Issue 613: Inverse shortcuts
In the 54th CIDOC CRM & 47th FRBR/LRMoo SIG meeting, upon discussing the proposed definition of R78 has alternate (a property that will be used to map IFLA LRM property LRM-R29 has alternate), the SIG became aware that the property is the inverse shortcut of F3.R4:F2.R4i:F3.
Given that the concept of inverse shortcut is not commonly used in the CRM (base & family models), the SIG resolved to start a new issue, where to check inverse shortcuts.
Furthermore, insofar as inverse shortcuts are mentioned in the properties' declarations, there should be a definition of the concept too. This is not only an issue for LRMoo (where to the SIG's knowledge there is at least one such property, namely: R78 has alternate), but CRMbase and the other models as well.
HW has not been assigned for this issue yet. CRM Editors group to appoint a group to do the task.
Rome, September 2022
Post by Wolfgang Schmidle (19 October 2022)
And another one: Are there really no "weak inverse" shortcuts?
Meghini & Doerr 2018 argue that weak inverse shortcuts are possible, although their example looks a little artificial:
E18 Physical Thing P53 has former or current location E53 Place
E18 Physical Thing P161 has spatial projection E53 Place P121 overlaps with E53 Place
The CIDOC CRM document, on the other hand, says: "An instance of the fully-articulated path always implies an instance of the shortcut property." So, there seems to be a change of opinion after 2018.
But this FOL expression that can be spotted in the wild looks to me like an example of a weak inverse shortcut:
E70 Thing P101 had as general use E55 Type
E70 Thing P16i was used for E7 Activity P2 has type E55 Type
P101(x,y) ⇒ (∃z) [E7(z) ∧ P16i(x,z) ∧ P2(z,y)]
The P101 scope note mentions it only indirectly ("This property associates an instance of E70 Thing with an instance of E55 Type that describes the type of use that it was actually employed for"), but I assume it is indeed ⇒ and not ⇔.
Post by Martin Doerr (19 October 2022)
Dear Wolfgang, we had just spotted the first weak shortcut at P7, and I think another one.
Post by Christian-Emil Ore (27 October 2022)
Some starting points:
weak inverse is a somewhat opaque term. The defintion in Meghini & Doerr is not clear:
"inverse weak shortcuts, that is weak shortcuts on the inverse properties, and therefore in them an instance of the shortcut property implies an instance of each of the properties on the shortcut path"
As far as I understand a "weak inverse" shortcut is a shortcut where an instance of the shortcut property implies the existence of an instance of the long path.
The example of a weak inverse property in Meghini & Doerr is dated:
P53 has former or current location: inverse weak
From E18 Physical Thing through P161 has spatial projection,
E53 Place, P121 overlaps with to E53 Place
From v 6.2.8 E18 Physical Thing is no longer a subclass of E92 Spacetime Volume. So the long path becomes longer.
P53 has former or current location:
From E18 Physical Thing.P196 defines: E92 Spacetime Volume. P161 has spatial projection: E53 Place.P121 overlaps with:E53 Place
Is this a weak inverse shortcut? Can the long path be inferred from an instance of the shortcut property inside the frame of an actual KB?
From the introduction to CRM (v.7.2.1):
Some properties are declared as shortcuts of longer, more comprehensively articulated paths that connect the same domain and range classes as the shortcut property via one or more intermediate classes. For example, the property E18 Physical Thing. P52 has current owner (is current owner of): E39 Actor, is a shortcut for a fully articulated path from E18 Physical Thing through E8 Acquisition to E39 Actor. An instance of the fully-articulated path always implies an instance of the shortcut property. However, the inverse may not be true; an instance of the fully-articulated path cannot always be inferred from an instance of the shortcut property inside the frame of the actual KB
Post by Rob Sanderson (27 October 2022)
This seems inconsistent with the definition of shortcut:
A shortcut is a formally defined single property that represents a deduction or join of a data path in the CIDOC CRM. The scope notes of all properties characterized as shortcuts describe in words the equivalent deduction.
"deduction", "equivalent" and "join" are not defined, but given other uses of deduction / deduce (e.g. in the definition of "primitive" and "join" is not used in that sense anywhere else in the document) it reads to me that the definition of a short cut allows one to deduce the existence of the longer data path. Equivalent typically means, and indeed does in the rest of the document, "equal in value, meaning or function".
I would propose that the definition of shortcut be updated to not use the words deduction or equivalent, and to include the information from the second section about shortcuts that Christian-Emil has quoted.
Post by Wolfgang Schmidle (27 October 2022)
I think the definition in Meghini & Doerr sounds opaque because the bit "that is weak shortcuts on the inverse properties“ got lost here. It mixes up inverse properties and inverse shortcuts. Something like this would be fine:
inverse weak shortcuts, in which an instance of the shortcut property implies an instance of each of the properties on the shortcut path
Post by Martin Doerr (2 November 2022)
Inverse weak shortcut = an instance of the shortcut property implies an instance of each of the properties and classes of the corresponding long path.
I think the text in "Meghini & Doerr" is simply wrong. The "inverse weak shortcut" is not a weak shortcut of the inverse property. Weak shortcuts are neutral to the direction in which the property is read. The reading direction of the property corresponds to the reading direction of the long path.
The definition was dropped from the CRM introduction at a time we could not identify inverse weak shortcuts in CRMbase.
Post by Christian-Emil Ore (30 November 2022)
In an email exchange we have discussed the term 'weak inverse shortcut' introduced in Meghini & Doerr. The conclusion is that the term is problematic and Martin redefined it as 'an instance of the shortcut property implies an instance of each of the properties and classes in the corresponding long path', or simply 'an instance of the shortcut implies an instance of the long path'.
Under the discussion of issue 616 'shortcuts in P50 has current keeper, P52 has current owner, P55 has current location' it became clear to me that axioms of the form
are not ideal in a KB since the process to find a possible z is time consuming or worse. For an efficiency point of view we should try to avoid these constructs.
There are two tasks remaining in this issue
- Find a good term
- identify all possible 'weak inverse shortcut' in the CRMBase and the models
In CRM base one can find the possible WIS by a text search '⇔' or variants over 'implies'. Six of the shortcuts in CRM are of the form <the long path> is equivalent to <the shortcut property>. These are:
- P125 used object of type (was type of object used in),
- P156 occupies (is occupied by),
- P167 was within (includes),
- P171 at some place within,
- P172 contains
- P195 was a presence of (had presence)
In addition P7 and P156 implies a long path:
P7 took place at (witnessed):
Therefore, this property implies the more fully developed path from E4 Period through P161 has spatial projection, E53 Place, P89 falls within to E53 Place, where the intermediate place is also defined in the same geometric system.
P156 occupies (is occupied by):
This property implies the fully developed path from E18 Physical Thing through P196 defines, E92 Spacetime Volume, P161 has spatial projection to E53 Place.
(by the way the quantification (0,1:0,n) of P156 may be wrong here since "Therefore, there is a 1:1 relation between the instance E18 Physical Thing and the instance of E53 Place it occupies")
According to the above overview, there are at least be 8 WIS in CRM base.
In addition in the scope note text of P50 has current keeper, P52 has current owner and P55 has current location', the shortcuts and the long paths are said to be 1-1, but this is not reflected in the FOL. It is my opinion that these three properties should be deprecated (freely after Cato the older).
In CRMarchaeo, only one shortcut is declared:
AP21 contains (is contained in) is a shortcut for the more detailed path from E18 Physical Thing through AP18i is embedded, A7 Embedding, AP19 is embedding in, A2 Stratigraphic Volume Unit.
No FOL aciom for this one
The other models is outside my current scope.
Post by Martin Doerr (30 November 2022)
Dear Christian-Emil, all,
I think we need to distinguish KB and ontology. Right now, it appears to me that
P50 has current keeper, P52 has current owner, P55 has current location'
are knowledge base constructs, whereas our FOLs are ontological per definitionem. I believe the better order of reasoning is to question P50 etc. in an ontology, and not this form of FOL in general.
We repeatedly have defined that KBs have some distinct differences in reasoning from the underlying ontology.
Therefore, I regard that 'lefthandside(x,y)⇒ (∃z)[righthandside(x,y,z)]' should not be a question of efficiency, but being.
In general, the existence of a real intermediate z is, to my understanding, not a topic for trying to find it in the KB. In general, such things are not documented, and will never be found. Reasoning, with implied unknown instances can be quite different from finding it in a KB, e.g., that people had been at a place where such things existed, as with the flowers I sent previously.
So, I'd regard this a question of implementation guidelines, exactly as we do with the quantifiers, and a reconsideration of P50 etc., rather than the FOL.