As many of you know (all, like, 12 subscribers or whatever there are here), I took a job with Metaweb, who run Freebase, in January this year. So since then I’ve been in charge of the Freebase blog and haven’t been blogging here.

I’ve been wondering what to do with this domain, and I’ve decided to keep it running and to use it for personal/unofficial notes on Freebase-related topics. While the Freebase blog is a sort of official channel, I’ll use freebasing.org for less formal topics.

So, I suppose a disclaimer is in order here: Views and opinions posted on this blog do not necessarily represent those of my employer.

You know how when you look at a topic, some fields are in the main area of the screen just near the image, and some are on the right hand side?

I used to think that anything that permitted multiple values went on the right, but it turns out that’s not so. Here’s the skinny on what’s really happening, thanks to Jeff:

You can actually specify which column a property appears in in the schema editor. Left is “horizontal” (because multiple values are displayed next to each other, separated by a comma), right is “vertical” (because multiple values are displayed above and below each other). The default layout is that properties for which the expected type is a primitive type (date/time, integer, text, etc.) display on the left, and everything else displays on the right.

(from this discussion about countries’ official languages.)

So, to summarise:

System primitives on the left.
Horizontal multiples on the left.
Vertical multiples on the right.
Non-primitives, even if single, on the right.

On Wikipedia, one of the great tensions is between deletionists and inclusionists. The debate is over whether Wikipedia should include everything, or only things that are important enough for a “real” encyclopedia.

On Freebase, we don’t have this problem. As far as I can tell, Freebase is about as inclusionist as it’s possible to get. I usually express this to people by saying, “As far as I can tell, they want 6 billion person topics — or probably twice that, to allow for dead people.”

On the other hand, the first tension I’m seeing emerge on Freebase is that between normalisers and normal people, and I’m seeing it play out in a few discussions that are going on at the moment.

In the comments on my post about gender I linked to a thread on the data modelling mailing list: Should some of the Person attributes be moved to Organism?

Here’s what I said in comments:

Aristotle: There was a related discussion thred: Should some of the properties of Person be moved to Organism?

Someone facetiously commented, “I immediately began wondering which celebrities would get marked as hermaphroditic or asexually reproducing”, which of course is one of the risks of over-generalising.

I think there’s a tension in Freebase, taking a very broad view, between models that tend towards abstraction and those that tend towards how people think day-to-day. We encountered this when Jeff P was modelling Government, and had an abstraction which was kind of theoretically valid across many government forms, but left some of us (esp. those from countries whose governments use the Westminster system) scratching our heads and saying, “but nobody in Australia/Canada/UK *talks* like that!”

In the case of sex/gender of human beings, there’s probably 99% of cases where people just want to be able to say male/female, and confusing them with a list that includes “intersex” and “asexually reproducing” would just boggle them. Male/Female/Other is a reasonably well-precedented pattern for use in forms etc (though I’d want to get some trans people’s views on that to make sure it’s what they’d want) while not being overwhelmingly technical.

As mentioned, we had a similar problem with governments. Jeff Prucher had put together an abstract model that was theoretically workable but which didn’t match well with the way I thought about government. I found myself having to twist my brain to make things fit. I’d get there eventually, but I couldn’t enter the data naturally and smoothly. It was like having to translate into a different language on the fly.

Jeff’s since put together a second government test domain which seems to work better for a wider range of government styles. The differences are relatively small, as far as I can tell without being able to look at the first model (it’s been made private again), but the difference to me as an end user are like night and day.

From what I recall of the two government models and the changes between them, and from other normalizers-vs-normals discussions that I’ve seen, I’d like to propose some techniques that seem to minimise the conflict:

• Talk like a lay-person. The use of good type and attribute names, ones that don’t require translation by users, makes a model friendlier to “normals”. Even if you make no other design changes, this can make a big difference.
• Support the vast majority. Like Perl’s motto of “make the easy things easy and the hard things possible”, you should first try to support the vast majority of cases in a way that’s easy to use, then deal with the outliers. Make your main type support the vast majority, and make a second type for the rare misfits and co-type as necessary.
• Love your CVTs. The extensive use of Compound Value Types seems to be a common feature of models that feel natural to normals while still being elegant data modelling solutions.

Anyone got any other suggestions?

“I KNOW LETS MAKE A TYPE FOR UNICODE CHARACTER!!!1″

Too many types. Must sleep now.

I wanted somewhere to blog about Freebase, Metaweb, and the like, so here it is.

I’ll be your host for this blog. My name’s Kirrily Robert, commonly known as Skud, and you can find me on Freebase under that name. My interests so far on Freebase have included tall ships and Australian location data.