On 25.11.2010 13:16, Simon Grant wrote:

>     3. A combination (set) of values. This could be different from a
>     range, for instance if {2000-12,2000} is the set of the month
>     December 2000 and the year 2000, which is not the same as the year 2000.
> Maybe that's OK, but I don't yet understand the practical application of
> a set of date/time values. Perhaps it would be clearer (to me at least)
> to have some concrete examples?

For instance if a book has been published several times, it's 
publication date is a set of dates.

>     4. An approximate value. Approximation is a flag that marks a value
>     independent from other properties. The concept is based on fuzzy set
>     theory. For instance the approximate year 2000~ is not equal to any
>     strict interval.
>     5. a value that is questionable
> This is really what I am questioning. Do you (does anyone) think that
> fuzzy set theory is really something we want to have as one of the
> foundations of the spec? I certainly have never yet got the sense that
> fuzzy set theory was uncontentious, and I question the wisdom of basing
> any standard on a contentious foundation. But maybe this is for debate.

Fuzzy sets are not magic and they have proved to describe real-world 
applications (which are based on natural language, which is fuzzy perse) 
much better than strict sets. If you know exact values, fine. But in 
many cases you only know an approximate value.

> Leaving aside fuzzy set theory, it isn't hard in many cases to translate
> an approximate value into an expression that it is one value from within
> a range.

The whole point of approximate values is *not* to specify a strict set. 
That would be like putting squares in a whole.

> Can anyone propose any generally acceptable distinction between
> an approximate value with no definite range, and a value that is
> questionable? At present I cannot imagine any pragmatic difference.

An approximate value has no definite range or strict value. You only 
know that any exact value will be near the given center (without knowing
a specific probability distribution).

"around May" is most likely May, but can also be a few month earlier or 
later. But it definitely is a month.

A questionable value has a strict value, but this value may be wrong.

"questionable May" could be May, or any other value, a year, a specific 
day, an interval, or no month at all.

> So, for a single value date/time, my personal conclusion is that there
> are just four cases:
> 1. A definite particular value
> 2. One value drawn from a specified range (copes with cases like "one
> Christmas in the 1890's")
> 3. One point in a specified interval

But an interval is not the same as a particular point in this interval.

> 4. A "my/our/best guess" value.
> Are all intervals adequately represented by exactly one start and end
> value, from the cases above?
> For ranges, I'm aware that others have spent a lot of time thinking
> about this. But isn't a range just a discrete set formulated in a
> particular way? I'll try to look again at what there is in the spec draft.

I thought so, but it is not. ISO 8601 defines an interval as one of:

- a start-point and a end-point
- a start-point and a duration
- a duration and and end-point
- a duration

Neither of those is equal to a single point in time. You must 
distinguish for instance

- the year 1976 as point in time
- the year 1976 as interval

As far as I read ISO 8601 right, this distinction is already there.

You can say the the whole year 1976 lies in 1976 as interval, but they 
are not the same.


Jakob Voß <[log in to unmask]>, skype: nichtich
Verbundzentrale des GBV (VZG) / Common Library Network
Platz der Goettinger Sieben 1, 37073 Göttingen, Germany
+49 (0)551 39-10242,