I could offer an interface:
https://toolserver.org/~magnus/thetalkpage/
<https://toolserver.org/%7Emagnus/thetalkpage/>
Yes, Magnus, I would also have suggested this next :-)
Mingli, inspired by Magnus' demo, we have indeed been talking about such
QA interfaces to Wikidata a bit already. But as you also note, there is
still a lot to be done for this to work. I agree that reasoning will
play an important role for this (because we don't want to store all
"grandfather" relations in Wikidata when they can be inferred easily).
We are also in contact with NLP people regarding generic language
interfaces ("understanding" questions instead of indexing generated
questions). A number of groups have been doing work in this area using
DBPedia or Yago, but Wikidata adds new challenges due to the more
complex data model.
Markus
On Wed, Aug 7, 2013 at 10:00 AM, Mingli Yuan <mingli.yuan(a)gmail.com
<mailto:mingli.yuan@gmail.com>> wrote:
Thanks, Markus,
About the background:
One is related with my current work and I can not say it too much.
But another story, I can say it publicly.
After playing with wikidata for a while, I realised that we have the
potential to create a WolframAlpha-like application. To achieve
this, what we need are only a indexer, a generator, a reasoner and a
traversor.
Take a look at simplenlg (
https://code.google.com/p/simplenlg/ ),
it can take claim triples and transform them into sentences of
questions. We then index these the answer by question sentences, the
answer could be provided by wikidata claims.
And then we get a QA engine with weak abilities. i.e.
* they only know who are Enistein's father and children, and do
not know that grandfather is the father's father.
* they only know which planets belong to the Sun system, and they
do not know which is the biggest, the farest, etc.
So if we add some axiom into the system and also with a reasoner and
a traversor. We have the potential to enumerate all possible
*simple* questions and answers of human knowledge. Then we get a QA
engine with strong abilities.
These are what in my brain now. I know these kinds of things are
never easy, but they are possible in the near future.
Regards,
Mingli
On Wed, Aug 7, 2013 at 4:14 PM, Markus Krötzsch
<markus(a)semantic-mediawiki.org
<mailto:markus@semantic-mediawiki.org>> wrote:
Hi Mingli,
thanks, this very interesting, but I think I need a bit more
context to understand what you are doing and why.
Is your goal to create a library for accessing Wikidata from
Clojure (like a Clojure API for Wikidata)? Or is your goal to
use logical inference over Wikidata and you just use Clojure as
a tool since it was most convenient?
To your question:
* Do we have a long term plan to evolve wikidata towards a
semantic-rich dataset?
There are no concrete designs for adding reasoning features to
Wikidata so far (if this is what you mean). There are various
open questions, especially related to inferencing over
quantifiers. But there are also important technical questions,
especially regarding performance. I intend to work the theory
out in more detail soon (that is: "How should logical rules over
the Wikidata data model look work in principle?"). The
implementation then is the next step. I don't think that any of
this will be part of the core features of Wikidata soon, but
hopefully we can set up a useful external service for Wikidata
search and analytics (e.g., to check for property constraint
violations in real time instead of using custom code and bots).
Cheers,
Markus
On 05/08/13 17:30, Mingli Yuan wrote:
Hi, folks,
After one night quick work, I had gave a proof-of-concept to
demonstrate
the feasibility that we can combine Wikidata and Clojure logic
programming together.
The source code is at here:
https://github.com/mountain/__knowledge
<https://github.com/mountain/knowledge>
An example of an entity:
https://github.com/mountain/__knowledge/blob/master/src/__entities/albert_e…
<https://github.com/mountain/knowledge/blob/master/src/entities/albert_einstein.clj>
Example of types:
https://github.com/mountain/__knowledge/blob/master/src/__meta/types.clj
<https://github.com/mountain/knowledge/blob/master/src/meta/types.clj>
Example of predicates:
https://github.com/mountain/__knowledge/blob/master/src/__meta/properties.c…
<https://github.com/mountain/knowledge/blob/master/src/meta/properties.clj>
Example of inference:
https://github.com/mountain/__knowledge/blob/master/test/__knowledge/test.c…
<https://github.com/mountain/knowledge/blob/master/test/knowledge/test.clj>
Also we found it is very easy to get another language
versions of the
data other than English.
So, thanks very much for your guys' great work!
But I found the semantic layer of wikidata is shallow, that
means you
can only knows who are Einstein's father and children, but
it can not be
inferred automatically from wikidata that Einstein's father
is the
grandfather of Einstein's children.
So my question is that:
* Do we have a long term plan to evolve wikidata towards a
semantic-rich dataset?
Regards,
Mingli
_________________________________________________
Wikidata-l mailing list
Wikidata-l(a)lists.wikimedia.org
<mailto:Wikidata-l@lists.wikimedia.org>
https://lists.wikimedia.org/__mailman/listinfo/wikidata-l
<https://lists.wikimedia.org/mailman/listinfo/wikidata-l>
_______________________________________________
Wikidata-l mailing list
Wikidata-l(a)lists.wikimedia.org <mailto:Wikidata-l@lists.wikimedia.org>
https://lists.wikimedia.org/mailman/listinfo/wikidata-l
--
undefined