"EPUB Creators MUST" and "EPUB Creators SHOULD" in EPUB A11Y 1.1

[murata2makoto]

EPUB A11Y 1.1 has several sentences beginning with "EPUB Creators MUST" or "EPUB Creators SHOULD". I propose to reformulate them as requirements on EPUB publications rather than those on humans.

First, the scope of EPUB A11Y 1.1 is the accessibility and discoverability of EPUB Publications. Humans are outside the scope.

Second, it is not possible to create tests for these sentences. Ace does not check EPUB creators either.

Third, these sentences do not impose any requirements on EPUB publications, since programs or other users are not prohibited from changing what EPUB creators did.

3.4.1.3.1 PAGINATION SOURCE

The 1st para, "Meeting this Objective"

EPUB Creators MUST identify that source in the Package Document metadata.

Reformulate the above sentence.

3.4.1.3.2 PAGE LIST

The 2nd para, in "Meeting this Objective"

EPUB Creators SHOULD ....

Ditto.

The 3rd para, in "Meeting this Objective"

EPUB Creators should

Is this just a non-normative sentence? (Note: "should" rather than "SHOULD" here).

3.4.1.3.3 Page Breaks

The 2nd para, "Meeting this Objective"

they SHOULD include page break markers for ....

What is the subject ("they")? It cannot be "an EPUB Creator" since it is singular. Reformulate this sentence without using "EPUB creator" as the subject.

they should include

Is this just a non-normative sentence? (Note: "should" rather than "SHOULD" here).

The last para, "Meeting this Objective"

EPUB Creators MUST identify the page numbers in the markup that controls the playback.

Reformulate the above sentence.

3.4.2.3.1 COMPLETENESS

1st para, "Meeting this Objective"

EPUB Creators MUST provide synchronized audio playback via Media Overlays for all textual content.

Reformulate the above sentence.

3.4.2.3.2 Reading Order

Ths 1st para, "Meeting this Objective"

EPUB Creators SHOULD ...

Reformulate the above sentence.

3.4.2.3.3 Skippability

The 1st para, "Meeting this Objective"

EPUB Creators SHOULD ...

Reformulate the above sentence.

3.4.2.3.4 Escapability

The 1st para, "Meeting this Objective"

EPUB Creators SHOULD ...

Reformulate the above sentence.

3.4.2.3.5 Navigation Document

The 1st para, "Meeting this Objective"

EPUB Creators SHOULD ...

Reformulate the above sentence.

3.3.2.1 Page and Publication

EPUB Creators MUST evaluate their accessibility as part of the larger work.

Reformulate the above sentence.

EPUB Creators MUST evaluate the WCAG guidelines for content to be ...

Reformulate the above sentence.

3.3.2.2 Applying the Conformance Criteria

EPUB Creators MUST NOT use EPUB's fallback mechanisms to provide a conforming alternate version [WCAG2], ...

Here is my proposal.

Every EPUB and Foreign Content Document in the spine plane (i.e., either in the spine element or in the fallbackchain) MUST meet the conformance requirements of the level claimed.

3.5.2 Publication Conformance

There are four occurrences of "EPUB Creators MUST". All of them should be reformulated.

Here is my proposed rewrite for the first one.

Values matching the pattern A11Y-VER indicate the version number of the EPUB Accessibility specification the publication conforms to, not including the decimal points. The value "11" indicates conformance to this version of this specification.

[mattgarrish]

If we do this, it should be done consistently for the core spec, too. I'd put this in the post-CR cleanup bucket since it's editorial and not critical to have done now.

[iherman]

• For the reformulation: I understand the logic of what you propose, but I agree that this should be then done for all the Recommendation track and Note documents we publish. This is a significant and highly error-prone process, and I am not sure it is worth the effort.
• For the testing thing: we propose a different set of exit criteria for A11y which might alleviate some issues. However, if none of the A11y checker tools check some of these requirements, then we may indeed have a problem either with the exit criteria or with the spec...

Cc @avneeshsingh

[murata2makoto]

@iherman

I certainly think that this change should be done for all of our documents.

[mattgarrish]

• This is a significant and highly error-prone process, and I am not sure it is worth the effort.

That's what we need to look at, too. We are splitting hairs here. Of course all the requirements are testable (and are tested) as they are still requirements on what the individuals or organization (it's not necessarily a single human) have to put in the publication.

It's simply a matter of phrasing preference: must I add a metadata property or must the metadata have a property. Should the error say "you forgot to add" or "the metadata is missing from".

It's slightly clearer to say what has to be where, but, like I said, it's not critical.

[murata2makoto]

@mattgarrish

With respect, I disagree entirely. Requirements on humans are outside the scope. Testing humans are not possible by Ace. The only way to do it is to interpret them as requirements on data. Since nothing prohibits other programs to touch what EPUB creators did, these sentences fail to impose requirements on data. Distinguishing data conformance and application conformance and avoiding the latter wherever possible are very important things in creating standards on documents.

[dauwhe]

First, the scope of EPUB A11Y 1.1 is the accessibility and discoverability of EPUB Publications. Humans are outside the scope.

The spec is written by humans, for human content creators, in support of human readers. I would say humans are in scope.

Third, these sentences do not impose any requirements on EPUB publications, since programs or other users are not prohibited from changing what EPUB creators did.

Are you saying that if, for example, a reading system had an ingestion script that removed all image descriptions, page breaks, and accessibility metadata it would be compliant with the spec?

[mattgarrish]

Requirements on humans are outside the scope.

That an EPUB Creator must include a property in a metadata section is not a requirement that involves testing a human, though, just as one example. There is no effect on the creator to verify. There is only a question of whether the metadata section contains the property they had to set. That is fully testable.

I understand what you want are actorless statements, but that doesn't mean there isn't an actor creating the content, even if it's an automated process (someone had to write the automation, after all).

Since nothing prohibits other programs to touch what EPUB creators did

I don't understand what you're saying here. The EPUB Creator is whoever is responsible for the creation of the EPUB. If another party/program takes an EPUB and makes modifications to it, they are playing their own role as a creator, as the result is a different EPUB. Someone is still responsible for ensuring the new content is following the requirements, even if it's an authoring tool developer.

Anyway, don't forget that Ivan and I don't disagree with you in principle, only on whether this is worth the effort at this stage. The work involved is not trivial, it does risk modifying normative statements in unwanted ways, and it's not clear it benefits the average reader or is only the stuff of interest to insiders like us who have been doing this too long. 😉

Can we perhaps look at the epub creator definition and see if it can benefit from some tweaking and leave a major overhaul to another revision?

[murata2makoto]

First, I agree to postpone this issue after the CR.

Having said that, I would like to give some background information behind my comment.

Document conformance and application conformance are completely different approaches in standardizing document formats. Document conformance is based on requirements on documents, while application conformance is based on requirements on behaviors of document processing applications.

Application conformance is tempting, since users see document processing applications rather than documents, which are byte streams.

In my more than 30 years of experiences with document format standardization, I have seen three major attempts for application conformance. The first one is ODA (ISO/IEC 8613), which was developed in eighties. Then, ODF (ISO/IEC 26300) and OOXML (ISO/IEC 29500) appeared around the same time. During the development of these document formats, some people wanted to persue application conformance and ensure interoperability of applications.

Application conformance in these standards are not miserable failures, since their definitions of application conformance are sound (although sketchy and abstract). But these attempts certainly did not achieve what people wanted. Nobody cares whether or not applications achieve application conformance. Application conformance is difficult to test and does not ensure usefulness of applications.

I thus firmly believe that we should avoid application conformance wherever possible and stick to document conformance. Application conformance is tempting but is a red herring. "EPUB creators MUST" is application conformance.

[clapierre]

In your statement

I thus firmly believe that we should avoid application conformance wherever possible and stick to document conformance. Application conformance is tempting but is a red herring. "EPUB creators MUST" is application conformance.

I would think “EPUB creators MUST” is a content conformance statement, whereas “ReadingSystem creators MUST” is an application conformance statement.

[clapierre]

Maybe instead of "EPUB creators ..." if we instead just use "EPUBs ..." we don't really care how they are created.
the HTML living spec does say "authors must/should".

[iherman]

@murata2makoto, I cannot comment on ISO specifications, I have not been involved with ISO standards for over 20 years now. However, if I take your classification, I believe that our spec is actually closer to what you call "application conformance". Indeed, the very essence of the W3C CR procedure is to go through tests, and implementations abiding to those tests. This is what the RS specification does, and this is what the current test suite will do when it will be complete.

Of course, EPUB is a mixed beast, because it is a file/document format as well as spec defining the core application behavior. I believe the comparison with ISO which (as far as i know) does not have the equivalent of our CR process may be misleading.

As for the original issue:

[@mattgarrish:]Anyway, don't forget that Ivan and I don't disagree with you in principle, only on whether this is worth the effort at this stage. The work involved is not trivial, it does risk modifying normative statements in unwanted ways, and it's not clear it benefits the average reader or is only the stuff of interest to insiders like us who have been doing this too long. 😉

I would be a bit more forthcoming here: I do not think it is worth the effort at this stage. If I consider the work it would take to do this and the danger of making mistakes and inadvertently modifying normative statements (that can backfire on us much later) I think it would not be a responsible choice to do this at this point.

[iherman]

Can we perhaps look at the epub creator definition and see if it can benefit from some tweaking and leave a major overhaul to another revision?

I was looking at the term definition of EPUB Creator and I believe we can "depersonalize" it, ie, make it clear that it can contain automatic processes. My first shot at it is:

EPUB Creator

The person(s), organization, or processes responsible for the creation of an EPUB Publication.

Depending on the process used to produce EPUB Publications, EPUB Creator may sometimes refer to responsibilities of the organization (e.g., the publisher), the individuals preparing the publication (e.g., technical editors), or automatic procedures (e.g., as part of a publication pipeline).

[clapierre]

I like this I would also include "conversion vendor" as this is standard practice for a publisher to hire a conversion vendor to build the EPUB to their specifications.

[mattgarrish]

My first shot at it is:

Ya, that's the general idea I had, too, when I looked at it again.

I do not think it is worth the effort at this stage.

What I find working against it is I don't see any practical benefit to the community coming from a change. Who is this solving a problem for? If it's all theoretical improvement, now is not the time to address it

Looking at other W3C/WHATWG specs, what "authors" have to do is all over them. We're fully in keeping with organizational practices.

[mattgarrish]

I like this I would also include "conversion vendor"

I think this is similar to what @murata2makoto mentioned about other parties changing an EPUB. I don't want to introduce terms for every possibility, but perhaps we can emphasize that creation may involve multiple parties.

But a lot of this is explanatory, so maybe we can turn it into a note to simplify the description. Writing this hastily as I have to run out for the evening, but I'm thinking of something like:

EPUB Creator

An individual, organization or process that produces an EPUB Publication.

NOTE

The creation of an EPUB Publication often involves the work of many individuals, and may be split across multiple organizations (e.g., when a publisher outsources all or part of the work).

Depending on the process used to produce EPUB Publications, responsibilities may fall on the organization (e.g., the publisher), the individuals preparing the publication (e.g., technical editors), or automatic procedures (e.g., as part of a publication pipeline).

As a result, not every party or process may be responsible for ensuring every requirement is met, but there is always an EPUB Creator responsible for the conformance of the final EPUB Publication.

[iherman]

To make things easier to discuss in details, I have created PR #2234