RE: [Heartlogic-dev] new idea

[William L. Jarrold]

On Thu, 5 May 2005, Joshua N Pritikin wrote:

> On Wed, 2005-05-04 at 22:22 -0500, William L. Jarrold wrote:
> > On Thu, 5 May 2005, Joshua N Pritikin wrote:
> > > I strongly suggest that we which generate fresh items from OpenCyc or
> > > ResearchCyc, as it exists today, and not rely on items which were
> > > generated by an unknown version of Cyc from the distant past.
> >
> > You are correct in the sense that If we want to make statements about Cyc
> > as is we will need a working Cyc and at least 1000 human judgements.
> >
> > However, we need to walk before we can run.  Consider this a pilot of a
> > method.
>
> Hm, OK

but you deleted my other even better reasons for doing this.  whatever.

> But then where do we get the proof steps to show people after they rate
> the item?

we should be able to get them from cyc.  not opencyc (which by the way
is MUCH MUCH better since the new release) ...recent emails suggest 
thhat the nl generation capability is only accessible with research cyc. 
i'll forward you and josh a msg about the licensing issues asocciated with 
research cyc.  but, like i say below we can still do cool stuff without 
cyc at all AND it will be quite some time before we will want cyc "in the 
loop."

> Let's just skip this part for now?

right.  lets crawl before we walk and walk before we run.

like i said (and you deleted - arg, i hate it when people do that,
espec w/o adding "<...stuff deleted...>" or somesuch, it mutates my
message) we can use this for other questions than "what parts of cyc
kb are valid, what parts synergize."  e.g. you can do it to validate
proposed assertions.  you can also use it to see how believability is
affected by contextual factors.  e.g. if someone sees "if water
touches x then x is wet." alone then we'd predict believability of it
to be higher than when presented in the context of "vienna is wet."
but how strong is this contextual effect...this is cog psychology
research but it may help improve processing of rumination data....hah!
dream on, bill!!!

> The other thing that occurs to me is that when we do this for real with
> a real version of Cyc then we'll probably encounter the same problem as
> with your dissertation study.  Namely that the believability of the
> whole item may be caused by some small part of the proof.  To narrow in
> on the part of the proof which is unbelievable, we might use the same
> kind of flow-chart idea which I proposed for a followup to your
> dissertation.

details details....(-:....yes, a flowchart approach may be the way to
go here and their are other things we can do too.

but i would like other heads than our three to think about this. 
hopefully we can bend peter's ear soon.

there's lots of other details...like i just learned at lunch that 
computing interrater reliabilty (kappa) is no small feat when you have 
more than two raters.

> Are we on the same page here?

i think so.  add that may be the problem.  my (and possibly our) habit
is to start a buncha projects but not push through until the finish.
the biggest sore spot here is that i have not published anything from my
dissertation.  another sore spot is that i have only started but not
finished analyzing the www data we have collected so far....it will be
a long long time before the work is mature enough to really make use
of something like research cyc or km's clib.  but that is the dream
that we will work towards.

> --
> If you are an American then support http://fairtax.org
> (Permanently replace 50,000+ pages of tax law with about 200 pages.)