[OSM-talk] Tagging hierarchies (was: RFC - lake)

[Dair Grant]

Frederik Ramm wrote:

>I'm not saying that "natural=water" should be deprecated. I'm just
>saying that if someone wants to introduce a tagging for lakes or
>other *special* kinds of water, then there is no technical
>requirement to tag everything that is tagged with the new lake tag
>(say, water=lake) as natural=water *also*, because the fact that
>"water=lake implies natural=water" could be stored externally.

There is definitely benefit in being able to say "X is a kind of Y".

Any client software using OSM has its own internal model (roads 
can be drawn in one of 7 styles, say) and has to map OSM's tags 
to that model to use it.

At present that means a big list of explicit tag matches, and 
anything unrecognised just has to be discarded.


Another approach would be to have a "conforms to" page, like Map 
Features, which describes how tags related to each other.

Software might not know what a reservoir or a dam were, but if 
it can fetch some rules that teach it what they're similar to 
(say a general body of water) then it can do something sensible 
with them.


>Obviously the option of vague tagging must remain, otherwise the
>lake tag would have to go once the biologists start differentiating
>between various types of lakes, and in the end we'd end up with a
>system where nobody can tag anything unless he's one of the world's
>three experts.

In the above case the software wouldn't have to care about what 
kind of sub-types of lake people wanted to define - if they all 
conformed to "it's a lake" then they can be processed as such.

I've been careful to say "conforms to" rather than hierarchy 
here, since I think that might be the bridge between 
pre-defining every tag for mappers (bad) vs every client having 
to hard-wire a list of stuff they understand and ignoring the 
rest (also bad).

The idea of conformance is that things conform to properties of 
something less specific than them - this is a "like a" 
relationship, not an "is a", so things can have multiple ancestors.


We kind of have this today, where something might be tagged with 
multiple tags to indicate different aspects of its nature.

That helps people capture things as they are on the ground, but 
building the inverse would make it easier to actually do things 
with the data.


I'm not sure what the best way to see if this is feasible would 
be; perhaps just sifting through the tags in planet to see what 
values people actually use.

My guess is that people often use the tags supported by the 
renderers, both because they're common things in reality but 
also because making something appear on screen is more 
fulfilling than capturing lots of data nobody ever sees.

A conforms-to system would allow renderers to use the small 
hierarchy of tags that they have rules for, but also allow 
people to tag things however they liked (provided their tags 
relate to something else, they don't need to adjust their 
tagging to match a renderer).

On the subject of tagging, was there any more development on the 
STAGS idea from SOTM?


-dair
___________________________________________________
dair at refnum.com              http://www.refnum.com/