cygri’s notes on web data

Some stuff that isn’t in SPARQL, but should be. I’ve two use cases in mind: building AJAX applications on top of a SPARQL store, and enabling discovery of the stuff that is in a SPARQL store you don’t know yet.

Expressions in SELECT: I want to be able to say SELECT DISTINCT LANG(?object) or SELECT DISTINCT DATATYPE(?object). This would also make extension functions much more attractive. (Update: Ideally, expressions sould also be allowed in CONSTRUCT blocks. This has many interesting applications if used with extension functions.)

More functions: NAMESPACE returns the namespace part of the URI. LOCALNAME returns the local part of the URI. CONCAT concatenates literals. SUBSTR returns part of a literal. MATCH extracts bits of a literal according to a regular expression. IF would select a return value based on a boolean expression. I could go on. These can (and probably will) be done as extension functions. As I said, to make extension functions truly useful, they’d have to be allowed in the SELECT clause.

CONSTRUCT *: The motivation here is extracting subgraphs from an RDF graph. I think this is hugely important, and I don’t understand why this is not allowed. I heard something about issues with blank nodes coming from multiple named graphs, but I’m sure this could be worked around.

COUNT: I understand there are semantic issues with this, but it would be super handy in some situations. Currently, if you want to know the number of triples stored in a datastore, you have to fetch them all. Does this suck or what?

Anyway, SPARQL is great and I’m really looking forward to seeing it supported in a wide variety of RDF tools.

Student job offer (in German)

The job is in the Semantic Web for Pathology project. I’ve previously written about it (in German).

Here’s what I dreamed last night. (True!)

This is a scene from a James Bond movie. One of the old ones, starring, maybe, Roger Moore as 007 and Gert Fröbe as the Goldfinger-like villain. This is towards the middle of the movie. Mr Villain has invited our charismatic top spy to his luxury boat and just welcomes him on board. The two well-mannered cosmopolitan gentlemen exchange pleasentries. We know, of course, that Mr Bond has a hidden agenda: He wants to find out how Mr V. plans to destroy the world.

Mr V., however, is preoccupied: His henchmen, a disorganized group of misfits and dimwits (but well-dressed), don’t seem quite up to the job of running business aboard the ship. There’s supposed to be dinner, but no one knows where and if it’s ready. Mr V. is annoyed.

Here, my point of view changes from an uninvolved observer to that of one of the henchmen. While the boss and his English guest continue to plod along the script, I’m hurrying through the belly of the ship, trying to find out who’s in charge and what I’m supposed to do. No one seems to know though. Just like me, everybody has just been recruited. It’s quite a diverse bunch — I meet some English, Spaniards, Italians, French, Germans, and a handful of eastern Europeans.

Word has it that there will be a team meeting in the ship’s spacious bar area. Slowly people start to arrive. All are twenty- or thirtysomethings, some women. I actually know a few folks, they are fellow Semantic Web hackers I’ve met over the years. As we and the new people introduce each other, which is not easy as not everybody speaks English, it turn out that all of us are youngish researchers, developers and students who have bet their carreers on the success of the Semantic Web.

Unfortunately, before I or Mr Bond can figure out why a supervillain would hire a ragtag band of ex RDF hackers, the story drifts away on a tangent about newspaper Berliner Zeitung defending the freedom of the press by reporting on an effort of the German Secret Service to suppress publishing of unflattering photographs of the new German chancellor Angela Merkel.

Weird.

TriG is a Turtle-based serialization syntax for RDF datasets. The only implementation I know of is in NG4J. An example:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix dc: <http://purl.org/dc/elements/1.1/> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .

{
    <http://example.org/bob.rdf> dc:publisher "Bob" . 
    <http://example.org/alice.rdf> dc:publisher "Alice" .
}

<http://example.org/bob.rdf> {
    [] foaf:name "Bob"; foaf:mbox <mailto:bob@example.org> .
}
 
<http://example.org/alice> { 
    [] foaf:name "Alice"; foaf:mbox <mailto:alice@example.org> .
}

This is an RDF dataset. The default graph contains metadata about the two named graphs. TriG is quite useful to write down small examples like this.

The point of this post is not to promote this nice syntax, but to act as a reminder about some minor syntactical issues with the current TriG spec. I’ve culled them from several email threads. This post summarizes my position, and I should probably prod Chris (who maintains the spec) to do something about it at some point in the future.

Namespace declarations

TriG currently allows namespace declarations (the @prefix lines) before, after and between graphs, but not within. I argue that namespaces should be allowed within graphs. This would make any valid Turtle file a valid TriG graph and vice versa, which is a Good Thing for several reasons (simpler spec, easier implementation, easier authoring).

Graph naming

TriG originally didn’t allow unnamed graphs, but after the RDF dataset was adopted for SPARQL, TriG was changed to allow a default graph.

Chris has gone on record saying that these syntactic variations should be allowed for named graphs:

<uri> { ...triples... }
<uri> :- { ...triples... }

And this is for the default graph:

{ ...triples...}

Currently, the spec also allows this:

:- { ...triples... }

This is a bug.

N3 compatibility

There was some push from Jeremy Carroll to maintain compatibility with Notation 3‘s syntax for formulae, which also uses curly brackets to denote graph-like structures, so we discussed adding some more options for named graphs:

<uri> { ...triples... } .
<uri> :- { ...triples... } .
<uri> = { ...triples... }
<uri> = { ...triples... } .

and for the default graph:

{ ...triples... } .

But of all these, only two are actually syntactically valid N3:

<uri> :- { ...triples... } .
<uri> = { ...triples... } .

So, if we want to align the spec more closely to N3, then we should allow only these two, and maybe this one

<uri> = { ...triples... }

because of symmetry.

(N3’s non-RDF (non-Turtle) features like formulae are quite obscure, and I’m not really convinced that compatibility with them is worth the additional complexity.)