The Truth About Auto-Tagging Webinar

Equidox Webinar logo

Can auto-tagging result in accessible documents? How can automation speed up PDF accessibility? Let’s take a look at Adobe’s auto-tagging feature including what it can and cannot do. Then we’ll see how the results compare with Equidox PDF remediation software and its Smart Detection Tools. Hosted by our own Dan Tuleta.

Webinar Video

Webinar Slide Deck

 

Webinar Transcript

All right so it’s just about two o’clock. So welcome in everyone. Thank you for attending another edition of Equidox Webinar Wednesdays. This month we are going to be talking about the Truth About Auto-Tagging.

A lot of people that we speak to seem to have a pretty heavy reliance on Adobe Acrobat’s auto-tagging feature. Which is not really intended to make documents fully accessible. It’s really just designed to get you started on the path of remediating that PDF. We want to talk a little bit about how that works, and its limitations, and different ways of going about using Equidox to fix the issues that auto-tagging can create. As always I’d just like to mention that if you would like to contact us feel free to reach out to EquidoxSales@Onixnet.com. Our website is www.Equidox.co.

We’re also very active on social media, so if you would like to connect with us on LinkedIn or Twitter, or Facebook we do post a lot of information and articles. So you can always stay up to date with what’s happening with the Equidox team. So with that said let’s get started. I’m just going to quickly mention a couple of things about our parent company. Onix is mainly a cloud consultancy. And we’re most known for our long-standing partnership with Google. And Equidox is a software product that is owned and developed by Onix, along with other professional services based mainly around PDF accessibility, and our mission here with the Equidox team is to ensure that digital information reaches everyone via accessibility solutions.

just a quick slide here about some clients of ours. We do work with notable names of various sizes and across all major verticals whether it be education, government, or enterprise. So we are here to help you regardless of how big or small your company is. So please don’t feel that you’re too big or too small to work with us. We are here to help you with all of your digital accessibility needs.

Okay, what we’re going to be doing today is we’ve just put together a very stock sample PDF document. This is one quick… (Oh, and I should mention before we go any further that there are a series of polls that we are going to be asking during the webinar, so please feel free to answer those. Just if you have a moment just to answer the little survey questions that pop up, that’s greatly appreciated. That will help us kind of tailor these webinars moving forward, to make sure that we’re addressing any questions or topics that you would like to see us cover.) But what we’re going to do today is take a look here at this sample PDF document. And so this document that we’ve put together, it kind of just contains a bunch of the common elements that you would find in any given PDF. So you’ll see things like text, headings, lists, tables, images, OCR, and of course the complicated task of setting the reading order for different pages. So we’ll be kind of working through some of these different elements and talking about ways that auto-tagging can go wrong, and ways of fixing it after the fact in Equidox.

Okay, the first attempt that we made at auto-tagging this simple document here, the simple one-page document… For whatever reason, Adobe Acrobat decided to tag the entire document as one big image. So this entire page, despite the fact that it has embedded text on it, and is full of all kinds of different elements such as of course the text, the headings, the list, as I mentioned before… Acrobat, for whatever reason, just decided that this whole thing is just going to be one big image. So there are a lot of problems with tagging an entire document like this with just something as simple as an image. Because the image itself is useless to a screen reader user.

There need to be tags around the content to give something for the screen reader to actually interact with and navigate around the page. There is no text-based content for the screen reader to read from in this current state. This of course would need to be fixed or changed or at the very least another attempt at auto-tagging it. So the second try, when we tried to auto-tag this again, and this kind of speaks to the inconsistency and the lack of reliability that you’ll get with auto-tagging in Acrobat…

So the second attempt, we actually were able to get some tags created around the content. So some of the elements were tagged. Some of the elements were missed. But in the GIF, if you see on the left-hand side here, we’re kind of scrolling through the tag panel in Acrobat… And you can see that it’s pretty complicated-looking. It’s messy. But this is what an auto-generated tag structure can look like. So if you’re a novice user, if you don’t know exactly what you’re doing here, trying to understand the complexities of this tag structure on the left-hand side and how this relates to making the document accessible… This can be a tall task. And especially if you’re trying to work through dozens of pages, dozens of documents, it can be a very tedious and time-consuming process.

The tags that we’re looking at here in the GIF are entirely inaccurate and this document is really a long way away from being accessible and usable to a screen reader user. And like I said before, the goal of auto-tagging is really just to get you started. But it is inconsistent. It is unreliable. And it definitely requires human intervention after the fact to fix the many elements that will be missed or inaccurately tagged.

So the next logical step, assuming that you’re a novice remediator, you’ve hit the auto-tag button and now you’ve got some tags… The next logical step would be to run the accessibility checker. Now it’s kind of funny because Acrobat is checking its own work. Acrobat applied the auto-tagging, and now we’re using Acrobat again to check the accessibility of those tags. So Acrobat here is checking its own work. And even when it’s applying its own tag structures to this document, it’s still locating multiple errors in the tag structure. Keeping in mind that there are also many other errors in this tag structure that the accessibility checker is incapable of identifying, there are many things that require human intervention to identify and locate, and of course, go back and correct…

If you are an experienced remediator you would be able to interact with this tag structure and navigate through the tag panel. And you would find dozens of issues that would need to be corrected. And the only way that they can be corrected is through manual intervention. You cannot just continue to hit the auto-tag button and find ways of fixing it. There is no magic solution. You simply have to work through the elements one by one and correct all of the errors that have been created.

Let’s see here… There we go, so one of the first errors that we’re going to point out here in this auto-tag structure is the text error. If you notice at the top of the page, all of that content across the two columns where we have two separate basic contact information or letterhead for the chief information officer and the chief executive officer of this fake company that we’ve invented… It’s going to group all of that text together inside of one tag. And the screen reader would be instructed by the tag structure to read directly across the two columns which are separating the two groups of contact information.

If you were a screen reader user, it would be very confusing, because you would be hearing both names. You would hear Joe Smith and Bob Jones read directly after each other. Then both addresses… then both phone numbers… so on and so forth, all the way down these two separate columns. So it would render those two bits of contact information pretty much useless to a screen reader user because it would not be read in the correct order. And this is something that of course Acrobat just did automatically. so it needs to be fixed manually.

Another error that we are able to easily locate here is a heading error. So within logical heading structure, the first heading in a document needs to be set as a heading level one… And for some reason, Acrobat decided to tag this very obvious heading level one… the top portion of the page as in as a heading level three. So this is objectively incorrect. The first heading in a document needs to be a heading level one. Adobe knows this, and still, their auto-tagger is kind of just you know grasping at straws as it’s applying tags to these different elements. So it assigns arbitrarily an H3 to this element, despite the fact that it has to be a heading level one according to all accessibility rules. So the reason it’s setting it as an H3 instead of an H1, it’s anyone’s guess, but it’s probably due to the font size or the font style that was used. But whatever the case may be, it does need to be corrected. And it can only be corrected manually. Okay, so the next major error that we’re going to see is another heading error. But this is a nesting error. So the second heading on this document has been identified as an H2. Now you cannot have an H2 directly following an H3 so this needs to be fixed as well.

Another problem here is that we have this element on the page which is actually a block of text, but it is sat inside of an image. So this is something that can happen depending on the way that the document was designed. So if there is text locked inside of an image that is just recognized as an image, as far as a screen reader is concerned. So OCR is required (Optical character recognition) to actually convert that text into something that is usable and can be read by a screen reader. Now Acrobat is able to locate this figure tag, however the content inside of it it needs to be converted into something that is actually embedded text, that can be read aloud by a screen reader. However, using OCR like on this one particular page, when we’re this far down into the actual remediation of it, can be a little bit risky.

There are strange occurrences of Acrobat trying to re-OCR the entire page or block the user from continuing to tag the rest of the elements underneath it for the rest of the document. This is something that is it’s kind of a challenge to deal with in Acrobat, and it needs to be corrected, however. So definitely another little challenging issue here with just the plain old auto-tagging.

Now another obvious area, as we just continue to work down the page, every single element you’ll notice seems to have something wrong with it… And this is just the nature of auto-tagging… So the next problem is the lists. So lists are almost never tagged correctly from auto-tagging. And the entire list of these seven different items, it’s going to be read and grouped together as a paragraph. So all seven of those list items are just going to be read as a strange run-on sentence. And doing this in Acrobat is extremely tedious. It requires dozens of steps, and even for a small simple list like this, there are so many clicks and steps associated with tagging it correctly that if you’re just a novice user and you don’t know exactly how to navigate the complexities of this Acrobat program in the tag panel, you can have a very very difficult time just trying to tag this one very simple list.

The next thing I want to point out is the table. So tables are almost never correctly auto-tagged, and they also require manual fixes of course. So the table cannot be left in this auto-tag state, because the screen reader is just going to be… it’s going to read straight across the cells without any indication of how the user is currently oriented in terms of rows and columns and like where that screen reader is actually focused on. So it renders the entire table useless because it simply cannot be navigated as a table is intended to be navigated.

So how do we go about fixing this? So we’ve pointed out obviously at least half a dozen major flaws with the auto-tagging of just this one simple page. So what we wanted to show you today is just opening up Equidox, our PDF remediation software, and we are going to use the different features in the platform to fix these types of issues. We’re going to use things like our Zone Detection tools, which is effectively its own type of auto-tagging, where we’re able to group together and find different clusters of content to get the user started. But of course, we have a much easier interface and a much easier workflow to navigate through to actually make these corrections to the document itself. So this list of bullet points is just a handful of things that we are going to do as we go through the Equidox demonstration.

I’m going to back out of the slide deck here, and I’m actually going to open up Equidox. And I already have the document pre-loaded in my Equidox account. If you take a look here, this is what the document looks like when we just simply import it into Equidox. so we have the tag structure that came from the original this is Adobe’s auto-tagging. If we remember, that issue before of the content being read straight across the two separate columns up here… We have our heading level three, our heading level two, our text zone capturing this entire list, this image of text is is going to be a problem as well… And then down here, we have this table, which is going to be read simply as text itself.

Before we make any changes or alterations to this, if you’ve never seen an Equidox demonstration before, if I press this button up here which looks a bit like a computer monitor, this will actually open a separate tab in my browser. As soon as my browser wakes up… Interesting… there we go! So it will open a separate tab in my browser, and what I can see here is an HTML rendering of the page that I’m currently working on. So the HTML is, basically, what it’s doing, is it is showing me exactly how a screen reader is going to interact with these tags as they currently sit. If I take a look here, if we remember the issue with the text jumping across the two columns, a screen reader would read that as Joe Smith Bob Jones and then would read their titles…

It would read their corporations back to back, it would read their headquarters, and their different addresses in terms of the street address, the city, and the state, their phone numbers… This is very convoluted and confusing and not at all intended how it should be read by a screen reader. I’m working down the page, you can see other issues here… We have our list down here at the very bottom, which is going to be just read as a strange paragraph with little stars inserted inside of it for list labels. And then this is actually what our table looks like currently. So our table is also just going to be read as a strange run-on sentence that will make no sense to a screen reader user… not even being aware that this is supposed to be a table until we tag it as such.

What I’ll do is, I’m going to come back into the actual Equidox Page Editor, and I’m going to make some adjustments to this page. the first thing that I want to adjust is the page. If I go to the Page Tab, what I’ll do is I’m going to use the Zone Detector tool. So the Zone Detector, what this is designed to do is ignore the existing tag structure, so that tag structure that Acrobat gave us with the auto-tag feature that’s not helping us at all… And I don’t want to try to work with those tags. I don’t want to have to undo all of the mistakes that it made. So we can let Equidox just redefine this page by using this Zone Detection slider. So if I move this slider back and forth, left to right, you’ll instantly see how we’ve cleaned things up pretty significantly.

We’ve gotten rid of all of those extra tags that were just really kind of getting in the way. And now we’re left with a much cleaner starting point that doesn’t require as many manual adjustments. So if I were to go into the HTML Preview again, already I’ve made some improvements simply by the fact that I’ve grouped together Joe Smith and Bob Jones’s contact information. At least now it’s going to be read properly.

I do have to make some additional adjustments, of course. We have things like images, we have heading structure, we have to apply break lines to these addresses, we have the list, and the table and we also have the OCR. So let’s just start working through the page, top to bottom. Up here in the top margin, we have just the version number of this document. Things like this can often be artifacted because they’re not necessarily adding much value to the tag structure.

I can start by just hitting backspace to get rid of that extra zone. I have this image here, which I’ll just begin with because it’s the next tallest zone. But what I like…what I can do is actually apply an alt description to it. So I can type in “letterhead mark maker sample company logo.” So I’m able to apply an alt text to that image so that it will actually have some context for the screen reader user as to what this image is. The contact information for Joe Smith and Bob Jones… What I want this to do is, I want to apply a small pause at the end of each one of these lines so that their contact information will be read in a bit smoother of a flow for the end-user, rather than having it all just be treated as one big sentence. If I apply the break lines checkbox to these two different elements, that is going to improve the end-user’s experience as they read through this page.

The next thing I’m going to do is focus on my heading structure. So if I click on this item here, where it says “Memorandum,” if we remember from the auto-tagging, this was set as an H3. This is our first heading on the page so we really don’t want this to be set as an H3. We want this to be set as an H1 because it is of course a rule for our logical heading structure that all documents have to have a heading level one. And that heading level one should be the first heading on the page. So all I have to do to set this as a heading level one is simply tap “1” on my keyboard.

Now I’ve made some adjustments to the page so far, and just so that we don’t get too far ahead of ourselves if I press the HTML Preview again… I mean this is not something that’s required… you don’t have to press the HTML Preview every time you make a change…

It’s just simply something to show you kind of how things are evolving here in terms of how a screen reader is going to interact with this document as we go. So if you can tell, now our actual contact information for the Joe Smith and Bob Jones… they’ve actually been treated a little bit differently. Where each one of those lines of text is going to be read with a slight pause at the end of it, just making it a bit easier to understand.

For example, where the phone number starts versus where the actual area code ends. The memorandum is now a much larger font because I’ve set this as an H1. So you can see that change reflected here in your HTML Preview, and then everything else below we have not yet impacted. So we still need to make some changes down here. But you can kind of see how things are changing as we go.

If I come back into the main page here, I want to capture this “What is Lorem Ipsum” as an H2. So zones can be easily adjusted. If I just resize that zone and drag it up, I can then grab a new zone, or grab this content by just clicking and dragging to capture that text again. And then if I just simply hit “2” on my keyboard, that will identify that as a heading level two.

Down below here, here’s that bit of OCR. So as you can see, right now it’s sat inside of an image zone. So what I can do is just get rid of that image zone by just hitting backspace on my keyboard. And if I just draw a new zone on top of that text, it’s now a text zone. But I can change the zone source to OCR. OCR, of course, standing for optical character recognition. If I hit the “Convert to Text” button here, this is going to process this basically a screenshot of this area of the page, and it is actually going to turn it into something here in this little text field where you have real encoded text that can be actually read by a screen reader. The alternative to that would be literally typing out every single word of this in the form of an alt text for that image. That’s going to take you a lot longer than simply just pressing the “Convert to Text” button in Equidox.

Over here we have another heading level two, so I’m just going to tap “2” on my keyboard. And this is identifying where this list is located, so this list of course needs to be addressed right now. Again it’s just sat there as a text zone, so it’s going to read this as just one big run-on sentence, which is going to be strange and not make much sense to an end-user. What we can do to fix this if I just hit “L” on my keyboard that will change the zone to a list, and then I can use the List Detection Slider. If I use the List Detection Slider, Equidox is able to automatically identify where those list items are located. And it is basically doing all of the heavy lifting for me. So it’s going to change what was something that is simply a text zone, it’s now going to tag this as a list, and when I export the document, Equidox goes through the steps of building out the complicated tag structure that is required for this list. So that I don’t have to do that manually.

If you remember before, I was talking about how a simple little list like this has only, I don’t know, eight items in it, there are so many steps required to properly tag this. It’s very manual, it’s very technical, and you have to really have a lot of patience and just a deep understanding of what you’re doing to be able to efficiently work through lists like this. So with Equidox, it’s simply a matter of identifying that this zone is a list, and moving the slider from left to right, and you’re done. The next thing we want to worry about here is the table.

As you can see, after I’ve used that Zone Detector the table is still not identified as a table. It’s just kind of grouping the individual columns and rows together in their own individual text zones. So I can fix this by simply ignoring all of those zones and just drawing a single zone over the entire table. And if I hit “T” on my keyboard, that will change the zone type to Table. And I can now open up the Table Editor. The Table Editor is going to isolate the table for me. And if you notice, the great green grid lines here, they’re kind of identifying where the rows are. It’s trying to figure out the basic structure of this table. This is the initial guess that Equidox has given me.

And it did an okay job. I can always make manual adjustments to this if I’d like to, but I prefer to use the Table Detector. So using our artificial intelligence, a combination of machine learning and computer vision, if I move these sliders around, Equidox is going to trigger that AI to automatically identify the layout of this table. Now this table has only column headers and no row headers, so what I’m going to do is change my row header from the default one (we have a default one to one ratio, meaning that you have your top row will be your column headers and the farthest left column will be your row headers).

In this case, I’m going to change this, instead of these being row headers, I’m going to just change them to table data. So if I just change my one to zero, that will update my table summary automatically, which is another thing that Equidox can automate for you. Automatically putting together your table summaries based on how you have structured your table. And the green grid lines to locate the different cells. So after doing that, if I take a look at this preview of just the table, we’ll see that before I had a very strange sort of like series of paragraphs like in run-on sentences trying to read through this table. But I’ve actually very quickly just converted it into a nice clean table.

And again the beauty of Equidox is that it will take this table that I’ve built here, and I can see in my HTML Preview, and it is automatically going to build the table tag structure for me when I export the document as a PDF. So if I save this table now and close out, I will return back to the Page Editor. And all of those zones that were underneath the individual text zones, that were underneath, they have been overwritten. I’m just now left with one single table zone.

Now if I take a look at the HTML Preview for the full page, what I’ll see is things are looking a lot better. But I do have some reading order issues. If you remember, I created this zone manually. This was the OCR zone. And when I created it manually, it took over the lowest available number in the reading zone. So it jumped up to number one because I deleted the zone that was capturing this little thing up here in the header telling me what version of this document this is. So to fix the reading order, it’s really straightforward. I can simply go to the Page Tab and press the “Reorder” button. And that is going to reorder things in a top-to-bottom, left-to-right sequence.

The first thing that will be read is my image here because this is the tallest zone… Then I have the Joe Smith and the Bob Jones contact information, my heading 1, and working down the page just as a top-to-bottom natural layout. If I go back to the HTML Preview, this will be a little bit more reflective of what I would like to see. Where I have down here… I’ve got my list and my tables, everything kind of in the order that I would expect to see it.

And one other little thing that you might choose to adjust… best practice would call to have your heading level 1 be the first thing read on the page whenever possible. So in this case, just to show you another way to impact the reading order… This is currently reading zone four. So if I wanted to make this the first zone read on the page, you might think that I would then need to make turn one into two, and two into three, and three into four… Just so that I can make room for this. But we can actually use the decimal system.

If I just put a period over here in the little order field… if I put a period in front of the 4 that will make that zone 0.4. And that is going to reorder it and put it at the very top of the reading order because 0.4 is of course less than 1. So I don’t have to have a perfect sequence of numbers. I’m able to impact the reading order of this page very quickly.

And now if I take a look at just the full sort of final HTML version of this document, I’m very satisfied with what I see. Everything here is basically exactly what I would expect and how I would anticipate a screen reader should be reading through this page. So I’m done with this document, and I can save it and I can close out at any time. And once I’m done closing out, I could of course export the document as a PDF simply by going to the Output Tab and hitting “Generate PDF.” Now in the interest of time let’s jump back into the slide presentation here.

So we’ve covered the Equidox demonstration, and again if there’s anything else that you’d like to see… if you’d like to have a more like one-on-one type of demonstration where we take a look at your own documents, your organization’s documents, or talk about the existing workflows that you have, please feel free to reach out to us through our contact information which I’ll definitely put up again on the screen. But it’s definitely something that we would love to chat with different organizations about.

We’d love to know how you’re using PDFs, how you’re currently remediating, or if you’re not remediating at all. And if you’d like to incorporate a tool like Equidox into your workflow. In full disclosure, we are, of course, just to keep things kind of simplified, we are just using screenshots. But this is a screenshot of the accessibility report after we have remediated that document. Just to be honest, we’re not actually opening up Adobe here on the slide presentation. We just took a screenshot of another version of this one that we did prior to the demonstration today.

But if you take a look here at the Tag Panel, now on the left-hand side. This is after running the accessibility checker, and there are only two issues located. But keep in mind that these issues will be present in every document that you ever remediate and ever attempt to validate in Acrobat because Acrobat is not capable of checking for the logical reading order or the color contrast of the document. So these are really just sort of warnings or reminders that those types of elements need manual checks. But we have already kind of handled that, especially from the reading order standpoint, by working through the document in Equidox and checking things with our HTML Preview.

For the color contrast, you can always use a third-party color contrast-specific checking tool. But these are things that are always going to be present. But if you can tell now you have a very clean-looking accessibility report and this document is now perfectly usable, perfectly accessible, and ready to be posted or distributed to your website, or any of your clients, or wherever it may be off to.

Just a quick reminder, a couple of the things that we covered in the Equidox demonstration. We went through a couple of our Smart Detection Tools. One click to set our heading levels, we were able to very quickly adjust our reading order, so there’s a number of things that we’d love to chat with you more on a one-on-one basis. If you’d like to have your own personal demonstration of Equidox, again please get in touch with us. And you can get in touch with us by reaching out to EquidoxSales@Onixnet.com. Check out our website Equidox.co and of course, we are always active on social media. So please feel free to connect with us there.

It is 2:30 on the dot. So I want to be respectful of everyone’s time. I really appreciate everyone attending today, and come visit us again next month for our next edition of Equidox Webinar Wednesdays.

[end of transcript]

 

Subscribe

Consent*
Clicking Subscribe will take you to the Thank You Page.
Hidden
Status*
Hidden
Lead Source*
This field is for validation purposes and should be left unchanged.

Categories

Accessibility Culture

Accessibility News

Company News

General Accessibility Resources

Laws and Regulations

PDF Accessibility Resources


Tammy Albee

Tammy Albee | Content Marketer | Onix Tammy joined Onix after four years experience working at the National Federation of the Blind. She firmly maintains that accessibility is about reaching everyone, regardless of ability, and boosting your market share in the process. "Nobody should be barred from accessing information. It's what drives our modern society."