Issue 196: Causation and events
Posted by Martin Scholz on 6/10/11
how can causation be modeled between two events A and B?
The eruption of Vesuv (E5 Event) caused the destruction of Pompeii (E5).
A stabbing (E5 Event / E7 Activity) caused the death of X (E5).
As far as I can see, there is no direct way, i.e. no property for that.
In the last example P15 influenced could be applied if the stabbing is modeled as E7. But influence is weaker than if not different from causation. Also, it cannot be used for the first example, as P15 only applies to E7 Activities, not E5 Events in general. Is there a reason for that? Events could be influenced, too, for example a flood by a dam.
P20 had specific purpose again can only be applied to the last example. Although it implies causation, the meaning would shift to willful killing and exclude accidental death.
A circumscription would be to define an event C and state
C P10 contains A
C P10 contains B
A (P120 occurs before OR P119 meets in time with OR P116 starts) B and infer that therefore there must be some causal connection between A and B. But this is awkward and very indirect.
If there are no better solutions, I propose the introduction of a property PXXX caused (domain & range: E5).
Posted by Martin 0n 6/10/11
Thank you for your suggestions!
Causation is a concept very hard to define. One may regard that the "cause" is the one of all the circumstances that could have most easily be avoided. We do not blame for the destruction of Pompei being built there, even though current research shows that there is a point to it, nor do we blame attribute Caesar's death to having gone to the Curia. In case there would have been a trap, we may blame the trapped for beinf trapped or the trap or the trapper, depending on the social circumstances.
This is generally out of the scope of the CRM, because it goes beyond documentation of facts.
But you can easily make such an extension, if you have good semantics for it.
The robust information behind such causes is the co-occurrence of processes: The process of the eruption of the Vesuv is simultaneously the destruction of Pompei. Indeed, you cannot separate the destruction from the eruption of the Vulcano. In your phrase we also mess up other semantics, such as: "The eruption of the Vulcano left Pompei in ruins" which would be the subsequent state.
The analysis cutting a process into parts were clearly the one is on the cause side and the other is on the effect side is even in physics unclear.
In the CRM we say: "Eruption of Vesuv (E5,E6) P13 destroyed: Pompei(E27)
"Stabbing Caesar (E7,E69 Death) P100 was death of: Caesar (E21)
Indeed Caesar was stabbed in a way that no individual hit should be lethal, such that no individual could be accused of murder. May be missing first aid killed him in the end, or missing modern medicine?
In all current social practice, causation is done by judges, whereas the facts are collected by the police. That clearly shows the different epistemological status of causation from documentation.
This holds for material cause-effect chains. "Causing" humans to do something is described in the CRM as "influence", assuming something like a free will (possibly to die rejecting an order).
We have repeatedly discussed the topic and found that "cause" is a concept for an interpretation model, but not good for the CRM. We found it useful to separate the description of facts from causation.
Please let us know what your concrete application is, and which sort of reasoning you would like to support.
May be there are other solutions to your problem.
Posted by Agnes Thomas on 6/10/11
isn't it that both P15i_influenced and P17i_motivated have domain and range E1_CRM_Entity? So it shouldn't be a problem to use it with E5_Event as well as E7_Activity.
There is also P123_resulted_in, but with domain and range E77_Persistent_Item. It would be really useful to have it for E5/E7, too.
For the Hellespont Project (modelling historical events taken from an ancient greek text), a further question would be how to distinguish exactly between E5_Event and E7-Activity. Is the Persian War an E5_Event or an E7_Activity?
Posted by Martin on 7/10/11
On 10/7/2011 12:53 PM, Martin Scholz wrote:
(by Agnes: Agnes: isn't it that both P15i_influenced and P17i_motivated have domain and range E1_CRM_Entity? So it shouldn't be a problem to use it with E5_Event as well as E7_Activity.)
(By Martin Scholz:Unfortunately this is not the case. The range of both is restricted to E7 Activity (for the inverse, i.e. the domain of "normal" property is restricted to E7 Activity)
Exactly. This link is for cases in which some person or group take up impressions or orders which we regard as having had an effect on the reported activity.
(by Agnes: There is also P123_resulted_in, but with domain and range E77_Persistent_Item. It would be really useful to have it for E5/E7, too.)
(By Martin Scholz: P123 has its domain restricted to E81 Transformation, which is a subclass of E5 Event. This is an interesting property. If I remodel my examples and replace the caused event E5 by some material or immaterial event outcome (an E77), I can express the causing event as cause for the outcome: The eruption of Vesuv (E81 Transformation) had as result (P123) the ruins of Pompeii (E77).
A stabbing (E81) resulted in (P123) a dead person X (E77).)
This is correct, and an additional detail to the solution I have described: It again takes "causal" and "effectual" events as one process, the "transformation". It describes the view I mentioned, that we regard as efect of the event not another event, but the persistent state of things after it. It is correct to characterize an event as Activity, Death and Transformation simultaneously, if the dead body is an object in a museum or otherwise subject of extended documentation afterwards. Please note, that the CRM is not prescriptive!
We should first ask ourselves, what we want to document and what the formal queries for a research question should be. To create in a knowledge base an instance of a dead body for the beauty of being able to say that is not useful.
(by Martin Scholz:This is not exactly what I was looking for (event to event), but it may be an interesting alternative! Although, it's fine for the examples above, I wonder if the restriction to E81 will lead to problems in causal relations.)
(by Agnes: For the Hellespont Project (modelling historical events taken from an ancient greek text), a further question would be how to distinguish exactly between E5_Event and E7-Activity. Is the Persian War an E5_Event or an E7_Activity?)
This may not be the right question. One cannot "distinguish" between a class and its superclass. Rather, the question is, what properties an instance must have to qualify also as instance of the subclass. Hence: the Persian War is an E5_Event. Is the Persian War also an E7_Activity? We can fairly assume plans for any war. So, any war is also an Activity.
(by Martin Scholz: I think, the key phrase in the scope note is that E7 is an "action intentionally carried out" by a single person or a group. As to my understanding, a car accident would thus not be an E7 unless it is provoked on purpose. Neither would be manslaughter/unintentional killing. In the case of a war, I think there is always the possibility for both parties not to battle, so it's an E7.In documentational practise, I can imagine that intentionality is really hard to prove or reject.)
Exactly. In the CRM-SIG discussions, we decided that intentionality is not to be seen to require pre-existing plans, only "active" participation. I may a car accident rather as activity as long as it is not provoked by heart stroke or failure. But in case of doubt, the more general class is always the correct choice, as Martin suggests. The question of classification is secondary to the use of relationships. I'd argue that registering a car-accident as Event with participants, possibly being also death of somebody, is adequate to describe the case.
If someone describes a car accident due to high speed as E7_Activity, a query assuming only E5_Event for an accident will still give the correct answer, due to the smart subsumption...
Posted by Athina Kritsotaki on 7/10/11
I agree with Martin Doerr and I believe in archaeology we make hypotheses on events that might have caused states, but this is part of the interpretation of the archaeologist (which is part of the definition). In a previous project/proposal about how to model periods (a Period Thesaurus- an extension to CIDOC CRM), we used the notion of "starting" and "terminating events" (as a relation between them).
Some distinct events are related to the definition of a period. We questioned that a single historical, religious, military, political or physical event can have a definitive affect on a period or an event. We regard that an event may be catalytic to social change and thus be loosely synchronized with the end or start points of a period/event. We mark them as "starting event" or "terminating event". We do not regard those events as causal and the change of a period or an event may quite well happen without such an event. Therefore we use these events as chronological markers rather than as part of the definition.
Posted by Martin Scholz on 10/10/11
I have no real application. The question rather evolved from a discussion about family relationships and how to relate the conception event and the corresponding birth event (see scope notes P96/97). IMHO, influence is too weak; causation would be more appropriate (apart from the fact that P15 can't be applied). As for our discussion, we do not worry about conception any more, but the question remained. That's why I formulated it abstract and general.
As there are a lot of properties in the CRM that imply causation or interpretation in general (some of these already mentioned) I consider their existence now being due to documentational practise. Interpretational propositions in general, including causation, are dicouraged, though. Right?
Then there is a remaining question:
Why can only E7 Activities be influenced? Things or people could well influence E5 events in general, like forces of nature. Of course, one could ask, how the event would have taken place, if there wasn't that influence, but that's also true for E7.
No need for "cause". Part-of relationship between events and enumeration of constituents is enough for information integration. The notion of "cause" should be part of another ontology with different epistemological assumptions.
CRM-SIG, Amsterdam 17-11-2011