +1
<laura> +1
<GN015> +1 to continue, as I think we haven't come to an agreement yet (I'd put it into the methods, though)
<Jon_Avila> +1 useful
<bbailey> 0
giacomo-petri: I don't think we shouldn't add method specific for automation, at the end it could be a platform that says their tool is 98% accurate in providing captions but at the very end you don't know until you test it. At the end if we have a method, it might be invalid and still needs to be tested. It could still fail regardless of the statement of the platform.
Rachael: Should we continue this conversation or just put this into the methods process?