RE: [huml-comment] PC-33 -Section 4.4.6-race

["Bullard, Claude L (Len)" <clbullar@ingr.com>]

I'll try this but only informally.  Much would depend 
on the particular application created using the HumanML 
primary types.  This information takes time to create, 
should be inspected by members of the group so described, 
and must be layered for detail.

>That's an useful case.  Thank you.  I can relate to that.  Here's how it
>looks to me:

>1. I think what I would really do is be confused and unsure.  I don't know
>that I would have the sense to ask the right questions.  I would also be
>concerned about the communication being ineffective.  And it is a reliable
>prediction that I'd feel awkward and the experience would be unsatisfying.

Good.  The first problem of the clerk is to be self-aware.  They must 
realize that they are confused and recognize as soon as possible, what 
signs in the communication are confusing them.  Otherwise, the reaction 
is typically a silent "Oh what is this idiot on about?" rather than, 
"Something here is confusing" and creating a strategy to resolve the 
confusion.  Miscommunication gets farcical or even homocidal if it 
continues unabated.  So sayeth Shakespeare and the Marx brothers.

>2. Seeing this set forth as a little puzzle, the only thing I can see to do
>in the hypothetical situations is (in my job as the clerk) to ask questions.

Good.  Direct questions are often the best way to resolve confusion.  The 
problem is selecting the right questions to ask.

>2.1 Asking where they are from would not help, because I don't have
>sufficient world knowledge to have that help me.

Ok.  More later on that topic, because that is one thing a HumanML application 
could provide:  a context-indexed sign dictionary.

>2.2 I might do that just to have more communication and ask some question
>about their customs, especially gestures that mean yes and no.

Better.  One could use a prototype.  Lean forward and say to the person, 
"I am a bit confused.  To this question, you answered, 'Yes'" and nod 
up and down as you do.  "Do you want the apartment?" continuing to nod 
up and down and reinforcing the gesture with YOUR question also slowing 
down a little so the rhythm of the nod and the question are in sync, thus 
emphasizing your intention.   

At least two problems can occur here.  When leaning 
forward, you may violate their interpersonal distance (the proxemic 
category) and by slowing down, you may create a sign of 'talking down'.
Interpersonal distance and speed of speech vary by culture and even 
locale within a culture.

So unfortunately, while the questions are good, there are still some 
risks here.

>3. So, since I am new on the job and I don't know about the different
>cultures of the people who will be looking at my apartments, it is all
>pretty chancy.

Yes, and it always is.  On any given day, with any given individual or 
group, it is easy to put the foot in the mouth.  

>4. My next question: Where is Human ML in this?  How would I, the clerk be
>aware of it, and if unaware of it, how would its existence bear on this
>scenario?

It will depend on the application.  This gets a bit pricey for this application, 
but a phrase and gesture dictionary with cultural indices can be helpful. 
Suppose you had an interface that includes the ability to quickly enter 
phrases of speech, plus pick from an iconized human, the part of the body 
gesturing and the direction of gesture.   You enter "yes" but also enter 
the nod.   The system attempts to match that to known cases of such 
a combination of gestures and returns:

"Origin country:  India.  65% confidence.
Interpretation:  affirmation.  Yes means Yes.
Would you like clarification? Y/N"

You select "Y" and it returns:

"Observations of natives of India indicate a 
habit of head bobbing when speaking.  Some 
westerners confuse this with gestures for 
yes and no.  The vocal sign is more reliable.

Would you like to enter additional context 
data? Y/N"

You enter N.  At this point, your ear for 
the speaker's accent makes you think the 
guess is good. (Another application would 
record responses and use Fourrier data to 
analyze and identify an accent, but that 
is expensive stuff for hotel clerks.)

The application continues:

Would you like more information on Indian 
contexts?  Y/N"

You ask the person at the counter, in a friendly 
casual way, "Are you from India?"  They say, "Yes!"

You enter "y" and the computer returns a 
short description for typical Indian HumanML 
categories such as haptic values for interpersonal 
distances ranked by social relationship types. 
You see from that that it won't be wise to lean 
into the face of the person at the counter.

Because it knows from your profile that you are 
western, it reminds you that it is not uncommon 
for some Indian males to hold hands in public 
in their own country and that this is a typical sign 
of friendship, not homosexuality.  It can also 
bring up local restaurants based on religious 
dietary restraints, some local movie theatres 
that specialize in Bollywood movies, where 
local temples and mosques are, and so on.  You ask, "would 
you like some information on local Indian activities?" 
The customer says, "Yes!" and you print out 
the second set of information for the customer 
as a gesture of friendship and help.

IOW, and this all depends on good information, it 
acts like an advisor.  HumanML categorized databases 
for different cultures can be created using XML by 
members of that culture, put up on the web for 
servers anywhere to access, and in fact, the 
application you have been using is a vanilla 
HTML web form.  Nothing exotic or quixotic.

len