[OSM-talk] Semantics layer for tags

[Martijn van Exel]

On Mon, Jan 10, 2011 at 2:50 PM, M∡rtin Koppenhoefer
<dieterdreist at gmail.com> wrote:
> 2011/1/10 Martijn van Exel <m at rtijn.org>:
> ...> is in cultural and language differences. What constitutes a trunk road
>> in Lithuania? What is a chemist in Spain? Not all tags even translate
>> one to one.
>
>> Ideally we would have a semantic layer between the user and the
>> database / API. This layer would comprise of an ontology of geographic
>> feature representations in different languages, think a structured
>> version of the different language versions of the Map Features page.
>> The ontology would also include synonyms of feature representations...
>
>> A semantic layer would solve a lot of problems with tagging ambiguity,
>> break down language barriers,
>
>> Thoughts? Are we doing this already?
>
>
> IMHO we are already doing something like this in our wiki, but the
> main problems won't be solved hereby. The main problems are the
> details ;-).
>
> Of course you could have a list (or ontology) that says that
> amenity=fuel would be de:Tankstelle, but it wouldn't tell you that you
> can at almost all de:Tankstelle get compressed air for your bike, or
> cigarettes at night, while in many other countries this is not a valid
> assumption.

Yes, we're trying to maintain the wiki with language versions of the
Map Features, but that does hardly solve any problems of accessibility
that we're facing because different Things in reality are represented
and classified differently in different cultures and languages. A
semantic layer between the database and the API (or in the API) would.
It could even play a role in describing the implied attributes that
you are talking about - consider maxspeed defaults for example.

And no, ontologies are not magic wands. At least not Harry Potter
grade ones. They don't free us from having to think about editing, but
they would make it easier to provide a local culture / language
interface to OpenStreetMap editing.
>
> This is just an example, but you will have these assumptions for most
> of the tags: for the local mapper they are included, but on a global
> basis they won't be valid. The meaning of a tag is somehow always
> dependent on the cultural background / area.

Yes, that is exactly where a semantic layer would come in! For
example, I would tag a feature in a semantics-enabled JOSM in my
native language, Dutch, as "provinciale weg". A lookup in the ontology
would expose an ambiguity: a provincial road could be highway=primary
or highway=secondary, depending on the road number. Human
disambiguation would be required, the attributes of the semantic
relation between 'NL:provinciale weg' and 'highway=primary' and
'highway=secondary' could provide a clue to do this. In other cases,
it could be automated based on the context. For example, if the road
number was already entered by the user.

Many other cases would be non-ambiguous, I'm just picking an ambiguous
relation to explain how that could work.

Martijn

Martijn van Exel +++ m at rtijn.org
laziness – impatience – hubris
http://schaaltreinen.nl | http://martijnvanexel.nl | http://oegeo.wordpress.com/
twitter / skype: mvexel
flickr: rhodes