[Data-modeling] [Developers] word puzzle

[Kurt Bollacker]

NOTE: I'm cross posting this to data modeling, because I believe the
discussion should move there.

I am caretaker of a multilingual repository of metadata of Swadesh
words.  For example, I have the word "fire" in probably over 400
languages, including (varingly populated) orthography, transcription,
dialect, and other metadata.  If (some of) these words make it into
Freebase, they'll have to be topics themselves, although I'd love to
also make them display names for the few hundred topics for which I
have words.  Not that there are language settings in the Web UI yet,
but it'd be more natural to access them that way.

								Kurt :-)


On Tue, Mar 24, 2009 at 12:08:49AM -0400, spencer kelly wrote:
> while playing with skud's namesake type and pumpkin's 'word' type i've
> noticed a funny problem with mapping word information.
> it's a 'teddy bear <http://en.wikipedia.org/wiki/Teddy_bear>' in english,
> which is named after teddy Roosevelt, but that doesn't mean in japan the
> word for stuffed bears has anything to do with him.  There are lots of
> things linguists like knowing about words besides etymology (presumably) and
> theres no reason why any of this would be consistant amung translations.
> 
> The topics that exist in freebase cannot map word information cause of the
> wordnet-esque mapping of 'meaning', not words.
> 
>   --a cvs'd namesake property wouldnt work either, because cvs's don't
> interconnect--
> 
> so maybe, the only way to map a topic's word-life correctly is to give words
> their specific topic, essentially multiplying every topic we have by how
> many languages we decide to map.
> crazy. stupid.
> 
> is freebase categorically non-linguistic?
> 
> are there prizes for finding silly bugs?

> _______________________________________________
> Developers mailing list
> Developers at freebase.com
> http://lists.freebase.com/mailman/listinfo/developers