RE: Model and sources of uncertainty

Paulo,
 
I don't think we in any serious semantic disagreement. All I see is that you
would prefer somewhat different tags. Please see my replies below.
 
==Mitch
 


  _____  

From: public-xg-urw3-request@w3.org [mailto:public-xg-urw3-request@w3.org]
On Behalf Of Paulo CG Costa
Sent: Tuesday, June 19, 2007 10:49 PM
To: mkokar@vistology.com
Cc: public-xg-urw3@w3.org
Subject: Re: Model and sources of uncertainty


Hello Mitch, Vipul, and All, 


So first of all I'd like to say that all I tried to do is read a couple of
papers (two of them are listed on the web site) that discuss various types
of uncertainty and then make an attempt at reconciling multiple views of the
terms used in the various approaches into a single ontology. In other words,
the notions I put in the ontology are not my inventions - they are defined
in various papers.


So did I. All the stuff about evidence came from David Schum's book (already
mentioned by Kathy) and numerous papers. He is a well known expert in
evidential reasoning.



During this weekend I tried to understand the feedback I received from the
group and then tried to see whether the current ontology is sufficient or it
needs to be extended. I came to the conlcusion that we might need to extend
it a bit. I have not made the changes in the graphics, as yet, since I
didn't want to put too much effort into the changes before first coming to
some agreements. 

However, in order to make the discussion more focused, I added some textual
descriptions of the terms used in the ontology (on the Wiki).

So here are my replies to the issues raised by the group, mainly by Paulo
and Vipul. I believe gstoil is in agreement with me.

1. Unreliability of the source: In the current Uncertainty Reasoning
Ontology, let's call it URO for now, this would be modeled by representing
the source as an instance of World. Thus there would be a sentence about the
source A "Source A is 33% reliable." Then the notion of "33% reliable" would
have to be specified using one of the Uncertainty Types.


My main issue with the graph is that I see uncertainty as a characteristic
of a given phenomena, not as a phenomena by itself.  
 
MMK: And the ontology captures this. Uncertainty is attached to a sentence
throught the property "hasUncertainty". The uncertainty is about a world
that the sentece is about ("saidAbout"). The sentence is "saidBy" an agent.
 
 As an example, a deterministic phenomenon can be perceived as uncertain due
to lack of knowledge about it, to a biased or limited perception, or to any
other cause that prevents our complete understating about it. On the other
hand, we might have complete information about a phenomenon and still remain
unable to tell its outcome, given the intrinsically stochastic nature of
that phenomenon. I called the first case as an epistemic "kind of
uncertainty", and the second case as an existential "kind of uncertainty".
Thus, when defining types of uncertainty, I'd list the above two.  
 
MMK: I modified the ontology a bit. I changed the name from
"UncertaintySource" to "UncertaintyNature". This is connected to Uncertainty
through the property "nature". I guess I could call it "kind", instead, but
I was afraid to use this name since this might get confused with "class" and
"type", both of which are special words in OWL.
 
MMK: I used the terms "epistemic", same as yours, and "aleatory" (instead of
existential), which I have seen in most publications that distinguish
phenomena that are random by nature, vs. those that seem to be
random/uncertain only because someone lacks knowledge about them.
 
I see all the terms that you list here (from 1 to 5) as characteristics of
the information we have about a given phenomena that might lead us to label
it uncertain. Thus, these are in fact sources of uncertainty, not types of
uncertainty. 
 
MMK:  I'm not sure I understand this. If all you are suggesting is to link
Uncertainty class with those five Characteristics by a property called
"sourceOfUncertainty", then this is just an issue of labelling. But perhaps
you are suggesting something more than this? Do you speak OWL? It would
really simplify the discussion if we could use a formal language instead of
only English. After all, OWL is a language supported by the W3C, and our
effort is part of W3C efforts.




2. Dissonance: I added one more type to UncertaintyTypes - Inconsistency. I
believe this would capture dissonance, but if not, we could then think a bit
more about this issue. I mean here logical inconsistency, i.e., when there
is no model for a sentence. I don't mean existence of evidence for and
against a specific hypotheses (which is just fine within the probability
theory).


Dissonance captures both contradictory and conflicting evidence.
Inconsistency happens when a KB has two axioms that imply mutually exclusive
hypothesis. 
 
MMK: This is exactly what I also mean by "inconsistency". I don't have a
special term for "dissonance" as yet; this could be added. But I would like
to settle the top level classes and properties, first. 


3. Incompleteness: I would also say that incompleteness is not a type of
vagueness. But it might be interpreted as a kind of ambiguity due to the
lack of sufficient information for resolving the question of whether a
specific world is a model of the sentence or not. But then we could also add
another class (Incompleteness) to the types of uncertainty.

4. Inconclusiveness: I believe I understand what it is, but I don't quite
see how this is a type of uncertainty. Perhaps I am missing something here?
But it looks to me more like redundancy, i.e., a sentence does not add to
the existing knowledge since it's already in what can be inferred from the
knowledge we already have. Please correct me if I am wrong on this.

5. Interpretation: I believe this is about sensors through which we perceive
the world. Similary as in point 1 above, this is a sentence about an
instance of World (sensor) whose accuracy would have to be specified in
terms of this ontology using one of the probability types.


As I implied above, I don't see any of those as types of uncertainty, but as
characteristics of evidence that result in uncertainty about an event. We
sense the world with sensors, the output of the sensors can be seen as
evidence we have on how the world actually is. If we have imperfect
information from our sensors we will have uncertainty about the phenomenon
they're sensing. 
 
MMK: I just uploaded the OWL representation of the ontology to the
Uncertainty Ontology page.
 
MMK: As I said above, I am not sure how to interpret your suggestion. My
thinking is that we need a class Uncertainty so that we can have uncertainty
objects. In case we need to send some information about uncertainty of a
specific statement, we can just send such an object. Such objects can have
various properties, like for instance the modeling approach used to capture
the uncertainty (e.g., probabilistic or fuzzy). I don't quite know how I
would do this if I attached only some features of uncertainty (what you call
"sources")  to sentences (if this is what you are suggesting). Other than
this issue, I think we are in pretty good agreement, except perhaps for some
labeling issues.
 
==Mitch
 

Cheers,
Paulo