Print

Print


Ok, I've changed the subject line again, and we'll refer to "reliability"
rather than "uncertainly".

 

I am still not clear on whether you want extensibility, or are satisfied
with a closed, controlled vocabulary.

 

You said:   

'I can well envision profiles where people have finer semantics such as
"most certainty correct (but possibly wrong)".'

 

I don't know how we can support a model that allows people to make these up
without building URIs into the model, unless we establish a registry of
reliability terminology, which I'm fairly sure we don't want, do we?

 

You  said:  

"How about the simple cases I've mentioned with a additional numberical
confidence level."

 

So:

'a'   = Known to be correct (observed, documented etc.) 
 ' b'  =  Likely correct ( p> 50%) 
 ' c'  = Possibly correct (Might be but not likely) 
 'd'  = Likely incorrect (The date is expected to be wrong p ~ 0) 
' e'   = Unknown (certainty unknown).

'fxxx' = confidence level xxx where xxx is a three digit integer 0-100
representing  percentage reliability.

 

In which case you could drop  " Known to be correct" since it would be
accomodated with a value of 100, or retain "Known to be correct" and instead
restrict the value to two digits.

 

But I don't see how this approach squares with the need for extensibility
(if we really need it).

 

 

On the "volatile dynamic dates" topic. I' ll try to talk about that in a
separate thread.

 

--Ray

 

From: Discussion of the Developing Date/Time Standards
[mailto:[log in to unmask]] On Behalf Of Edward C. Zimmermann
Sent: Monday, February 07, 2011 11:09 AM
To: [log in to unmask]
Subject: Re: [DATETIME] uncertainty and precision

 

On Mon, 7 Feb 2011 09:25:13 -0500, Ray Denenberg, Library of Congress wrote 
> Ed, 


> 

  


> You raise two issues:  


> 

  


> (1) The default should be "unstated reliability" rather than "certainly
unknown" as I had suggested.    That's fine with me.  So '1984?'  would
carry the implicit semantics "uncertain, but level of uncertainly unstated".



>  

  

I think we should try (I am especially guilty) to concentrate on not
speaking about certainty or uncertainty but reliability---- perhaps even
trust. 


> (2) Extensibility or not.  There are two ways to do this, I think. One is
to declare this model to be explicitly non-extensible and define a closed
controlled vocabulary.  The other would be a model based on assigning URIs
for the levels. 

 

I think that is too much of a good thing. How about the simple cases I've
mentioned with a additional numberical confidence level. This would, I
think, be sufficent to cover the needs of emprical artefact dating and event
forecasting.

>     The first is what I thought you wanted, and since you are perhaps the
only member of this forum interested in uncertainly - that is, going beyond
simply asserting that a value is uncertain, 

I think there were a few people here. 

> 

While we are on the subject of "needs"... is there any way we might be able
to also include some manner to talk about volatile/dynamic dates----
carrying over a topic we discussed a few months ago in the context of SRU/W
facets--- such as today, now, yesterday, past two weeks, last month, next
year etc...   

  


> 

  


> 

  


> 
> From: Discussion of the Developing Date/Time Standards
[mailto:[log in to unmask]] On Behalf Of Edward C. Zimmermann 
> Sent: Friday, February 04, 2011 2:27 AM 
> To: [log in to unmask] 
> Subject: Re: [DATETIME] uncertainty and precision 


> 

  


> On Thu, 3 Feb 2011 11:24:30 -0500, Ray Denenberg, Library of Congress
wrote 
> > (I changed subject of this thread, it was "uncertainty".) 


> I would change the topic to "data reliability". Uncertainty is an
attribute. Precision is also something else.. 


> > Take "uncertainty" first.    If we say that ? may be followed by a
measure of uncertainty we have to be careful because ? does not always end
the entire string.  For example: (2004)?06-1.  However if we say that
uncertainty is indicated by one of the letters a, b, c, d, e, then we don't
introduce ambiguity because in the current draft there isn't any case where
? would be followed by one of these letters. 


> 
> > 


>   


> 
> > So the question is:  Are Ed's suggested set of values acceptable to
indicate uncertainty: 


> 
> > - a) Known to be correct (observed, documented etc.) 
> > - b) Likely correct ( p> 50%) 
> > - c) Possibly correct (Might be but not likely) 
> > - d) Likely incorrect (The date is expected to be wrong p ~ 0) 
> > - e) Unknown (certainty unknown). 


> 
> > Actually I would eliminate (e); "certainty unknown" should be the
default - it should not be mandatory for the ? to be followed  by an
uncertainly level.  


> 
> >  


>   


> 1) I think the default is "unstated reliability". For the reciever there
is perhaps little difference but they are not the same just as the (a) case
is equivalent to the reciever but not the "same" as the non-specified case,
e.g. (X)?a versus (X) 


> 2) I am not quite sure that using letters is the best approach since it
locks things too much down  and prevents extension... I can well envision
profiles where people have finer semantics such as "most certainty correct
(but possibly wrong)". Recall these are also dependent upon states of
knowledge. The assertion, for example, that that the Earth was the center of
celestial spheres  was "known" to be correct and confirmed by observation
(including those of Ptolemy and Copernicus). Tycho Brahe established a new
standard of observations and that formed the basis for Keppler's work (and a
paradigm change).... 


> 
> > Next, precision.  Similarly, ~ could be followed by a level of
precision.   Simon, would you like to propose a scheme? 


> 

  


> Here we should to talk instead about "readability". 


> > 


>   


> 
> > --Ray 


> 
> 
> 




-- 

Edward C. Zimmermann, NONMONOTONIC LAB 
http://www.nonmonotonic.net <http://www.nonmonotonic.net/>  
Umsatz-St-ID: DE130492967