Issue 439: Approximate Dimensions

Starting Date: 
2019-10-15
Working Group: 
3
Status: 
Open
Background: 

Posted by Richard Sanderson on 15/10/2019

Dear all,

In recent history, we have added P189 approximates for the practically ubiquitous scenario where we have recorded the approximate “declarative” place of an event, but not the exact “phenomenal” place. P189 allows us to say that the event took place at the phenomenal place, which is then approximated by the declarative place.

Thus: 

  Birth_of_Rob a E67_Birth ;

    p7_took_place_at [

        a E53_Place ;

        rdfs:label “The exact place Rob was born” ;

        p189i_approximated_by [

            a E53_Place ;

            rdfs:label “New Zealand” ;

            // …

        ]

    ]

This gives us two significant advantages:

    We can have multiple declarative places associated with the single phenomenal place. This allows us to be clear that the event took place in one location, but we have multiple ways to describe that location in our information system.
    If we can be precise (enough) about the phenomenal place (e.g. we have the GPS coordinates from the digital camera that took the photograph), then we do not have a different model … we can simply ascribe those coordinate values to the phenomenal place.

While the E53 Place scope notes do not talk about approximation, there is another class that does … the very next one, E54 Dimension.

An instance of E54 Dimension represents the true quantity, independent from its numerical

approximation, e.g. in inches or in cm.

However, there isn’t a property that allows us to use this same approximation pattern for Dimensions.

The same advantages would apply:

    We can have multiple declarative dimensions (10 inches, 25 centimeters) that approximate the true dimension, rather than implying there are two different dimensions.
    If we do not have this case, because the dimension is measured very accurately and has only a single numerical representation, then we can simply use a single Dimension.

This is also useful for conservation when the same dimension is measured to different degrees of accuracy with different instruments or techniques … there is only a single height (for example) but it is measured with a laser, or by estimation.

Thus I would like to propose the addition of a new property, Pxxx_approximates_dimension, that mirrors P189_approximates, that would be used to associate true dimensions with their approximations.

It would be used in exactly the same way as P189:

painting a Human-Made_Object ;

  has_dimension [

    a Dimension ;

    p2_has_type <aat:height> ;

    pxxxi_dimension_approximated_by [

        a Dimension ;

        p90_has_value 10 ;

        p91_has_unit <aat:inches>

    ]

  ]

Thank you for your consideration of this issue!  I’m happy to write up a draft scope note for discussion if the general issue is considered to be worthy of inclusion.

 

Posted by martin on 16/10/2019

Dear Robert, All,

Your proposal well taken, but the recent change in the scope note was exactly that "The properties of the class E54 Dimension allow for expressing the numerical approximation of the values of instances of E54 Dimension. ".

The point is, that true numerical values of Dimensions do not exist for continuous value spaces. Therefore, any measurement and opinion about the values are approximations.So, there is no need for another property. Measurements have typically known tolerances, which may be statistical, as mean deviations, or absolute.

The property P189 was introduced because of the huge number of geo-referenced resource with no indication how distant or different the approximating area is from the real place. For any approximation with known inclusion or overlap properties to the real place, P189 should NOT be used. A "real place" can be confirmed by multiple observations for things that do not move or have not moved.

This scenario does not exist in the same way for dimensions in general.

I recommend to adjust scope notes and guidelines adequately. If a dimension is given as 10cm, it is per definitionem an approximation, because no natural thing has dimension 10,00000000000000000000000000000000000000000000000000000000000000000000000 cm.

A fine example of measurement tolerances is the recent problem of determining the proton radius:
https://en.wikipedia.org/wiki/Proton_radius_puzzle
See also:
http://pdg.lbl.gov/2012/reviews/rpp2012-rev-history-plots.pdf
https://www.quantamagazine.org/proton-radius-puzzle-deepens-with-new-mea...

I think it is a question of guide lines how to interpret the absence of P10a,b.

Opinions?

Posted by Robert Sanderson on 16/10/2019

Thanks Martin!  A couple of clarifying questions, please …

> The point is, that true numerical values of Dimensions do not exist for continuous value spaces.

Could you explain how you see this being different for E53 Place? The true Place also doesn’t exist as space is also continuous. Doubly so as the definition of place says it is independent of matter. No matter how precise I am about a lat/long/altitude, I still could be more precise. Or more precise about a location relative to an object as a frame of reference; notably as this frame of reference would need to be measured … which would mean that Place would rely on the Dimensions. So it seems like we can reduce the Place approximation to a Dimension approximation, at least in the case of relative coordinate spaces.

> For any approximation with known inclusion or overlap properties to the real place, P189 should NOT be used. A "real place" can be confirmed by multiple observations for things that do not move or have not moved.

And also for this … how would we have multiple observations of the Place, such that it was clear that they were all approximations of a single phenomenal place, without using P189?  For example, I have a bounding box for my city of birth, and a centroid pin for it … I wasn’t born in two places, yet without using P189, I would need to have two P7s … no? What am I missing? 

Posted by martin on 16/10/2019

Hi Robert,

I have been a bit sloppy, as always.

A phenomenal place is thought to be recognizable within some fuzzy limits. So, indeed, all spatial coordinates for a phenomenal place are approximations. For those approximations, we normally use the properties "has former or current location" or "falls within", which both include the true place. That means, that the intersection of all those is still includes the true place. With these properties, I can query absolutely where the place is guaranteed not to be, and within which limits I find it. With P189, we mean an approximation of unknown guaranteed relations to the approximated. So, we cannot query yes or no where the real place is in relation to the approximation.

The same reasoning holds for many dimensions, but there is no typical practice as vague as that of providing a point near a place.

On the other side, many dimensions are not stable over time. For those, each measurement provides another dimension. Many measurements are given with statistical deviation values. The scenario intersecting all measurements to get closer to the real value normally does not hold. It will be a combination of measurement deviations and varying "real value", and intrinsic fuzziness of the property measured.

Therefore I suggest to regard any dimension as an approximation, except for counting stable aggregates of things.

Would that make sense?

Posted by Robert Sanderson on 16/10/2019

Yes, that makes sense, thank you.

One further observation…

> The same reasoning holds for many dimensions, but there is no typical practice as vague as that of providing a point near a place.

I think there’s some very similar practice however of providing multiple values for the same dimension, that at least are roundings from the same measurement.

For example the Met’s descriptions have “H. 14 5/16 in. (36.4 cm)” and similar [1], ours are the other way around “23 x 16.5 cm (9 1/16 x 6 ½ in.)” [2] as does MFA Boson [3], the NGA [4] and many others.

With P90a and P90b we could give a margin of error, but indeed that is not common practice that I can find.

So while the true place falls_within the declared approximations, we cannot say that both 14 5/16 in. and 36.4 cm are close approximations of the same height. They may have both come from different Measurement activities, rather than one being calculated from the other, so we can’t use that as a joining entity.

> I suggest to regard any dimension as an approximation, except for counting stable aggregates of things.

Do you mean then to remove the “true quantity” description from the scope notes?

Posted by Martin on 18/10/2019

Dear Robert,

On 10/16/2019 9:39 PM, Robert Sanderson wrote:
> Yes, that makes sense, thank you.
> One further observation…
> > The same reasoning holds for many dimensions, but there is no typical practice as vague as that of providing a point near a place.
> I think there’s some very similar practice however of providing multiple values for the same dimension, that at least are roundings from the same measurement.

I wold see this as different. Measurements use some device and procedure. Properly document, we understand their behaviour. A spot marked on a map near something has no particular procedure associated.
> For example the Met’s descriptions have “H. 14 5/16 in. (36.4 cm)” and similar [1], ours are the other way around “23 x 16.5 cm (9 1/16 x 6 ½ in.)” [2] as does MFA Boson [3], the NGA [4] and many others.
>
> With P90a and P90b we could give a margin of error, but indeed that is not common practice that I can find.
 

Well, in natural sciences it is. That's what physicist learn to do... Serious publications require it always.
>

> So while the true place falls_within the declared approximations, we cannot say that both 14 5/16 in. and 36.4 cm are close approximations of the same height. They may have both come from different Measurement activities, rather than one being calculated from the other, so we can’t use that as a joining entity.
>
> > I suggest to regard any dimension as an approximation, except for counting stable aggregates of things.
>
>
Do you mean then to remove the “true quantity” description from the scope notes?
Indeed

Posted by Robert Sanderson on 18/10/2019

(Snipping to avoid the list’s length filter)

I agree that it would be lovely if everyone recorded the measurement activities in detail, including with margins of error. But this isn’t a physics documentation group, it’s museums. That others do it better than we do is great, but doesn’t address the current practice of the domain and its information systems.

That said, the removal of the “true quantity” part of the scope note addresses at least the semantic part of the concern, that it’s currently impossible to abide by the definition.  The implementation concern of asserting that two approximations are related can be considered separately

Posted by Martin on 18/10/2019

On 10/18/2019 9:48 PM, Robert Sanderson wrote:
>
> (Snipping to avoid the list’s length filter)

>
> I agree that it would be lovely if everyone recorded the measurement activities in detail, including with margins of error. But this isn’t a physics documentation group, it’s museums. That others do it better than we do is great, but doesn’t address the current practice of the domain and its information systems.

Well, we are concerned with conservation analytical methods, and archaeometrics, which do use error margins. In case of sloppy data by museums, I'd argue we really do not need to care about precise and true values at all, isn't it? We can interpret the absence of error margins (P90a,b) as an approximation without known errors. Isn't it?