RE: [Heartlogic-dev] OHL v2 alpha test

[William L. Jarrold]

On Tue, 22 Mar 2005, Joshua N Pritikin wrote:

> On Tue, 2005-03-22 at 00:30 -0600, William L. Jarrold wrote:
> > * most of the text i had no idea what it meant....e.g. this...
> >
> > "orange juice is a matter of might (as opposed to law)."
> >
> > ...huh?  might, as in might makes right instead of rule by law
>
> Yes.
>
> If daddy doesn't give Toby the orange juice that he's after then on what
> basis is daddy able to deny Toby?  Is it a matter of law, which compares
> withholding orange juice to stealing a car?  Or is it merely that daddy,
> as the parent, can make a unilateral decision (might)?

So, this is a kind of open ended thought that Toby might have about how he
feels.  It is a very complex cognition.  Little kids don't think about 
laws llke this.  If you believe in Kolhberg's stages of moral development 
(probably a pretty old-fashioned idea by now) then Toby is probably at 
what Kolhberg (K) woudl call the "me first" stage.  I'd say eventually we 
do want to get to that level of sophistication.  But right now we still
don't even know if something as simple as goal substitution will work
successfully in a knowledge based system.

So, you are tapping an interesting area but it is at a much more macro 
level of psychology.  I would rather focus on very simple appraisals
that are easy to generate with current knowledge based systems.

> Suggestions about how to make this more clear are welcome.

Now that I get it a little better I could suggest better wordings.
E.g. "Toby feels confused, he is unsure whether daddy is denying
OJ because he is bigger and has more power than Toby or is it
do to a rule that Toby and daddy must obey."

But still I think that generating something like that in a non-hack
authentic way is way beyond our abilities.

> > * as you prolly recall, in the dissertation, what you have as "story" was
> > split up into "overriding goal" and "situation."  it might be better to
> > keep that distinction.  i'm not sure.
>
> I haven't integrated all the levels of appraisal but this would be easy
> to do.  Each level (should) depends on earlier levels.  So by the time
> we get to level 'm' (the level of your dissertation), there is a
> empirically tested believable overriding goal from level 'g'.

I'm not sure what you men by empirically testing the overriding goal?
Are you testing that the goal is one that is believeable for kids to have?

> As soon as you report that you have toured all the levels then I will
> adjust the database so that it grows by level instead of jamming
> everything together, like it is now.

Well, I have sort of toured all the levels.  I feel pretty confused when I 
do.  I will keep at it.  But wanted to respond at the risk of doing so 
prematurely.

> > * regarding "tortured prose". e.g. this...
> >
> > In tobys opinion, his dad wants to avoid something about orange juice for
> > himself. (TODO: how to disambiguate whom the trailing himself/herself
> > refers without tortured prose?)
> >
> > ...those seem like issues that are very tightly intertwined with the tasks
> > of re-building the model, aligning the model with CLib (e.g. in my model
> > I used the term gender but in CLib they use sex).  I would not work on
> > any detailed KM model stuff until we decide how tightly we want to align
> > with CLib.  (I'm assuming you know that KM has a text gen facility and
> > are using that to generate the text and not your own hand-rolled text
> > gen system).
>
> Incidentally, my database uses sex instead of gender.  However, that's
> beside the point.

Okay.  ANd Peter has said that we do not need to be fuly aligned with 
CLib.

> What is worth stating is that there is a pretty clear separation between
> the model and the database.  I can provide a list of queries I run on
> the KM model.  The database only stores the cues, possible appraisals,
> and, optionally, the expected believability.

Ah. Okay.  Eventually we might wanto have a live model.  Whatever.

> What I like about the new database design is that the appraisals are
> stored in a tree-like structure.  Statistics are kept in real-time.

Cool!

> Branches which are not believable are not searched.

But getting ratings of things that we have hypothesized our unbelieable
is essential to verifying our hypothesis.  If our theory predicts Blah
is unbelievable, then we need to test it.  Maybe it would help to
have re-skimming of my dissertation of the table of contents?  I should do 
this too.

Bill

>