Emotions and the Semantic Web

[John Flynn]

I posted a related email to the Creating AI group, but this email is more emotion specific. I am attempting to create an ontology, using the Web Ontology Language OWL, that represents a variety of human charateristics, including emotions. My ultimate goal is to develop a web-based system using ontologies and software agents to simulate selective human behaviors. I believe a key component of such a system is the representation of emotions. I am not a cognitive psycologist but I suspect that a selected subset of emotions may be adequate to reflect at least a reasonable first-order influence on behavior. I also suspect that adding additional emotions to the core set over time would improve the accuracy of the simualtion, but would probably only reflect only marginal improvements in simualtion accuracy. Of course emotions are only meaningful in the context of a target. I love my wife. I hate my mother-in-law. I love my car. I hate my wife's car. These are examples of targets for emotions. So, to be meaningful in a computer simulation, it is necessary to capture not only specific emotions and their intensity levels, but also the context of the emotion in terms of targets. There is also a temporal aspect to emotions. I may love my wife in the morning but due to some stemuli I may hate her later in the same day. I have expanded some of these ideas, including a first cut at an ontology, on my website at http://web.tampabay.rr.com/flynn/COPE.h . I would be very interested in your comments.

 

Thanks,

 

John Flynn

[Brandon Franklin]

Hm.

On 1/11/06, John Flynn <jfly...@gmail.com> wrote:
> Of
> course emotions are only meaningful in the context of a target.

"Of course"? I do not believe that's a foregone conclusion. What is
the target of happiness? What is the target of sadness?

> I love my
> wife. I hate my mother-in-law. I love my car. I hate my wife's car. These
> are examples of targets for emotions.

I would argue that you are getting confused by some sloppy language
conventions that we use. When you say you "love your car", I do not
believe you're describing an "emotion" but rather an "opinion".

There's general agreement that the four basic emotions are happiness,
sadness, anger, and fear. I would argue that none of those has a
"target", though they may all have a "cause" which is not the same.

If you hate your mother-in-law, I would assert that what you are
saying is that she causes you to become angry.

>So, to be meaningful in a computer
> simulation, it is necessary to capture not only specific emotions and their
> intensity levels, but also the context of the emotion in terms of targets.

I do not agree with this at all, and neither does the science.

> There is also a temporal aspect to emotions. I may love my wife in the
> morning but due to some stemuli I may hate her later in the same day.

Put differently (and, I think, more accurately) in the morning she
made you happy, and later she made you angry.

-Brandon

--
My LinkedIn Profile: http://www.linkedin.com/profile?viewProfile=&key=1615193

[miaou]

I think the context analysis should be the first step. It must be done
acording to a choosen cultural environment, because semantic analysis
relies on culture, which gives a basis of of already known relations
between the elements of a context, and also a structured map of words
linked together: semantic fields (I'm using terms directly translated
from french, sorry). When a phrase has been analysed acording to the
cultural environment it belongs to (a first, simple analisys, based on
words ocurences would do it), then we get this map. We must then know
how deep is the contrast between these different fields, because this
will give us the intensity of each specific emotion. Again, this relies
on already known data. The use of databases seems mandatory to me. The
values we get can be used to draw vectors. After playing around with
those, to reduce their number, the resulting ones give directions and
intensities. The trick would be to be able to make the values point at
newer semantic fields in order to follow the human mind: What hapen
when your wife makes you angry ? Do you leave the house and never come
again ? Of course not, because she's your wife, because you probably
love her, because you might have children etc... So, not putting
yourself away from the cause of a negative emotion can create new
emotions, much more difficult to capture, I think. You might feel glad
not to leave because this reminds you a beautifull aspect of human
beeing, you might feal angry at yourself because you don't dare to
leave, you might a lot of things, but these things are sources of new
emotions that continue to mix with the previous ones, modifying the
global result. Synthetising human behaviors doesn't seem to be reliable
unless you have a LOT of computer ressources. Simple simulations can be
done but I wouldn't expect spectacular results.