When I demonstrated a very simplified RDF/RSS model last week, in the comments attached to the post, Ziv asked the following question:
One question of an RDF newbie: Why do we need that (rdf:Description) element? Why can’t we simply put the @rdf:about attribute on the (item)?
As I started to answer the question in the comments, I kept finding myself taking the question deeper and deeper into the meanings of RDF:
-The rdf:about attribute can’t be used directly on a property led to
-The RDF/XML follows a striped XML syntax of class/property/class/property, regardless of shortcuts led to
-The striped XML syntax is based on the pattern of node-edge-node in RDF led to
-The node-edge-node of RDF is based on a model
All of which led me to a truly definitive question about RDF — why? Why the use of rdf:about here rather than there. Why the syntax? Why the model? After all, XML is a piece of cake — an element here, an attribute there, slam dunk in some text and hey now, we got data. Why make things more complex than they need to be?
Why? Because XML is great about collecting data but is lousy about recording knowledge. There is no facility inherent within the plain vanilla flavor of XML that allows one to write or read assertions in such a way that these assertions (read this as ‘statements’) can be machine produced and machine-readable. And the machines need all the help they can get.
We humans don’t need a rigorous model to communicate. We have phonemes that form words that make up a vocabulary, members of which are then used to form sentences through the use of this really irritating set of rules called “grammar”. We’re programmed to apply these rules through years of instruction, using a neural networking technique called ‘education’. When programming is finished, and after passing certain quality assurance tests, we’re set upon the world. Once loosed from the constraints of the lab, we promptly and as quickly as possible throw out much of what we’ve learned in favor of imagination, creativity, and a dangerous little nugget called innovation.
I love it.
Dorothea wants to discuss her specific mindset related to ‘sexism’ and the concept of sexiness and uses a new word: grunch. This word doesn’t exist, but we as humans adapt to it, add it to our vocabulary (phonemes: grrr + unch). In future writings based on Dorothea’s original discussion, we know what grunch is. Humans adapt.
In 1986, Hans Gabler made 2000 ‘corrections’ to James Joyce’s Ulysses. Well, thank goodness he did because nobody read it the way it was, all those grammatical errors and typos kept getting in the way. Most likely no one even heard of this book until Mr. Gabler took it in hand. As grateful as I am, though, I have recently discovered an even better re-write of this classic: Ulysses for Dummies.
I digress. XML and RDF.
With XML I can record pieces of data such as date, an excerpt, a title, author, category and so on. The structure of the markup allows machines to read these individual facts, to verify that the recording meets certain simple rules. But what if I want a little more than just plain facts. What if I want to be able to take these facts out for a spin, kick the tires, check under the hood?
I have a web page. Facts about this page are: title, URL, date edited, category, and author.
Page has title. Page has URL. Page has edit date. Page has author.
Tarzan has Jane. Jane has Cheeta. Cheeta has banana. A pattern is beginning to emerge.
Every sentence has a subject and a predicate. The subject is the focus of the sentence, and the predicate says something about the subject. These two basic components work remarkably well in allowing us to communicate, to share amazingly complex knowledge.
Returning to RDF and XML, using straight XML is equivalent to only allowing communication with one verb — To Have. Following this, an XML translation of the previous paragraph would be:
Sentence has subject. Sentence has predicate. Sentence has focus. Subject has focus. Predicate has information. Subject has information. Predicate has subject. Components have power. Communication has components. We have each other.
As you can see, after a time, the simplicity breaks down — we need to increase our capabilities, even though doing so adds complexity.
Enter RDF, providing a structure and a meta-language to XML, a grammar if you will.
RDF has one pattern: (subject)(predicate)(object). However, this pattern gives us the tools to record data in such a way that knowledge can be inferred mechanically, merged via a well understood and defined logic with other knowledge, and so on. The subject is the noun, the focus of the statement; the predicate says something about the subject; the object is what is said.
Taking the test paragraph, it can be re-written into the following RDF-like statements:
(Sentence) (has a component)(which is a subject)
(Sentence) (has a component)(which is a predicate)
— no, no, don’t worry — it does get better
(The subject)(is the focus of)(the sentence)
(The subject)(is described by)(the predicate)
(Sentence Components)(enable sharing)(of knowledge)
By providing the ability to record this subject-predicate-object pattern, RDF allows us to expand on the depth of information we gather. The more complex the information, the deeper the pattern is applied, but it is still this triple. In a graphical context, the subject-predicate-object form into a node-edge-node that allows us to build new statements on previously occurring ones.
The focus OF the sentence IS the subject DESCRIBED BY the predicate WHICH IS a component OF a sentence. Consider in this sentence that the predicates are the capitalized value, the graphical notation of this could be: node-predicate-node-predicate-node-predicate-node-predicate-node-predicate. Nothing more than a repetition of our friend the triple, connected end to end.
Representing this within XML requires a set of syntactic rules that ensure we don’t accidentally shove a predicate next to a predicate and so on. There are rules for how to identify a subject, and how to add a predicate. There are rules for how to repeat properties (predicate-object pairs), and how to group properties. There are even rules for how to create a statement about a statement (known in RDF as ‘reification’, though I prefer ‘RDF’s Big Ugly’, myself). But fundamentally the rules break down into nothing more than node-edge-node-edge-node, forming a particularly interesting XML pattern called The Striped RDF/XML syntax.
Rule’s that basically say that predicates can’t be nested directly beneath predicates (edges next to edges) or that whole node-edge-node thing gets blown out of the water. And rules that state when an rdf:about attribute can be applied. In my simplified RDF/RSS, the rdf:about attribute can’t be applied directly to the ITEM element because ITEM in this instance is acting as a predicate, with an implied URI of “item” — it can’t act as a new subject, too. Edge-edge.
(Check out the example RDF files with the RDF Validator to see a graphical demonstration of node-edge-node.)
Once you’ve described one data set with these rules, interferences can be made to other data sets made with the same rules.
As an example, RSS is nothing more than a quick news blurb that gets consumed in less than 24 hours and doesn’t persist. The power of RDF isn’t necessary for RSS used by aggregators, primarily because the data doesn’t persist and one thing about the search for knowledge: it does require that the bits of the knowledge stick around long enough to be discovered.
However, RSS captures a rich set of information about a specific web page or weblog posting: the author and creation date, as well as category, and possibly even links to other resources. What a pity to put this into a form that will only be thrown away.
Well, who says it has to be thrown away? We’s all bosses here, we is. If I says to keep it, I’s boss, and you listen up or Bird be real angry, she will. Real angry. Hissy fit angry.
I modified my individual weblog posting archives to include a bit of RDF in the header that contains the same information used to produce the RSS files that aggregators so callously consume and toss aside. Since this modification was in the template, this RDF is generated for each page automatically. And once persisted in the archive page, it’s there for anyone to discover, providing a richer set of data than just that assumed with keywords pulled from the text.
In this RDF is an identification of the author, an entity which is rounded out by a FOAF (Friend-of-a-Friend) RDF file; knowledge of me, who I am, adds depth and categorization to my Book Recommendation list RDF, and so on and on — a vicious cycle of knowledge acquisition.
(Archived page and comments at Wayback Machine)