[Clipart] tag you're it (fwd)

[Alan Horkan]

On Thu, 7 Sep 2006, Bryce Harrington wrote:

> Date: Thu, 7 Sep 2006 01:45:04 -0700
> From: Bryce Harrington <bryce at bryceharrington.org>
> To: Michael Moore <stuporglue at gmail.com>
> Cc: Open Clip Art Library <clipart at lists.freedesktop.org>
> Subject: Re: [Clipart] tag you're it (fwd)
>
> On Thu, Sep 07, 2006 at 01:43:42AM -0600, Michael Moore wrote:
> > >> >True...do you have any ideas? Our new interface would be awesome and
> > >> >easy to add to.
> > >>
> > >> Maybe, at least maybe somewhere to start at.
> > >> http://stuporglue.org/downloads/ocalgame.png
> > >
> > >I like it!
> > >
> > >But maybe there should also be a second feedback loop element, where
> > >people look at a set of potential keywords and try to identify the best
> > >ones.
> >
> > I thought about that, but I couldn't figgure out a good way to work it
> > into the game. I think that if people are typing keywords without
> > knowing what the current ones are, then the best keywords will
> > eventually bubble to the top of the list. That's assuming a count was
> > kept of how many times each keyword was entered.
>
> True, that's possible.  I've always loved the idea of how feedback loops
> can correct for errors in the primary process.  It'd be cool if some
> 'metamoderation' system could be used to control the primary game to
> ensure it remains more interesting.
>
> Another way to think of it is if you're playing against a computer,
> quickly you learn there are tricks to achieving success.  But if a human
> is on the other side, and will be adapting to any predictable gameplay
> you adopt, well then it starts to get very interesting...

The google image labler telepathy game does put some words on an off
limits list after a while, presumably based on their frequency.

I have entirely failed to get a match against the "astronomy" for any of
the space images which come up frequently.  The game is far too heavily
biased in favour of speed over accuracy or quality but I guess as the game
goes on longer and the list of off-limits words covers all the basics
users are forced to provide alternative descriptions.

> As an example, imagine if a clipart of a camera came up.  The person
> knows that it's a Kodak 4321, since she owns the exact model.  But she
> simply types in "camera" as a keyword because she suspects no one else
> in the game will know its model.  But then after the game she'd like to
> add that model name as a keyword.

When screenshots of celebrities come up it is far easier to match
man/woman/person than to actually hope the other player knows who the
actor or character actually is.

> > Example:
> > After the game, I view my results and see that "circle" was one of the
> > tags for the dartboard. I could flag it as wrong since it's not an
> > obvious tag for a dartboard. If circle were flagged as wrong a certain
> > number of times, it would be dropped from the tags list.

With SVG it should be possible in theory to check for things like that,
but in practice Inkscape uses paths for everything rather than circle,
rect, polyline, polygon.  I drew some art using only polygons in an effort
to make things look stylised which makes me think perhaps some researchers
will mass process a copy of the collection and analyse the usages of
shapes and styles, perhaps quantify the artist styles, who knows.

-- 
Alan