On 19.08.2014 16:13, Lydia Pintscher wrote:
On Tue, Aug 19, 2014 at 11:19 AM, David Cuenca
Thanks for the stats, Gerard. Two thoughts:
- With so many items without description I wonder why we don't have the
automatic descriptions gadget enabled by default.
I am a bit worried about enabling this by default for everyone as a
gadget. We need the descriptions in a lot of places where people
search for items. The next big one will be Commons. But _a lot_ more
will come in the future. Think for example of tagging your blog post
on Wordpress with Wikidata concepts. You'll need the descriptions. If
we enable automatic descriptions on Wikidata now we will actively
discourage people from entering more descriptions. That would be bad
as 3rd parties then don't get the benefit of them.
I am also hesitant to build this into Wikibase directly as it'd need
quite some domain-knowledge for all I can tell at this point. That's
something we need to avoid.
Anyone got ideas how to get out of this?
It had been suggested recently on this list to store the
autodescriptions in the data, e.g., using a robot. The question there
was whether this would make future automated update too troublesome
(since one would have to check if the description has been overwritten
by a human in the meantime). I think this can be solved (see below). For
thousands of non-described items this would be a large improvement.
An important point is that there are really two kinds of "descriptions"
that we should keep separate, since they have two different purposes:
(1) to provide a clue for distinguishing items with the same label
(2) to give a human-readable, informative summary of the data
The descriptions that we want to have stored on Wikidata are there for
the first purpose ("type-1 descriptions" :-). Their main virtue is to be
as much to the point as possible, so you can read them quickly in a
small dropdown menu etc. (short and accurate, but just enough
information to clarify what we are talking about).
Descriptions of the second kind are a completely different issue. They
should not be stored on Wikidata (or anywhere), since they will
continuously evolve. The more data you have, the better your type-2
description will get. For new kinds of data you will even have to extend
the code that generates the texts. Also, these descriptions could be
much longer, and indeed their "optimal" length would vary from
application to application.
This is why I think that it is fairly safe to import (some) type-1
descriptions without this reducing in any way the importance of type-2
descriptions. Of course even type-1 descriptions will change over time
(esp. for living persons or ongoing events), but most of them should be
fairly stable (cities, species, many people, ...). Since we are only
interested in a very concise description for this purpose, it might be
possible to see if an item has enough data yet to create an "ultimate"
description (in the sense that more data would not have an effect on the
concise description anyway). Magnus will know more about this.
A simple, low-tech way of enabling future updates would be to store the
list of imported descriptions somewhere so that future update robots can
check if the description has been changed by someone in the meantime
without having to consult the page history.