Issue 548: Non-human Actors

Starting Date: 

Posted by Robert on 21/9/2021

Dear all,

In working with our natural history museum, we have a need to assign non-human "actors" to "activities", which is not currently possible.

I think the easiest case to discuss is the construction of a (collected) nest by a (known individual) bird.

We have an identity for the bird (and indeed, we have the remains of the bird!) and we have an identity for the nest that the bird constructed. We can estimate the time when the nest was made, and we know exactly where it was made (due to where it was collected from).
For example: 
Or a dinosaur nest, where the adult and the eggs and the nest are preserved.

If the bird (or dinosaur) could be an Actor, then it would be easy - the bird carried out a Production, during the TimeSpan, which produced the (coughcough)MadeObject, at the Place. However the only thing that can carry out activities is a human or group thereof.

Similarly, the nest might have been built by a mated pair of birds, thereby requiring a Group-like construct for non-human actors as well. 

At the moment it seems like the best we can do is (beginning-of-existence-of-nest)  P12 occurred in the presence of (bird-as-biological-object), which seems woefully inadequate semantically as it likely occurred in the presence of a lot of things, including other birds that didn't actually do anything. The closer subproperty is P11 had participant, which we can't use as birds cannot be actors.

This might also relate to other discussions, in particular:
* Instruments -- the instrument is somehow more responsible for the measurement than the thing being measured. It is at least "instrumental in" the measurement, be it digitally or mechanically.
* Bias -- that animals cannot take intentional actions is a pretty biased viewpoint. Canis virum mordet, not only vir canem mordet. This might be extended to un-observable agents -- a culture might believe that a ghost, spirit, god, or other non-physical entity carried out some action.
* Software "agents" -- even if the software is acting totally deterministically at the behest of another actor, a hard determinist might argue the same for humans.

We could add a property either something like "instrumental in" with a broad range (Persistent Item, as super-class of Actor?) that is less about intent and responsibility, and more concerned with the required-ness of the entity for the event. Or we could go further and create some new classes between E77 and E39 that allow limited performance of activities by non Humans.

Current Proposal: 

Posted by George on 21/9/2021

Hi all,

I think this is a fruitful and interesting topic to take up. While it will undoubtedly take a good deal of work to tease out appropriate, objectively grounded modelling, I think this is an area that long needed looking at. Since we have 7.1.1 firmly under our feet, this seems a very nice modelling direction to follow. I likewise encounter various situations where non human agents are in scope of the CH data but as of yet not in scope of CRM. Would be great to work on expanding the ontological umbrella to manage those elements. 

Posted by Martin on 21/9/2021

Dear Robert,

I support this.

I suggest the non-human Actors to go into CRMsci. It is a straightforward extension of scope, and has been discussed in the past. Non-human actors cannot be hold liable, and will not report. They are obviously a sibling to the human actors, and fall under a common generalization. In the same way, we have generalized over physical things in CRMsci.

I think any opinion that animals in general cannot take intentional actions has been proven non-sense. Conversely, human actions are often enough instinct driven.

So far, I do not think we have evidence of conceptual objects created by non-human actors. Whales may turn out having oral traditions in the future. Bird songs are, however, partially tradition and not innate, but we miss the creator individual...

Posted by Mercedes Menendez Gonzalez on 22/9/2021

Dear all,


Although I am quite new to this, I would like to contribute my opinion on this interesting topic, if I may.

I agree that the most suitable option seems to be to create a class or some new classes for non-human actors. Going back to Rob’s example, I would say that the bird carries out an intentional action when it designs and builds the nest with very specific purposes (to lay eggs that have a specific size, to raise offspring).  We could even think on nest construction as an individual action as well as a collective behavior.





*I take the opportunity to thank you for the invitation to participate in this forum and to introduce myself. I am Mercedes Menéndez, PhD candidate in Art History at the University of Oviedo, Spain.



Posted by Martin on 22/9/2021

Dear Mercedes,

Thank you for your good comments! What we would need now most are real data examples tracing individuals.

Posted by Mercedes Menendez Gonzalez  on 25/9/2021

Thank you for the kind words, Martin.

A brief try, could we find a good example in chess artificial intelligence? The human and the computer perform equivalent roles as (participants) players. For instance, the IBM computer named Deep Blue beated Kasparov in a well-documented match on May 11, 1997, at the Equitable Center in New York.

Also, with my apologies if I am misunderstanding things.

Posted by Martin on 27/9/2021

Dear Mercedes, all,

My position is that machines are not actors. They are robots, that work on behalf of human actors, following human instructions. Their use is regulated by laws concerning those activating them, and not for suing the machine for its initiatives. There is no fundamental difference to setting up traps, no matter how complex the machine and its instructions are. Non-human actors should be restricted to living beings. Robots and traps and events set in action by them should be each a different category, and this is a nice, but different, challenge to model as well. Opinions?

Posted by Robert Sanderson on 27/09/2021

Could it be kept open until there's a clear cost / benefit established, rather than philosophy around free will? 

For example, if the ontology allows things that should be perdurants to become endurants through agency, then we've messed up a fundamental design decision. For example, a fire might "carry out" the destruction of an object, but it's not an actor. But a self-driving car seems to have more "agency" than the cyanobacteria "responsible" for creating stromatolites ( A tiger escapes its enclosure at a zoo and eats a child ... the tiger carried out the eating, but can't be held legally accountable. The zoo on the other hand maybe could be ... but the zoo did not eat the child.

There's lots to unpack ... it would be good to determine how far we can unpack it as part of the process, while respecting core design values.

Posted by Mercedes Menendez Gonzalez  on 28/9/2021

Dear Martin, Rob, all,

I chose this case because it seemed to me a potentially problematic example, when it comes to think about possible equivalences between human and non-human agency. I assume that the (ontological) difference between living and non-living beings must be established, in some way.

In any case, I would not like to monopolize too many lines, as I am aware that my experience in this field is very limited. For the moment, I can acquire very much knowledge, rather than contribute it (so thank you all).

Posted by martin on 28/9/2021

Dear Robert,

Please excuse my sloppy shorthand! Of course I meant that a machine capable of causing events in reaction to external stimuli in a controlled manner is a new class model, AND the reactive events are another new class which should be related, it didn't come to my mind it could be one

I just expressed my opinion. I have not made any decision. E39 Actor clearly excludes machines and animals so far. My argument is neither philosophy about free will, nor an interpretation of the word "agency", which would be a linguistic argument.

From a methodological point of view, the only thing  that matters are the properties we associate with these things in documentation practice. Practice, and not philosophy, is, e.g., that a machine cannot be sued, but those setting them up in this manner. This is different from suing the owner of a tiger.
The first thing to look at, in a bottom-up manner we are committed to, is to make ontological distinctions, not extending existing concepts into new domains. There are, to my opinion, much more things that differentiate Actors and Activities from robots and their reactions which I have not listed.

Only after we have carefully investigated that there are enough commonalities between originally distinct concepts, we can decide if they warrant a common superclass.

Both I have not seen yet.

Would that make sense?

Posted by Robert Sanderson on 28/9/2021

Yes, understood and agreed! Was just trying to clarify the process. And in particular, the properties (and class hierarchy) are very important. Scope notes can be ignored by humans (at their peril), but it's much harder to ignore the ontology definition.

For documentation practice, I think most systems I've seen would say that software does things, especially in digital preservation where the software's actions must be  auditable (if not accountable).  I do worry about legal responsibility as a factor in deciding agency/non-agency however, given different jurisdictions and legal systems, but I also understand the rationale.

Posted by martin on 28/9/2021

Dear All,

Robert, all, I think it would be good to have progress in reviewing the PARTHENOS model. It contains a quite elaborate model of e-services, and makes subtle distinctions beteen maintainers, machines, and software installed. A lot of aspects may already be resolved there. The model has been implemented and used in a large EU Project.

Post by George Brusker (11 October 2021)

Dear all,

In preparation for the discussion of non-human actors as related to use cases arising in Linked.Art (inter alia), Rob and I have sketched some ideas back and forth to try to find a monotonic was to add the agency of animals in the first instance into CRM (proceeding in an empirical bottom up fashion) and then see where else we might also get added in (searching for the sibling class that Martin suggests and the generalization that it would need).

The linked sketch provides a proposal for discussion. The background is given already in this issue. ( (png)

Here is some argumentation.

Up to now, CRM takes its scope as related to documenting intentional acts of human beings. Its top level class then has been E39 Actor which gives properties which allow the assigning of responsibility for an intentional activity. It has two subclasses, E21 Person and E74 Group. These two kinds of being have different behaviour, therefore properties, therefore classes.

If we expand the scope (in base or in sci or wherever) to include animal agency in the first instance, then we must have a way to monotonically generate this extension (we don't want to just expand the scope of E39 Actor because then we will end up with rabbits being responsible for financial crises and murders and all sorts of nonsense).

So we want to introduce a sibling class for E39 Actor. Call this biological agent. Instances can be anything biological. This would obviously be some sort of a superclass of E21 Person, since all persons are biological actors as well. It would be a subclass of biological object since all biological agents must be biological. (but not all things biological are biological agents)

Then we would want a general class that subsumes the agency of purely human actors and biological agents. This would be our top class. Here we come up with a more general notion of agency. Whereas E39 Actor was declared in order to account for a 'legal persons notion' of agency common to Western legal systems etc. (and is perfectly adequate for the scope of CRM Base), this would be a broader notion of agency. 

In order to avoid impossible philosophical arguments around self consciousness, we can give a more externalist scope note / intension to this class. Agency has to do with those entities which display self organization and action towards an end from an external perspective. This way we avoid having to know if the other really has a self. If it looks like it is acting intentionally and people document it as such, then so it is

This now gives us a super class (and eventually super properties) for all agents. 

But wait... we need more.

CRMBase distinguishes between persons and groups. Whereas persons must have both agency and be individuated corporeal beings, groups do not. Persons are atomic and irreducible (can't be made up of more persons, can't be spread over multiple bodies / time zones). Groups are composed of persons and groups. Groups are inherently collective.

If we wish then to have this same distinction reflected into the biological domain we would need a class for individual biological agents parallel / sibling to person and a class for collective biological agents, parallel / sibling to group.

Doing this one would then need the superclasses to subsume these divisions. Hence:

Individual Agent: subclass of Agent, superclass of individual biological agent

Collective Agent: subclass of Agent, superclass of collective biological agent and human group

This finally allows us to have:

Individual Biological Agent: subclass of Biological Agent and Individual Agent: used for individual birds, trees, and other biological actors

Collective Biological Agent: subclass of Biological Agent and Collective Agent: used for flocks, forests and other group biological actors (unlike human groups, such groups are inherently corporeal)

And at that point we might consider renaming our existing classes to 'human' xxx


E39 Human Agent: subclass of agent, no real change in intension, the kind of entity that can take action for which legal responsibility can be attributed within human cultures societies

E21 Human Person: no real change in intension but its superclass becomes individual biological agent and human agent (ie an animal that can be held legallly responsible for its actions)

E74 Group no real change in intension, but it gains a super class Collective Agent so it can be queried together with other agent groups.

This analysis does not get into the properties which are, of course, fundamental but sketches a possible path for creating the structure necessary to create this extension of scope in such a way that it would respect the principle of monotonicity in revising the model while allowing the growth of the model to handle the many use cases of documented animal agency that fall within CH institution's documentary scope.

Hope this is a good starting point for a constructive discussion!




Posted by Martin on 11/10/2021,

Dear George, Robert

This makes generally sense to me as a discussion starting point. However, I‘d like to remind you that our methodology requires first a community practice of doing documentation about such things, and second domain experts for concepts that are not our primary knowledge.

To my best knowledge, there does not exist any reliable concept of what individuality means across the animal kingdom, nor what a collective of such individuals is. There is an unbelievable complexity to these questions. We know from experience that any global widening of scope can blur all distinctions ontology enginerring relies on. Therefore I'd regard it as most important to find the experts first and let them speak.

The reasons why we did not model animal actors is precisely the lack of an experts group to communicate with.


...I'd like to add that behavioral studies with birds and mammals may be a reasonable scope, but, experts need to speak I think!

Posted by Robert Sanderson on 11/10/2021

Could we clarify what sort of expert we're looking for to move the discussion forward? In particular, natural history museums seem to be at the critical intersection between CIDOC and the activities of animals. I can represent the sorts of documentary evidence from that side, and happy to reach out to colleagues at other NHMs. So I think the first aspect is covered, but I question whether we (as modelers of museum knowledge and documentation) /need/ to understand animal individuality or behavior in order to take the first step of describing an animal performing some action. Conversely, my experience has always been that when there is something to react to, it is much easier to engage with outside specialists.  It is easier to ask for opinions on something than it is to ask them to help come up with the interdisciplinary model.


I also don't think it makes sense to model animal actors in great detail, down to the same level as the differences between classes in CRMTex for example. The baseline that we need to start with is much simpler.  If there isn't a fine grained concept of animal individuality, I don't think that means we can't model an individual animal at a coarser granularity, just that we shouldn't allow the ontology to describe anything that we don't understand. Even as a non-biologist, I know without any hesitation that the bird laid the egg in the nest in the Peabody Museum of Natural History, and that the herd of dinosaurs created the footprints preserved in Dinosaur State Park up the road from us. I know that a sheepdog can herd sheep and makes decisions about which way to run to accomplish the aim of getting the sheep into the next field (and when I was a little lad played the part of such a sheepdog for my uncle in New Zealand). How does the sheepdog know? Does it know that it knows? If we study 100 sheepdogs individually and in groups, what do we learn about sheepdog behavior? I don't care, and I don't think any other museum oriented documentation system would either :)

Posted by Martin on 11/10/2021

Dear Robert,

Having collaborated with natural history museum colleagues for some years and designed a research infrastructure for biodiversity in Greece, I understand that they normally do not describe the actions of an individual in a way that information integration on the base of the individual's animal actions would be needed. They would rather state the fact that an individual of type A, showed individual behavior pattern B. They would integrate these data on a type base, and not on an individual base. We have at FORTH converted Darwin Core data of occurrences of individuals into CRMsci representations. That had so far covered the needs.

A colleague in Britain had used, I think, CRM for modelling observations of Caledonian Crow observations. Since these crows do not travel, the relevant information access and exchange is still on a categorical level.

Migratory birds tracking may be an application, but normally they do not describe other behavior than move, in which case we can use a Presence construct for the migration paths.

Our collaboration with NHM showed that they often prefer not to use CRM for their observation data. In a large European Project, we were forced to cheat and rename all CRM concepts, so that they appeared under a "BIO" title.

So, in short, we need an expert that would show us practice of modelling animal actions individually, and be willing to consider CRM...

Posted by George on 11/10/2021

Hi Martin,


I think Rob listed in the introduction to the issue the use cases of documentation of individual action of animals. 


It would seem that natural scientists don't only study species but also individuals. 


Here's a smattering of pieces culled from casual reading in the past few weeks with nice motivations and examples for these new classes. 

Posted by Pat on 11/10/2021

Just to remark that the library world discussed non-human actors for many years (in the literal sense of actor as in the dogs that portrayed Lassie in the TV series, or that portrayed Sykes and Paddy from Midsomer Murders, somehow it is always cute dogs that are brought up in the discussion). 


The desire was to list the named animal actors in the credits for the cast of a film and provide access via their "real" names the same as for the rest of the cast, and so using the same mechanisms as for human actors.


This sounds like it might be fine until you realize that making the dog a valid LRM-E6 Agent means that it can have the full range of responsibility relationships to works, expressions, manifestations and items. Which becomes absurd.


And while is it understood that one can easily film an individual animal, it isn't clear that it is behaving as an actor intending to create a cinematographic work in the same way that the human participants. There was also no clear consensus on which sorts of animals were individually interesting enough to merit this treatment, rather than just being viewed as an instance of their species (as in nature documentaries).


The animal agent option was rejected in FRBR and again rejected in LRM, and a LRM-E6 Agent (= E39 Actor) remains restricted to either individual human beings (LRM-E7 Person) or groups of human beings (LRM-E8 Collective Agent, or F55 Collective Agent in LRMoo).


The current compromise is that the animal actors, if it is desired to provide access points for them, are established as instances of a subcategory of LRM-E1 Res that is disjoint from LRM-E6 Agent. There was talk of creating some guidelines for this at one point, but I have not followed the issue since then.

Posted by Robert Sanderson on 11/10/2021

Hi Pat,


While that is certainly true from a model-theoretic perspective, in practice authorities simply create Persons for them which is, in my opinion, even worse because there is a demonstrated need which the modeling is intentionally preventing.


For example in the Library of Congress:

Real animal/people:


  Misha the Dolphin:


And fictitious:

  Odie (from Garfield):

  Grumpy Cat: 


In ULAN, here's a racehorse/person: 


ISNI has a dog/person called Maggie Mayhem:


And so on.

Posted by Pat on 12/10/2021

Hi Rob,

Looking at the dates on Lassie and Misha, I see that they were created during the phase when people were trying this under an unwise modification to RDA, and not been revised since. This would no longer be valid under the latest RDA. And no one has bothered to propose MARC coding specific to this type of heading, leading to the ones that were created being shoe-horned into the personal name coding. The proportion of the huge LC names file is too small.


As for the fictitious, that was a completely different argument that has also lasted years. Stems from a difficulty in distinguishing between a name and the reality behind it. 

But these two issues are frequently conflated in the library world by people trying to use discussion related to why one was invalid to imply the position on the other issue didn't make sense.

The thing is that there is no problem about having a work about an animal or about a character (as a concept), or have photographs, films or sound recordings of an animal. but it doesn't make sense to set up a relationship where these own an item, publish a manifestation, write, compose or translate an expression, or create a work. So the relationship is other.

And a person can choose a pseudonym of any sort (even one that evokes a pet name or is the same as a fictional character), that still doesn't make the person into a pet. Same as two people having the "same" name doesn't fuse them into a single human being in some sort of weird siamese twin situation.

Anyhow, I just wanted to to point out that there has been a lot of ink spilled over these issues, to no real result.

Post by George Bruseker (12 October 2021)

Hi all,


Here are some examples of databases that deal with individual or collectivites of animals NOT as THINGS but as AGENTS:


EMU: Pest Tracking in Museums


Here's a database that tracks the migratory paths of individual birds:


Here's a database that tracks orcas:


Here's a database that tracks gorillas:


I would say that often something doesn't get documented because it is silenced by the information systems available (see the terrible gorilla database), arguably what CIDOC CRM is supposed to aid in getting out of (viz. Dominic's textual works issue and documenting context). The fact that people are forced to shoehorn identifiable individuals that they want to document and have discourse about into classes that do not suit them is for me the obvious argument for making classes and properties! 


Whether there are explicit fields for such data, the natural world is something which unsurprisingly Cultural Heritage is interested in and refers to. Orcas are, for example, highly important animals within different cultural systems in Canada, they are documented and they are documented not as things but as agents. So what is the pressing counter point to allowing this expressivity? That there are too many classes and properties. Many would make that argument about CRMinf or about any of our extensions. I suppose it depends on where you interest lies. By not opening these categories we effectively mute/suppress this voice. Because the limits of the world are my language when we choose to oppress a class we choose to oppress the ability to express that object. Or we indeed force the documentation of things that are considered agents as objects. This seems the greater harm to my mind.


On the expertise question, I am not sure if we required a biologist to be able to model the notion of Birth or Death. Did we not use a middle level understanding of everyday objects and their documentation in systems in order to be support the recording of standard kinds of facts of interest to a researcher? Birth and Death are not high concepts of when conception begins or when the soul leaves the body, they are rough and ready everyday ideas of, there was a person and an event led to its end, there was a person and an event led to its death. How the case of modelling animals differs is not clear to me. Did we bring in financial experts model the payment class? On which issues we need an expert and on which issues not is not clear, nor is that expertise counts. As Rob says, having many years of experience in cultural heritage documentation and analysis of such systems does not count? I would think in basic matters like this, it goes back to the ground of coming to a common sense modelling in line with what is considered the best state of knowledge regarding the world. We KNOW that the best state of knowledge is not represented by the present modelling because agency is not just attributed to human beings. Therefore, we are presently deliberately out of synch with the best state of knowledge. I would think it behooves (pun intended) us to step up to the plate and get on to making it possible to express basic facts about the world that can be and are referenced in CH data systems (such as the existence of animals!).





Post by Martin Doerr (12 October 2021)


Hi George,

I'd prefer to let the biologists talk about that. To my best knowledge of real cases, this is a much debated question. For the time being, I am sorry I have no time to provide details.

All the best,


On 10/12/2021 10:02 AM, George Bruseker wrote:

On the expertise question, I am not sure if we required a biologist to be able to model the notion of Birth or Death.

Post by George Bruseker (12 October 2021)

Hi Martin,


I'm also pressed for time but above wrote out an argument. 





On Tue, Oct 12, 2021 at 10:17 AM Martin Doerr <> wrote:

Hi George,

I'd prefer to let the biologists talk about that. To my best knowledge
of real cases, this is a much debated question. For the time being, I am
sorry I have no time to provide details.

All the best,


On 10/12/2021 10:02 AM, George Bruseker wrote:
> On the expertise question, I am not sure if we required a biologist to
> be able to model the notion of Birth or Death.

Post by Athina Kritsotaki (12 October 2021)


I am probably missing something here, but regarding these databases, in which cases these animals are documented as actors? It seems that there are documentations about births and traps and capturing events, but the discussion is about activities carried out by them, right? From my experience with gbif and darwincore, which a standard that is widely used for biodiversity databases, haven't seen definitions of this kind of relationships, but  maybe I am missing things
or I misunderstood something


Post by Dominic Oldman (12 October 2021)



Just for clarity (I'm not sure whether George's reference was to me), my argument was that there are many practical things we can do to encourage historical contextualisation with the CIDOC CRM without any model changes, which we should be paying more attention to, and this is related to the scope of traditional documentation practices and museum documentation worldviews. I think this is a significant and large issue. I will release a sub-committee statement addressing this during the week to the CRM SIG which covers wider contextualisation and issues like mutual respect. 


I think I understand at a high level the broad issues on both sides of this argument, and I can see from the tone of the discussion that the disagreement requires further investigation, and I would certainly like the opportunity and time to talk to people and read about these issues myself. 


It is key to me, and the reason I, and others, became involved in the ontology, that it adheres to certain scientific principles.  


I think it would be helpful to restate these principles clearly so that at the very least we have a proper starting point for these discussions, which would also be helpful to anyone reading these discussions and wanting to contribute. I will initiate something along these lines. 






Post by George Bruseker (12 October 2021)

Hi Athina!

What causes the bird to go from a to b, I guess is the simplest way to put it. Does it just happen? Is the bird just present in the event of its migration?


On Tue, Oct 12, 2021 at 11:22 AM athinak <> wrote:

I am probably missing something here, but regarding these databases, in
which cases these animals are documented as actors? It seems that there
are documentations about births and traps and capturing events, but the
discussion is about activities carried out by them, right? From my
experience with gbif and darwincore, which a standard that is widely
used for biodiversity databases, haven't seen definitions of this kind
of relationships, but  maybe I am missing things
or I misunderstood something



Post by Philippe Michon (14 October 2021)


Dear all,

I am not an expert in biology, but the question has aroused a lot of interest within my organization and I, therefore, allow myself to provide some animal examples within museum databases.
I don't know if this example has been mentioned before, but the animals that received the Dickin medal are in my opinion good examples that animals participated in wars.

As an example, the Canadian War Museum documents a Dickin medal of a dog named Gander. Although Gander is not widely documented on the museum's website, his Wikipedia page highlights several of his heroic acts.

Another similar example would be the rescue dog named Rex (also documented on Wikipedia) which is associated with 4 items from the Imperial War Museums. In addition to the Dickin medal, they also document his harness
Here is a short description from the harness record:
Harness associated with 'Rex', a British Second World War Home Front 'rescue dog'. 'Rex' was awarded the Dickin medal on 25 April 1945 for outstanding bravery in finding bomb victims, trapped under fallen debris.
Returning to Canada, the Canadian Museum of History documents 116 artifacts (mostly trophies) won by horses in various horse racing championships. The majority of horses are identified by their names such as Northern Dancer for example. His biography is really well documented on Wikipedia again.
It also seems important to me to underline that the information on these horses is currently recorded in the "Person/Institution" field with an "Associated Famous Animal" tag. I don't know the reasons for this choice, but I found the association between person and animal worth mentioning and probably questioning.
I hope these few examples highlight the desire of some institutions to document the participation of specific animals in documented events.
Finally, I also wanted to mention this Twitter thread where a few people are discussing the best way to describe a goat within their organizations.

Post by Martin Doerr (14 October 2021)

Dear Philippe, All,

These are good examples. Could you provide sample records? We should check the schema used for migratory species tracking, and find an expert for that. Presence in events may be enough. Kea, Caledonian Crow and other behavior studies typically are done in controlled Lab environments, and therefore do not pose a problem of historical fact integration. Integration needs are on the kind of abilities, rather than the actions. Would need an expert.

I think the examples show that we talk about some subset of animals that are relevant, and do not pose problems of individuality.

Further, I think machine events are much more important to discuss. I suggest to talk about machine "reactions" "delayed reactions" stimulus induced reactions, predetermined. So far, robots with random behavior are not particularly useful.
Software is instructions, not "active" or "reactive". What make a robot appear to have "agency" is nothing else that it work with a clock inside. There are traps, bombs triggered by clocks, mechanical vending machines etc, which generalize over the concept of deferred machine reactions.

By the way, the term "agency" as a property has no good translation to German.

All the best,


Post by Rob Sanderson (18 October 2021)

Hi Martin, all,

Perhaps I am misunderstanding your comment about individuality ... but there are clearly individual animals in both my use cases and Philippe's. The twitter thread that Philippe linked to demonstrated the very common nature of museum collections having to work around the lack of a non-human agent class.

Given the lack of discussion at the SIG, the seeming lack of interest in moving this forward in a timely fashion, and the demonstrated need to document interactions between animals and other entities described in at least natural history collections, such as the original use case presented of the bird (identified and collected instance rather than type) that built the nest (again, collected instance), plus the ones from Philippe, it seems inevitable that something outside of CRM base is required.  There is a precedent in CRMSci for inserting classes into the hierarchy, and as the proposal from George is monotonic, it seems an extension could define the solution as presented.

As such, we can take the proposed solution back to the Linked Art group to discuss in that context and if, in the course of time, the SIG changes its opinion or comes to a solution, then the extension can always be rolled into either a SIG based extension or into base.


Post by Dominic Oldman (18 October 2021)

Hi Rob / All,

Personally, I am still doing some reading around this which is quite varied (moral, legal, projections). I think that the development of a Social Ontology extension may also have a bearing.

In the meantime I came across the following model (Ollman) which reflects the generalisation/specialisation approach of the CRM in terms of different levels of generality. It goes something like this.


  1. There are unique aspects of particular people (individuals) and particular events they (individuals) participate in. 
  2. There are general activities related to groups of people. 
  3. General characteristics that can be applied to people within a particular phase of human history. 
  4. More general social division across different historical phases. 
  5. People and activities which form part of the human condition.
  6. Animal life more generally.
  7. As a material part of nature. 


The author assumes that we are set apart from animals, but we share some general characteristics ("life functions, instincts, and energies"). I'm looking at other conceptions also.

Very best,


Post by Martin Doerr (3 November 2021)

Dear Robert, all,

In response to your message from Oct. 18, "but there are clearly individual animals in both my use cases and Philippe's" and "Given the lack of discussion at the SIG, the seeming lack of interest in moving this forward in a timely fashion, and the demonstrated need to document interactions between animals and other entities described in at least natural history collections"

I kindly ask you to consider the following:

Please understand that the question of non-human actors is a fundamental widening of scope of the CRM, and cannot be answered as quickly as submissions of individual properties for well-understood classes.

I may also remind you, that the CRM-SIG is committed to the quality and robustness of proposed solutions, in order to minimize non-monotonic changes in the future. Given the amount of work, priorities can not always be accommodated to the needs of member projects to proceed, and CRM-SIG depends on extensive homework by members. Therefore, "... in a timely fashion" seems to be a misunderstanding of how fast a committee of volunteers as the CRM-SIG can act at best will, and who is in charge of elaborating issues until a sufficient maturity so that a majority of SIG member can develop a justified opinion.

Let me again stress that CRM-SIG follows with the CIDOC CRM a quite distinct modeling methodology. It is vital to follow this more or less, so that a well-structured discourse can be achieved.

Starting with defining the animal actor does not really provide a good entry point to the problem, because we first need properties for the individual use cases, in order to decide which precise form of such a class, if needed at all, is fit for the purpose. The undeniable fact, that "there are clearly individual animals in both my use cases and Philippe's", is, as in many other cases we have modeled in the CRM, on its own not yet enough to justify and implement such a new class in order to solve the interoperability problem the use cases pose, which is the ultimate utility of CRM modeling.

Rather, we would need to start with "need to document interactions between animals and other entities", because interactions yield properties. My critical remark above "if needed at all" about an classes should never be interpreted as a "lack of interest" nor questioning facts or the need for modelling such interactions (I have also declared my personal interest in this issue). Rather, is an application of Tony Gruber's principle of "minimal commitment", a criterion to be applied objectively, at the end of any modelling exercise.

It rather appears to me that questions of CRM-SIG members to your approach and advice for more background knowledge so far have unluckily be misinterpreted as lack of interest.


Please let me therefore go through the beginning of a modelling exercise, in order to illustrate the method, not to preempt the result, and to move this forward:

We do have a class for animal actors, which is E20 Biological Object, to which humans belong. Without specific properties, which must be supported by use-cases, we use p2 has type for specializations, because classes without properties do not support propositions, and there are published CRM-SIG rules to deprecated classes without properties at the end of modelling a subject.

Therefore you may not be surprised about lack of reaction to your approach so far.

Taking the birds nest, we do have the superclass of E12 Production, S17 Physical Genesis, with the property O17 generatedE20 Biological Object can also be used for the nest, the product.

Necessarily, the generation of the nest must be an instance of S17, directly or indirectly via a subclass.

Methodological Question 1:  Do we have data for the individual Genesis events?
                                   case A: We do not. Then, a property "E20 Biological Object - Oxx1 product of species: E55 Type" would be enough.
                                   case B: We do have contextual data of the Genesis event, which are distinct from the encounter event. Then, S17/O17 should be used.

We do miss the actor in this class. This is true.  We miss it also for plants. We have fruits etc. We even have carnivore plants.

Top priority for defining the cause of genesis for objects like nests is the generating species, which is often characteristic for the product. Undeniably, in most cases of Natural History specimen, the generating instance is not individually known or documented, even if photos exist.

Then, if we create a subclass of S17, "Sxx1 Biological Genesis", we will first need a property Oxx2 done by species: E55 Type.

       We are now confronted with the scope of  Biological Genesis!

                                     Is it animals? Why not plants? Why not Bacteria? (, the oldest evidence of life on this planet, by
                                                the way)

The answer to the adequate scope should be given by experts and/or by examining standard ways to document products of biological genesis.

In very rare cases, as you mentioned, the generating instance is documented beyond species level or even extent in collections, or still living in nature. For these cases, we would need:

                            Sxx1 Biological Genesis. Oxx3 done by: Sxx2 Living Being ?

Comment: Here we could use the Animal class. But what would be the point to reduce it to animals? By the way, human birth should be regarded as an instance of S17...

Note also, that a class like "animal action" does not solve the production question yet, because it misses the binding to generating something, so we would need to specialize it anyway to a form of genesis.

Methodological Question (recall-precision argument):
                                       If we document the instance not with "Oxx3 done by", but only with "P12 occurred in the presence of", and a suitable text, which is
                                        anyway required, what research questions would be compromised by too many unrelated answers? What is the expected              
                                        precision/recall ratio for a query : Get instances of "Sxx1 Biological Genesis" where  P12 occurred in the presence of: refers to an instance
                                        of  'E20 Biological' that is not also referred by O17 generated? (e.g. the dinosaurs preserved on their own eggs and birds with their own 

                                       If the precision is sufficient to easily sort out false positives, this scenario would not require Oxx3 for purposes of interoperability.

Methodological Question (efficiency for the most frequent cases)

                                       If we adopt Sxx1 and Oxx3 only, Oxx2 and Oxx1 are shortcuts of Oxx3. Whereas Oxx1 misses the whole individual context, Oxx2 provides                                         the complete context, and only misses the specificity of naming the property to the acting individual. On the other side, the P12 solution
                                        may deliberately let open the missing scientific knowledge about precise roles of involved individuals.

                                        Note, that the CRM is not intended as an exhaustive language to describe facts of this world. For this purpose, ontologies like WordNet,
                                        SUMO, Cyc are much more adequate.

Following the above analysis, the birds nest scenario seems to provide
                        A) a strong argument for a class Biological Genesis, and at least a property "done by species".
                        B) it provides a much weaker argument for a property "done by" individual.
                        C) Given the paramount importance of documenting whatever species is responsible for a specimen or feature, a restriction to animals would not                             yield a more expressive property that holds only for animals.

I'd also like to stress that the obvious analogies of human and animal behavior, and considerations of agency, are a concern of philosophy and behavioral biology,  but not in isolation a modeling criterion for developing models of instances that link human documentation about such subjects effectively.

This chain of argumentation should, to my opinion, be continued for the other use cases, such as:

                   *  Tracking animals (normally done by Darwin Core records we model as Encounter Events. I have asked a colleague for representative examples
                            from LifeWatch Europe)
                   * Race horse bios (note that race horse actions are normally under complete control of human activity)
                   * Bug damage in art conservation (normally documented as alterations/ deteriorations with categorical cause, individual bugs would normally                                           appear as sample taking)
                   * Behavioral studies (here we are indeed at the action level, need to understand the precise interoperability demands from experts. Is the URI of an                                    individual crow indeed the entity that will link data collected between different research teams, such as author URIs link library data across                                     the world? Or do other identifiers make the critical connection? Note that the CRM aims primarily at inter-institutional integration. This is
                                    explicit in the Introduction. Actions can have an immense complexity, from tools use to social deception and organized murder. Are there    
                                    powerful generalization?)


Given the complexity of the issue, we need to have a good understanding of all related use cases. Then, the last step will be looking at individual solutions together, find possible commonalities, contradictions, generalization, simplifications etc.

Please comment!

All the best,

Post by Oyvind Eide (4 November 2021)

Dear Martin,

thank you very much for this extensive overview of the modelling questions and challenges. I want for now to pick up on one small comment only. 


This chain of argumentation should, to my opinion, be continued for the other use cases, such as:

                   *  Tracking animals (normally done by Darwin Core records we model as Encounter Events. I have asked a colleague for representative examples
                            from LifeWatch Europe)
                   * Race horse bios (note that race horse actions are normally under complete control of human activity)

I have discussed this quite a bit with horse trainers connected to some work on agency some years ago, and also, the case of reindeer herding dogs.

I have not studies race horses so you might be right that they are under strong (if not complete) human control. However, I would be happy to show video evidence and professional views at a meeting (based on practitioners, not on researchers, I must add) indicating that the claim for rider control at the micro level in other and more complex horse operations is not plausible given the speed of movement in some types of action such as combat situations. Human-to-animal communication is just too slow explain the movements horses do when, for instance, galloping sideways or backwards in time critical operations. In my understanding, a level of double agency in negotiation, or some sort of symbiotic cooperation, explains the evidence better. I think we need to include horse decision making in such cases. 

As for dogs herding, autonomous decision making by the dog based on general guidelines, extensive training, and frequent cooperation with the human(s) also seems plausible. 

All the best,


Post by Martin Doerr (4 November 2021)

Dear Øyvind,

Thank you for comment. Actually I would have never assumed anything else than what you describe below.

What I wanted to make clear with my previous message is exactly this difference between integrating documentation based on unambiguous ground facts as they are expected to appear widely in documentation, versus our common understanding of the relevant reality beyond these facts. It appears to me that the aspects you describe are well-placed in narratives giving adequate credit the subject.

The CRM is not even suited to describe such details for human actions, nor is it intended to provide an adequate account of the details of reality. Firstly, the reduction of aspects of reality to FOL predicates is a horrendous restriction. It doesn't even work for the weather forecast or anything approximately described by differential equations. Even if it were possible, we would need a huge ontology, and I personally would regard it as a cultural nightmare. All statements I have made in my previous message have nothing to do with a simplified world-view of my own.

Let me repeat the methodological argument in my previous message:

Methodological Question (recall-precision argument):
                                       If we document the instance not with "Oxx3 done by", but only with "P12 occurred in the presence of", and a suitable text, which is
                                        anyway required, what research questions would be compromised by too many unrelated answers? What is the expected              
                                        precision/recall ratio for a query : Get instances of "Sxx1 Biological Genesis" where  P12 occurred in the presence of: refers to an instance
                                        of  'E20 Biological' that is not also referred by O17 generated? (e.g. the dinosaurs preserved on their own eggs and birds with their own 

and : "Note that the CRM aims primarily at inter-institutional integration. This is explicit in the Introduction."

Similarly, I implied, that we should discuss:

Methodological Question 2 : If we document race horse bios with CRM propositions only in the framework of the human-initiated actions, rather than all the
                                                    descriptions of the horse's own initiatives, (and describe such initiatives at a type level for example, a subject index to a
                                                    narrative for example), what research questions would be compromised by too many unrelated answers? What records would
                                                    we miss, if we get the full records back, based on queries that refer only to the human-initiated actions?

            Of course, "used specific object" for a race horse in a race is utterly ugly, but not wrong, and a ground fact.

Please compare the integrating power of the URIs of horse-own actions in formal propositions to the integrating power of the framing human activity within which the horse acted. These are the critical questions. The reality, as you describe, is hopefully well-known to all of us (I have enough smart animals in my garden, and a smart grand-daughter in the age of 18 months).

Please note, that exactly these kinds of arguments have been done regularly in the development of the CRM, in order to keep it at a comprehensible size and as objective function-based criteria. For instance, we accepted CRMarcheo only after identifying real scenarios of its integrating power across and beyond individual excavation records.

If members agree, I propose to make an issue to document the evolution of this issue as didactic material for the CRM methodology. We never gave a reasonable written account of this kind for previous decisions, rather registered the following consensus only.

Please note, that I do not answer here these questions, a simply state that CRM-SIG must provide answers to them, or change the method.


All the best,


Post by Martin Doerr (19 November 2021)

Dear All,

As background material for future discussion, I add here a real example of animal tracking and how this would be represented in a CRM-compatible way by colleagues in charge in the National Greek research infrastructure Node "LifeWatch Greece" of LifeWatch Europe:

All the best,