Allow localised/translated street type search

[richlv]

Nominatim special phrases functionality allows translating object types (for example, "atm").

It would be great to also support translation of street types.
This would make the search more accessible, and reduce the incentive to translate all street names in all the languages.

For example, "Brīvības street" would find "Brīvības iela" - "iela" being Latvian for "street".

Perhaps some consideration might be needed for offset or circular words - for example:
"street" in Latvian is "iela"; in Lithuanian - "gatve".
In Latvian "gatve" means "avenue". In Lithuanian, "gatve" seems to be "alėja".
But then Latvian has "aleja", which also seems to translate to Lithuanian "alėja".

These might not be a concern at all, no clue how the implementation might work.

[lonvia]

This problem exists not only for streets but pretty much all names in OSM. Another prominent example are the translations for administrative boundaries, e.g. in Poland.

So I'm thinking about replacing the special phrases (which are mainly for searching a POI by its type) with more generic category words. This is a rather vague idea at the moment but would work somehow like that:

• Somewhere there is a list which states which tag combination (potentially in which language and or country) triggers a category word. For example, a Lativian name with highway=* has category words "iela", "gatve", "aleja".
• When Nominatim imports data and parses an object that matches the tags in the list, it matches the name against category words and creates an additional artificial name with the category removed. So "Brīvības iela" would also get indexed as "Brīvības".
• When searching, category words are detected as well and handled as optional. So the search for "Brīvības street" would match against the indexed "Brīvības" because 'street' is listed as a category word for highway=* in English.

This solution would solve some issues with translation, around the question if a category word is part of the name or not ('Hotel Bellevue' vs. 'Bellevue') and also issues in Spain, where the street category is preferably left out.

However, there are a whole can of unsolved questions for the realisation. Not least the problem that in some inflected languages you can't expect to just remove a word and have a sensible new name (look at the translation in the Polish example).

[richlv]

That sounds great - Estonians also leave out street category.
Not quite sure about the word removal resulting in a less descriptive name. Didn't grasp it from the Polish example. What would be an example of it being a major problem?

[i-ky]

in some inflected languages you can't expect to just remove a word and have a sensible new name

Would it be possible for Nominatim to run short code snippets (in Lua, Python, Javascript, whatever easily embeddable language) to come up with "artificial name" for indexing? I am pretty sure that Polish example ("powiat zgierski" → "Zgierz County") can be coded relatively easily.

[lonvia]

Nominatim has a framework now for plugging in Python code to do arbitrary preprocessing of names. This would go somewhere in there.