[Aleader-dev] Pronouns

[Joshua N Pritikin]

[I'm going out of town for 4-5 days.  I thought I better send
this because I think you will get more optimistic after reading
the abstract.  Have fun with your neuropsych assessments!]

Abstract

Pronouns correspond with Baron-Cohen's TOM.  Combining pronouns and
desirability can result in an incremental extension to your thesis.
A KR model of pronoun-desirability is presented.  A selection of
examples scenarios from OCC are used to illustrate the utility of
pronoun analysis.  Potential empirical studies are outlined.

Discussion

There was an essential inspiration I received (a few years ago) in the
construction of Aleader.  In academic terms, I would describe it as
organizing affect analysis around a TOM and the proposition that
pronouns (I, we, you, he, she, they) are _the_ basic TOM concepts.

There is a natural correspondance:

2nd person pronouns : Intentionality Detector, Eye Detection Detector
and Shared Attention Mechanism.

3rd person pronouns : Theory of Mind Mechanism

Aleader theory is built from evidence found in language as is OCC.
However, the emphasis is on pronouns. All other linguistic evidence is
integrated around a pronoun architecture.

In your thesis, you are concerned with computing desirability.  I want
to combine pronoun appraisal with your goal appraisal.  This email
textually describes how this combination might work. Hopefully we can
meld the material into an empirical test.  I am hopeful that adding
pronoun appraisal is a small enough incremental step that we can build
on much of the groundwork laid by your thesis.

I included the plural pronoun "we" in my list (above).  Actually I
want to see how far we can get with only two people plus an witness.
This means three people, or two people if one of the participants also
takes on the role of the witness.  This is the minimum structure
needed to introduce 1st, 2nd, and 3rd person pronouns.  By limiting
ourselves to the minimum structure we carve out a smaller piece --
hopefully a tractable piece -- of the affect puzzle.

Let us take Figure 3.1: Example Clinical Workbook Item from your
thesis and re-write it in pronoun-desirability form.  For the sake of
brevity, I will choose an arbitrary interpretation instead of trying
to generate all possible interpretations.  We will add back
generativity later.

sentence #1: Tracy wants a banana.

pronoun-desirability form:
  I (1st person) = Tracy
    [will be] happy = want a banana 

sentence #2: Mummy gives Tracy an apple for lunch.

pronoun-desirability form:
  I (1st person) = Mummy
    indifferent = (Mummy doesn't want anything for herself.)
  you (2nd person) = Tracy
    happy = an apple for lunch
  overall (3rd person) desirability state = gives

The "appraising-agents" relation from your thesis is equivalent to the
1st person pronoun by convention.  Therefore, I have chosen Mummy as
the appraising-agent.  The rest is just a matter of assembling our
evaluation into English sentences.

Q. How does Tracy feel about receiving an apple for lunch?
A. Mummy thinks that Tracy feels happy about it. (Mummy's point of
view + goal substitution)

A few things to note:

+ Sentence #1 is too simple to involve even a 2nd person pronoun.
This suggests that sentence #1 does not involve TOM (or even shared
attention).  So from a TOM point of view, sentence #1 is merely a
logical assertion in the goal appraisal module.  On the other hand,
sentence #2 is exactly complex enough to provide bindings for 1st,
2nd, and 3rd person pronouns.  If sentence #2 is affectual then the
affect is a TOM affect.

+ The word "desirability" doesn't read very well for the 3rd person.
That's why I invented a new terminology "situational intention" to
describe 3rd person desirability states.  Each situational intention
state corresponds to a pair of individual desirability states.
However, there are also English words which describe situational
intention states.  Some examples include: give, take, admire, admired,
etc.

Going through the exercise of building Aleader & etc, what I found is
that given any two known desirabilities (1st, 2nd or 3rd) then you can
solve for the remaining unknown desirability.  So once we pick one of
the many interpretations of sentence #2 then we can retract knowledge
of one of the desirabilities and predict it using logic rules.  The
full PowerLoom (KIF) details are available at:

  http://savannah.nongnu.org/download/aleader/htdocs/aleader.ploom

with some example situations at:

  http://savannah.nongnu.org/download/aleader/htdocs/nausicaa.ploom

To get a better sense of how frequently we find simple TOM situations
of two people plus a witness, I have undertaken a pronoun analysis of
OCC emotions.  The analysis doesn't interfere or clash much with OCC.
It is more a matter of being explicit about something which was
formerly implicit.  Details follow.

  Well-being emotions (p 86)

  Joy's type spec: (pleased about) a desirable event

  Joy's example: The man was pleased when he realized he was to
  get a small inheritance from an unknown distant relative.

  Distress' example: The driver was upset about running out
  of gas on the freeway.

Joy's type spec offers bindings only for a 1st person.  Distress'
example involves only a 1st person.  This suggests that a TOM oriented
model should consider Well-being as part of the desirability appraisal
and not as an affect.  Although Joy's example offers bindings for a
1st person (the man), 2nd person (an unknown distant relative), and
3rd person (give), this TOM structure is certainly an artifact of
English composition rather than an essential feature of the joy
example.  So our rule-base would contain:

;; I don't really know how to write this in KM, but here goes!

(every Human has
       (goalAgent-of ((a Goal with 
                         (goalEventType (Inheritance))
                         (goalAgentRole (receiver))
                         (goalObjectTypeRole (transfered))
                         (goalObjectType (money))))))
(every Human has
       (goalAgent-of ((a Goal with 
                         (goalEventType (NotRunOutOfGas))
                         (goalAgentRole (driver))
                         (goalObjectTypeRole (fuel))
                         (goalObjectType (gas))))))

Now if we go back and do a pronoun-desirability appraisal of joy's
example then we can use our goals to determine the desirability.

pronoun-desirability form:
  I (1st person) = The man
    happy = because getting an inheritance (goal congruent)
  you (2nd person) = an unknown distant relative
    indifferent = assume dead
  they (3rd person) situational intention = take

As you can see, this is the same thing as before but with the TOM
structure exposed.  Now we move on to Fortunes-of-others (p 92).
These OCC emotions typically involve two instances of a TOM.  Since we
already discussed Well-being emotions, we will only consider the
immediate TOM structure.

Happy-for example: Fred was happy for his friend Mary because she won
a thousand dollars.

pronoun-desirability form:
  I (1st person) = Fred
    happy = aligned with Mary
  you (2nd person) = Mary
    happy = goal congruent
  they (3rd person) situational intention = ready

  Sorry-for example: Fred was sorry for his friend Mary because her
  husband was killed in a car crash.

pronoun-desirability form:
  I (1st person) = Fred
    sad = aligned with Mary
  you (2nd person) = Mary
    sad = goal incongruent
  they (3rd person) situational intention = protest

Resentment example: The executive resented the large pay raise awarded
to a colleague whom he considered incompetent.

Here is a naive analysis, but it is wrong:

pronoun-desirability form:
  I (1st person) = the executive
    sad = didn't get pay raise
  you (2nd person) = a colleague
    happy = got large pay raise

This is wrong because the appraising-agent is not consistent.  For the
executive, desirability is from the executive's point-of-view.  For
the colleague, desirability is from the colleague's point-of-view.  If
we need to represent both points of view then we need to create two
separate TOM situations.  Since this email is getting lengthly, let us
look only at the executive's point-of-view.

pronoun-desirability form:
  I (1st person) = the executive
    sad = didn't get pay raise
  you (2nd person) = a colleague
    indifferent = got large pay raise (past tense)
  they (3rd person) situational intention = uneasy

Or if the executive considers it like a theft then we should write it
like an OCC Prospect emotion (below).  Notice that resentment involves
an event status change from unconfirmed to dis/confirmed.

pronoun-desirability form:
  I (1st person) = the executive
    happy = due for raise (unconfirmed)
  you (2nd person) = a colleague
    sad = if gets pay raise (unconfirmed)
  they (3rd person) situational intention = steals

Or after the dis/confirmation:

pronoun-desirability form:
  I (1st person) = the executive
    indifferent = no raise (confirmed)
  you (2nd person) = a colleague
    sad = got pay raise (confirmed)
  they (3rd person) situational intention = observes

This example appears especially generative because we easily found
three interpretations with slightly different shades of meaning.

OCC Appraisal of Object emotions (p. 156) operate on the basis of
appealingness.  Your thesis did not consider appealingness but for the
sake of discussion we can assume an equivalence: it desirable to be
appealing and undesirable to be unappealing.

Liking example: Mary was filled with affection as she gazed at her
newborn infant.

  I (1st person) = Mary
    happy = her infant is appealing
  you (2nd person) = newborn infant
    indifferent = busy breathing and looking around
  they (3rd person) situational intention = admires

Disliking example: John disliked the concert so much that he left in
the middle.

  I (1st person) = John
    sad = her infant is appealing
  you (2nd person) = concert
    indifferent = are concerts intentional?
  they (3rd person) situational intention = uneasy

What is surprising about the Disliking example is that a concert is
bound to the same slot as the infant.  The output of the
intentionality detector seems to be ignored (is a concert a living
thing?).  Given what we know about autism, it seems unlikely that we
can assign the same affect regardless of the intentionality detector.
In OCC, it is clearly stated that Appraisal of Object emotions also
apply to taste (as in food).  One solution is to refine the Appraisal
of Object emotions by whether the object is judged intentional by the
intentionality detector.

Conclusion

Autism research leads us to believe that TOM is a real
psychological-brain component. Therefore, its operation should be
accounted for clearly in cognitive appraisal theory.  Looking at the
role of pronouns in OCC emotions highlights some mild ambiguity in
OCC.  Pronoun analysis guides us to a more precise model of affect
eliciting situations.

Looking towards an empirical study, I think the dependent variable can
be believability of appraisals (like your thesis).  Since we are doing
appraisals (not affects), it should work OK to work in textual form
(without video).  I'm not sure about the independent variables.  We
can vary the point-of-view (1st person / appraising-agent) or ask
about the situational intention.

What do you think?

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