[OSM-dev] Nominatim geocoding for Japanese address

Akiko Goto akiko.goto at onlineconsultant.jp
Fri Apr 11 02:13:31 UTC 2014


Dear Sarah,

Thank you so much. 
It is good to hear;
>Nominatim is now able to handle 'street-less' addresses .

It will help us.

I and Mr.Iida will have a meeting next week, so we will discuss about this.

Regards,
Akiko

> -----Original Message-----
> From: Sarah Hoffmann [mailto:lonvia at denofr.de]
> Sent: Thursday, April 10, 2014 7:56 AM
> To: Troy Wu
> Cc: Satoshi IIDA; dev at openstreetmap.org; Akiko Goto
> Subject: Re: [OSM-dev] Nominatim geocoding for Japanese address
> 
> On Wed, Apr 09, 2014 at 02:23:54PM -0700, Troy Wu wrote:
> >
> > On Apr 9, 2014, at 11:23 AM, Sarah Hoffmann wrote:
> >
> > > On Wed, Apr 09, 2014 at 12:52:49AM +0900, Satoshi IIDA wrote:
> > >>> Japanese address system
> > >> I made Japanese address structure for OSM, last year.
> > >> And got consensus between Japanese mappers. also reported in Tagging
> ML.
> > >>
> https://lists.openstreetmap.org/pipermail/tagging/2013-September/01
> > >> 4816.html
> > >
> > > I rememeber this mail but obviously misunderstood that it is an
> > > finished proposal. I also cannot find much data that follows the
> schema.
> > > Taginfo reports that there are more addr:block_number than
> > > addr:blocknumber but that might be unrelated.
> >
> > And then what about other countries?  In Taiwan, there's the idea of
> main road, alley/subroad, then building identifier.  I'm sure other
> places will have different address formats.  Someone just mentioned that
> the best solution is to create a geocoder for each "type" of address?
> Is that the trajectory of geocoding for OSM?  To have a
> per-country/region based geocoder?
> 
> I sincerily hope not.
> 
> The reason that geocoding works so badly outside Europe is first of all
> a data problem. The only systematic addressing proposal we have to date
> is the Karlsruhe schema and that was indeed developed with European street
> addressing in mind. There is no use in developing a localized geocoder
> until the tagging problem for other addressing schemas is sorted out.
> 
> So if your country addressing doesn't work, by all means, propose
> extensions to the current addressing schema that solve your problem.
> You should discuss these things in your local community first, then
> document it in the wiki and put it out on tagging@ for discussion to find
> out if maybe other countries have a similar problem which can be solved
> at the same time. And please, open a trac ticket for Nominatim to let
> the developers know why addressing in your country does not work and what
> you are planning to tag.
> 
> Of course, there is no gurantee if/when things get actually implemented
> because there are simply too few developers to implement the stuff.
> Again, writing a new geocoder won't really help with that problem.
> It would be much better if the few people interested in this topic would
> join forces and manage to work on one common project.
> 
> > And, someone mentioned that Nominatim isn't good in the US?  I'm trying
> to implement some basic geocoding (all US addresses), and that would be
> somewhat devastating.  Is this actually the case, before I spend time
> going down this path?
> 
> I'm looking at this from another continent, so I might be wrong, but the
> problems with geocoding in the US look mostly like a data problem as well.
> Boundary data in particular is really bad in the US.
> The good news is that the US community has worked hard on improving the
> data in the last year and I'm quite confident that these things will get
> sorted out soon. There are also some particularities with US addressing
> where the tagging still needs to be defined better (i.e.
> postal towns) but apart from that I don't think Nominatim would do that
> badly.
> 
> There is also now the address repository that Ian Deeds announced a few
> days ago. This data could be mixed into Nominatim as external data quite
> easily and further improve the rather patchy house number data.
> 
> If all that in its current state is enough for your needs, I don't know.
> You should run a trail with a small set of data on the publicly available
> instances to find out.
> 
> Sarah




More information about the dev mailing list