Raw

Ok, I'm looking ahead a little here in a bit of my personal knowledgebase organising pseudo-project (written down here before I forget). Early on I'll be wanting to create identifiers (/IFPs) for things, including concepts. Later on I want to use heuristics to discover implicit relationships between them. I have plans… But without worrying about all that just yet, I will definitely want to be able to capture as much information as possible regarding characteristics of things. I've got all the usual vocabs available for explicit stuff, including dc:description etc. for wordy literals, and the W6 vocab covering the more general, human side of properties. What I haven't yet got is something to cover another aspect of description which I want available to heuristics, where things can be described on bipolar scales. Rather than trying to pin this down in some precise way, I intend to go with the flow of human-oriented description. Here's a sketch of a first attempt at expressing this:

<Resource> hasTrait [
   skos:inScheme <uri for my scheme>;
   skos:prefLabel "label"; 
e.g. "temperature";
   level "X";  
(-1 < X < 1)
   scaleHigh "label"; 
e.g. "hot"
   scaleLow "label"; 
e.g. "cold"
   traitGroup <group uri>
] .

With Trait rdfs:subClassOf skos:Concept

The thing being described will have other descriptions and I'm hoping they'll be enough to give a context for the scale. So if the thing is a drink then 'temperature, 0.9′ (i.e. very hot, 90ºC) will mean something different if it were a star (i.e. very very hot, 90,000ºC) . I'm not after physical, semiotic or even logical precision here, just trying to get enough to be able to help discover associations that might make sense to humans.

I'm going to have to reread the N-ary Relations doc and around SKOS, but I think this should capture what I'm after. I'll probably write a simple form UI for this, try out examples once I'm happy I've got the structure roughed out.

One bit which might make an interesting experiment is the traitGroup part. Yesterday on a walk I was telling Caroline about this stuff, as one of the ideas I want to play with later was hers ("Interstices" - I'll describe when I get to it). She pointed out the way adjectives in English have a certain order: "Suzan has long brown hair" sounds ok, "Suzan has brown long hair" creepy. Caro reckoned it was probably a good trick for poets, the emphasis really changes: "Suzan has a blue small car" sounds like the car's miserable. This morning Caro pulled out some grammar books which tell how the order of adjective of quality generally depends on the type of adjective, roughly: opinion, size, general description (excluding emotion/personality), general description (emotion/personality), age [plus 'little'!?], shape, colour, origin (e.g. nationality), material composition, purpose (gerunds - walking stick). I've no intention of diving into the natural language side of this, but these do look like they might make useful groupings for the characteristics. What I'm not 100% sure about is whether it would be better to use subclassing - ColorTrait - or a tag pointing to the group. The former might well be easier to reason over, but I've a feeling the modelling might get messy.

As always, suggestions welcome.

[Danny]