[Aleader-dev] pre-conceptual definition

[Joshua N Pritikin]

On Mon, Sep 08, 2003 at 12:10:10AM -0500, William L. Jarrold wrote:
> > ... I'd like to
> > send one impractical email which I've been working on for more
> > than a week.  Perhaps you can review it first before finishing
> > the backlog?  What do you think?
> 
> Sure.  Send it on.

Again, please give this email higher priority than the other
emails in our queue.  I believe this email will change our
focus in a positive way.  Anyway, here goes.

> > "emotion".  I don't think emotion is very well defined (in general)
> > and "affective state" is a mouth-full.  What do you think?
>
> It depends on how the field has defined emotion.  Do what the field
> does before arbitrarily redefining emotions.  OCC defines emotion
> one way.  How do others define it?  Reading happens fast, so long
> term names are okay.

OK, I'll follow the literature.  Here is a _draft_ quality glossary:

emotion - The words which people typical consider as referring to
emotions: hope, fear, joy, relief, sadness, distress, frustration,
etc.  We generally agree with the OCC definition of emotion.

affective state - We consider classical emotions a subset of a more
general category of affective states.  Beside emotion, affective
states include moods such as curiosity, abandoned, criticizing or
invitation.

mindreading - To infer another person's affective state by running our
system ``off-line'': to place ourself in other person's shoes and see
how we resonate in the pretended context.  Perhaps the most obvious
example of mindreading is what happens while watching a film.  A film
is nothing but 3rd person perspective: image and sound.  However,
people can easily mindread the actors and feel a precise replica of
the affective states depicted onscreen. [also mention baron-cohen?]

empathy - Some vague mixture of mindreading, sympathy and emotional
bonding.  We avoid this word since we feel that it is not v well
defined.

appraisal - The cognitions which precede and trigger cognitive
affective states.  These cognitions are originally unconscious, but we
posit that we can gain awareness of them.

We distinguish affective states by cause.

+ Biological factors such as hormones, hunger or thirst

+ Cognition, meaning cognitive appraisals

+ TODO: expand list with help from the lexicon article

+ Mindreading one of the other causes

> It should not be hard to studies that have looked at the reliability of
> judging emotion in pictures of people's facial expression.

When I get time, I am going through _Cognition and Emotion_ just
reading stuff somewhat at random to absorb the style & etc.  It seems
like re-test reliability is a pretty standard statistical measure.

> > On the other hand, please recognize that if we stick with
> > Aleader's simple, narrow view of emotion then we still
> > get useful classifications ("useful" meaning that the
> > classifications subjectively feel correct) without
> > resorting to generativity.
>
> [Will's long and interesting response omitted.]

I think our discussion with respect to generativity boils down to one
metaphor: If you have a hammer then every problem looks like a nail.

+ I acknowledge that generativity makes a non-generative cognitive
model deeper and more believable.

+ You acknowledge that a non-generative cognitive model may be
important in the context of training, as long as it doesn't make us
"more rigid, inflexible or unidimension."

It seems like we are mostly in agreement.  I believe that your
discussion of generativity (with minor revisions) will fit gracefully
into our theoretical framework.  Even if there remain minor points of
contention, I want to focus other issues which I feel are more urgent.

> > > Here's another one...You really need to have a controlled list of terms.
> > > How many terms are emotion terms?  50, 5, 100, 500, 1000?
> >
> > Hrm .. I thought I explained in the prototype research paper
> > that emotion _terms_ are secondary to the collection of
> > examples of a given emotion.
>
> I disagree.  To be sure, note: a term is not a word.
> [... snip ...]

I read your whole reply. I think you are missing my point.  Allow me
to offer an analogy.

What does the phrase "the taste of suger" mean?  You know because you
have eaten suger.  Common suger is a particular family of chemical
compounds of a particular purity.  When we eat suger then we _know_
how suger tastes.  We experience the taste in a _pre-conceptual_ way.

In my opinion, this distinction is incredibly important.
Pre-conceptual experiencing is how we actually know reality.  A symbol
which is not grounded with a pre-conceptual experience may not have
any precise definition.  Even a Cyc KB with 100 million assertions
can't offer a more precise definition of "the taste of suger" than I
get when eating suger.

Everything you mention in your reply: term, word, 59092928098092,
snicklefritz, "Fears Confirmed", #$FearsConfirmed, label, name,
pointer, category, prototypes, all these things are conceptual.  The
difference between these concepts and film examples is the difference
between the phrase "the taste of suger" and eating suger.

In other words, when I define an affective state, I want to provide a
reliable (repeatable) recipe which causes you to _experience_
(mindread) exactly that state.  Why?  Because pre-conceptual
experience is always more vivid than conceptual description.  That's
why I am very picky about insisting that I have not identified any
affective state _terms_.

I feel like this point is absolutely crucial to a new theory of
emotion.  If I am using funny terminology then introduce me to the
standard terms.  Especially _this_ discussion needs to be crystal
clear to readers.

Perhaps I need to discuss more about "pre-conceptual definition".
Imagine you are in a foreign country where you don't speak the
language and they don't speak your language.  You may not have any
shared words, but you do have shared experiences.  So if we have an
experience of eating suger and my friend says "saker", then I
remember that "saker" has something to do with eating suger.  I guess
the same thing happens when we first learn language as children.

One of the reasons I wanted to come to India is just to remember how
it feels not to be completely surrounded by known words.  While in
USA, there is the constant bombardment of English.  Parsing and
understanding English becomes a habit which we seem unable to turn
off.  So I wanted to stay in India just to experience the world with a
minimum amount of conceptual thinking.  I have even tried to avoid
learning Marathi (the local language) so that I would have more
opportunities to experience pre-conceptually.

OK, language learning is not an exact parallel because the film clips
typically _do_ include a spoken conceptual component.  However, the
significance of the conceptual component is minimized when we _unify_
the examples together (like KM unification, but pre-conceptually).  I
mean, it is somewhat non-sensical to unify the conceptual components
specific to different films.  How can the ideas specific to Star Wars
make sense in the context of Good Will Hunting?  So after unification,
there remains _something_ which is _not_ a literal description which
causes the mindreader to experience a specific affective state -- this
is a pre-conceptual definition.

Understand?  If not, let's try again.

> Hmm, I thought you were interested in prosody.  If you aren't, then
> I think you'd be happy with representing textual stories and teaching a
> computer to infer the affect.

After the above discussion, can you guess how I would answer?

+ A cognitive affective mindreading model which is built upon
pre-conceptual definition has the potential to be more accurate than
a conceptual model.

+ Prosody (and facial expressions) are investigations which are
sub-operations of mindreading.  For the purpose of pre-conceptual
definition, we wish to send experiences through the "mindreading
module" hostically, without micro-analyzing the operation.

> My sense is that it will
> help publishability to define what is and what is not an emotion.  It
> would be useful to give examples....
>
> - what is clearly inside the category of "emotion"
> - what is just barely inside the category of "emotion"
> - what is just barely outside the category of "emotion"
> - what is just clearly outside the category of "emotion"

I keep fumbling around trying to answer this.  Now that I have read
the "affective lexicon" article, I sense how to press forward.  Thank
you for bringing this extremely important article to my attention!
Before I get started, I need to highlight the basic structural
difference between "affective lexicon" & Aleader.

OCC appraisal theory _starts_ with emotion terms, from which appraisal
rules are developed (correct?).  OCC discusses, for example, what are
the cognitive conditions preceeding the emotion X?  To contrast,
Aleader _starts_ with a "unified ancestor affective state" (_without_
any emotion targets). The appraisal rules sub-divide the affective
state space.  That's why I was describing the Aleader appraisal in
terms of a decision tree.  As we keep splitting the affective state
space then eventually we reach one of these leaf cases:

+ An affective state which seem to correspond to an OCC emotion.  For
example, your description of OCC admiration closely mirrors one of the
Aleader appraisals.  I am still waiting for my copy of OCC to identify
more instances of overlap.[*]

+ An affective state which corresponds to a word which is not an OCC
emotion.  I already offered some examples above which I repeat here:
curiosity, abandoned, criticizing or invitation.

+ An affective state which does not correspond with an OCC emotion and
also lacks a similar English word.  Consider the affective state: "I
am relaxed after you admire me."  Which succinct label should we
assign to this affect?  "Soothed"?  I'm not sure.

OK, I don't want to go into more detail at the moment.  However, one
thing I want to point out is that the whole "pre-conceptual
definition" discussion so far is _not_ Aleader specific.  Any affective
model may be able to benefit by a pre-conceptual style of definition.
In fact, I do _not_ care about the details of the Aleader model per se.
To me, what is important is using pre-conceptual definition.

Perhaps you said it best: "So far, the chief value of your system
seems to me to be able to provide a vivid and reliable depictor of
emotions."  Yes, absolutely!

[*] I think you suggested email'ing Gerald Clore about an electronic
version of OCC since I am not receiving quick service from the
bookstore.  I do not feel comfortable making this request.  I prefer
to wait.  If you really think that he will respond positively then you
are welcome to try.  (After one month, shipping status is:
"Unavailable in primary warehouse, being procured from secondary
warehouse" -- Ugh!)

> > > One important
> > > question is, is there more heterogenetiy between categories than within
> > > categories?  One way you could test this is ask people to view pairs of
> > > scenes and rate their affective similarity on a scale of 1-5.
> >
> > I have already done most of the work to automate this type of test.
> > I will get busy and make a few more preparatory changes to the software.
> > Perhaps within a week, we'll be ready (software-wise) to get started
> > testing human subjects.
>
> Cool.  Howz that gone?

[I might have mentioned some of this already.]

12 Aug: I met Dr. Shubhada Pange, one of the top psychologists in
Nashik.  It was pretty cool.  She doesn't know much about _cognitive_
psychology, but she does a lot of psychological testing in large
companies and seems interested in being helpful.  For example, I think
we can get 100s of English speaking subjects for testing once
everything is ready.  She asked for a bunch of background research
articles which I email'd today.  Let us see what happens.

22 Aug: Dr. Pange didn't reply to any emails.  I remember that she
said that she is not very comfortable with computers.  Typical
psychologist!  Oh well.  What I predict is that nothing will happen
until I go back to her office with the whole experiment 100% ready for
human subjects testing.

By the way, you keep mentioning that our experiemental strategy may
change.  That's why I have not put much effort into molding the
software to present any kind of specific test.  I have work on:

+ A preliminary KR model of the Aleader appraisal.  PowerLoom looks
like a good option (at least when the next version becomes available).
I can forward you an article about connecting PowerLoom & Cyc.

+ Brainstorming.  Maybe we can do an experiment only to test whether
pre-conceptual definition is more re-test reliable than conceptual
definition?  Or, who knows, maybe "pre-conceptual definition" is not
as innovative as I presume?

+ Waiting for stuff to happen.  The whole project seems stuck at
the moment.

Before writing any more, I await your reaction with respect to
"pre-conceptual definition".

-- 
A new cognitive theory of emotion, http://savannah.nongnu.org/projects/aleader