A battle of Beliefs: RDF, Natural Language Processing, and the future of the web

Last Week in HTML has been practicing its wicked ways, and pulled a quote from a comment I made to a post at Sam Ruby’s

Ian is wrong. Absolutely, completely, and dead wrong.

rather than Ian shouting out “Hurrah!”, he says we must have five different solutions to the five problems, because to do otherwise is to…what? Give up control? Fail to meet the Guinness Book of World Records for largest, most pedantic specification ever derived by man?

At first glance, this seems a repetition of an argument that is growing thin with overuse, but the recent discussions in the RDFa mailing list, about RDFa in HTML5, provides a clear demonstration of the basic disconnect between the parties. Enough so to make it of value to re-visit the discussion, again.

On the one hand, you have RDFa, which is a serialization of RDF, which is a formal data model providing support for a universal form of structured data. On the other hand, you have those whose ideology for the future of the web is based on natural language processing. This is an old, old battle and one we’ve been fighting since RDF was first proposed—prior, really, as I remember working ideological differences between natural language processing, as compared to structured data techniques, in various projects at Boeing in the 1980s.

One would think, then, considering the age of the debate that we wouldn’t fight this old battle in the lists for HTML5. Why? Because it exists above and beyond just HTML5. It is a debate about the fundamental nature of the web, at its most general and profound level, while HTML5 is really nothing more then the next generation of HTML. However, we are fighting this macro battle out in the micro lists of HTML5, but deceptively so.

Those who support RDFa have been continuously asked to provide use cases for RDFa, and have created a wiki page to record these use cases. But each time the use cases are proposed, we’re given a response that the use cases are inadequate, and different sets of criteria for how these use cases can be “improved”. It is frustrating to the RDFa adherents, stumbling about in the dark hoping to hit exactly the right “fit” in order to satisfy these never-ending requests.

In the new thread, though, the underlying ideological differences are peering out through the fabric of technical obfuscation, and we see the real purpose behind the demands for RDFa to justify its existence in HTML5. We’re not being asked to justify RDFa in HTML5; we’re being asked to justify RDF, and beyond that, we’re being asked to justify the concept of structured data. Not just once, but for every instance of a use case.

Ian Hickson writes in one comment in the mailing list thread:

I wouldn’t worry too much about the various solutions in each case — a list of solutions can never be complete, and people will never agree on what consists a pro and a con. What would be useful, though, is an example of how RDFa is expected to solve the problem, e.g. with sample markup showing how the relevant data might be encoded and code snippets showing how the data would then be processed; and a discussion of ways to deal with the likely problems (e.g., for this particular use case: how to deal with authors screwing up and encoding bad data, how to deal with apathy from sites that you want to scrape data from, how to deal with malicious authors encoding misleading data, how to deal with spammers, how to deal with requirements like Amazon’s desire to track per-developer usage, how to enable monetization for producers who are intentionally obfuscating the data today, etc. I expect other use cases will have different problems).

The first set of requests are reasonable and have been demonstrated. I use RDFa in my site to document each post with a formal title, author, date, and set of topics, each of which can be extracted using a PHP API that I’ve installed at my site. I plan on using this data in order to generate my front page eventually. This same data can be extracted with a Firefox toolbar, too, if I’m so inclined, and used to output an RDF document for other’s to consume. The data has also been extracted as part of Yahoo’s SearchMonkey effort, I do believe.

Others have provided examples of the Creative Commons licenses, and FOAF, and other uses of RDF/RDFa. Not only the purpose behind the use but even demonstrations of how the data can be combined across pages. These seem to meet the requests for demonstrating code to both incorporate the RDFa in HTML5, as well as code to pull such data out.

As for authors screwing up and providing bad data, well I have to assume the same mechanisms in place, in the browser, when a person inputs bad data into an alt attribute (if it survives in HTML5) would be in place for bad data in a property attribute. And if the data is coded incorrectly, applications expecting valid RDFa wouldn’t be able to process the data, but that’s little different than applications not being able to process a bad script, or malformed piece of SVG, or even a crappy video file, embedded in the page.

The questions I just responded to are legitimate questions. They serve a purpose, and a person can determine by looking at these questions what needs to be provided to ensure the success of the use case. But then we start getting into murkier territory. Ian asks, how to deal with apathy from sites that you want to scrape data from, how to deal with malicious authors encoding misleading data, how to deal with spammers, how to deal with requirements like Amazon’s desire to track per-developer usage, how to enable monetization for producers who are intentionally obfuscating the data today, …

My god, how do we deal with these on the web today? HTML, itself, fails badly with all of these, so do we give up on HTML? If not, then why are we demanding a state of rigor from RDFa that we’re not willing to apply to HTML5, itself?

If you think this latter set of questions were tongue-in-cheek, perhaps a bit of markup levity, Ian repeats them, later, in the same thread

Do we have reason to believe that it is more likely that we will get authors to widely and reliably include such relations than it is that we will get high quality natural language processing? Why?

How would an RDF/RDFa system deal with people gaming the system?

How would an RDF/RDFa system deal with the problem of the _questions_ being unstructured natural language?

How would an RDF/RDFa system deal with data provided by companies that have no interest in providing the data in RDF or RDFa? (e.g. companies providing data dumps in XML or JSON.)

How would an RDF/RDFa system deal with companies that do not want to provide the data free of charge?

How would an RDF/RDFa system deal with companies that want to track per-developer usage of their data?

One could ask all but the first question about HTML, and not find satisfactory answers. Yet we’re being asked to provide sufficient answers to these questions for a small subset of attributes in HTML5, which would form the basis of support for RDFa. As for the first question, Do we have reason to believe that it is more likely that we will get authors to widely and reliably include such relations than it is that we will get high-quality natural language processing?, this, again, brings us back to a fundamental differences in ideology, natural language processing as compared to structured data, and how can one deal with such profound differences in something like a use case?

To repeat what I said earlier, the issue isn’t about RDFa in HTML5. It is about the existence of structured data on the web. It is the underlying purpose behind RDF. It calls into question a decade’s worth of work, based on the input of hundreds if not thousands of developers and designers. It is questioning the fundamental separation of ideology between the web of the future based on natural language processing and the web of the future based on structured data. But where the structured data folks, those who support RDF, and RDFa, welcome natural language processing as a complementary process, the natural language processing folks seem to see the very existence of structured data woven into web documents to be anathema.

Now, someone tell me how we can break through this wall with use cases?

Dan Brickley chastises those on the RDFa group who see this as a battle, writing

This is not a battle. Battles kill people. It is a dispute amongst technologists who have varying assumptions, backgrounds, collaboration networks and agendas, and who are slowly learning to see each other’s perspective.

Please (and I am very serious here) stop using such bloody metaphors to describe what should be a civil and mutually respectful collaborative process. You will not improve anything if you foster this kind of perspective on our shared problems. Battle talk results in a battle mindset. I do not want to hear any RDFa advocates talking in such terms.

Really, enough with the battle stuff. Go find someone who works on HTML5 and be nice to them, find common ground, try out their tools.

Play nice…try out their tools.

I have tried the tools, and in fact just tried the HTML5 validator with the SVG, MathML, and RDFa (minus Curie) preset, and aside from the fact that it tossed my DOCTYPE, didn’t like my profile attribute, some of my meta elements, and the use of “none” as a value for preserveAspectRatio in my SVG, the validator had no problems with any of my RDFa. I would have to assume, then, that we have seen a demonstration of RDFa in HTML5…and found it good? And lo and behold, the RDFa extractors have also found the same page, and the same use of RDFa, to be good. Hands across the water.

But evidently, not sufficient. What else must we do to play nice? Well, Sam has laid out the “nice filter” in comments to his post that began this particular thread

What would it take for inclusion of the RDFa attributes in HTML 5 to be tracked in the W3C HTML Working Group issues list? Given the links I provided at the top of this post, I’d say that pretty much all of the pieces are in place except for a discussion on the public-html mailing list.

What work would be helpful in getting this to be resolved successfully? Fleshing out the use cases addressing as much of these concerns as are relevant.

How can you help? Join the WG and/or contribute to the wiki.

Just so that it is clear, as we move towards summer I plan to become ruthless in clearing out issues which have been raised but don’t appear to have any substantive proposals or support. There is much good work in HTML5 and it would be positively criminal for it not to advance due to procedural maneuverings. I don’t intend to let that happen either.

And this then leads us back to the questions posed by Ian, above. For each use case, must I then justify RDF? Structured data? Must I give details about how spammers will be vanquished, and evil corporations not allowed to monetize such effort? Must I provide a 12-step program in how to lure the reluctant microformat user into the fold? Does the fact that Virgin Mobile misused the Creative Commons license to publish photos of people without getting model releases, mean that the use of RDF/RDFa to document a Creative Commons license can never be a valid use case? After all, it fails the evil corporate use case requirement being demanded of RDFa.

There seems to exist a gentleman’s agreement in these specification email lists, whereby the participants humor absurd questions such as those proposed by Ian. Well, thank goodness I’m no gentleman.

If the RDFa in HTML5 adherents will be required to provide not only justification for RDFa, but also justification for RDF, as a whole, in addition to a dialog and debate about the fundamental differences between natural language processing and structured data with each and every use case, then I fail to see the “niceness” supposedly in play here. It’s difficult, too, to see exactly what we’re supposed to do to bring about this so-called “common ground”. Ultimately, structured data people see natural language processing as complementary, and that there is room on the web for both ideologies. The natural language processing folks see structured data as competitive, and that the web of the future will be based on one or the other, but not both. How do you work through that kind of difference?

Print Friendly, PDF & Email