[Tagging] How to overcome lack of consensus

[David Earl]

> This illustrates, to me, that an attempt to add an ontology on top of
> the tagging is likely to be vulnerable to the problems it aims to
> solve.

Any such thing would want to express relationships, so that e.g. 
renderers which have never seen the tag before can say "aha, they say 
that I could render this like a restaurant if I haven't added an icon 
for it it specifically".

> However, it's possible we could start to have a fairly low-key move
> towards ontology by simply using mediawiki categories. For example if
> you add [[Category:placestogetameal]] to the wiki pages for fast_food,
> restaurant, cafe, then it's rather likely that data consumers can use
> some of the pre-existing mediawiki tools (as used in
> http://dbpedia.org/ for example) to extract structured data that
> expresses tags' relations. So if you start with wiki categories you
> could get fairly far without having to impose a strong ontology.

This seems entirely the wrong way up. Surely much better to start with a 
machine-readable, properly structured, extendable, schema and fderive 
the human readable documentation from it, than try to extract 
information from unstructured data intended for human readers.

The point of a schema is to bring the documentation into the world of 
semantic web - share data not articles, so that programs can do 
something useful with it as well as people. Defaults, varying by 
nationality ("what's the speed limit here if it isn't stated 
otherwise"), descriptions in the same place in multiple languages, 
alternative names for tags in different languages, class hierarchies of 
objects, suggested related properties ("I see you're adding a post box 
here, you might want to add its identifier", "it's a hotel - if you can 
please tell us the operator, how many bedrooms", etc) and so on - rather 
than embedding this knowledge separately and independently and often 
differently in every program that works with OSM.

David