SUMO
1022 December 2004 by Ian Davis
Danny points to Ontology Portal – The Suggested Upper Merged Ontology which has been updated. SUMO, CYC, WordNet and similar efforts are always kind of fun to explore but I’m always left with a feeling of disappointment.
For example, the SUMO page above has a search box with a deliberately provocative default search term: Buffalo. It waits there quietly whispering “come on, type something if you dare; I’m more than just prehensions and qualities and actualities, I know about buffalos and real things like that; come on and try me.” My eyes swept the desk looking for something obscure: CD, screen wipes, coffee mat, translucent flexible gap filler, Sony-Ericsson K500 docking cradle. In the end I stopped being vindictive and tried something pretty commonplace: Paper. It gave me some definitions from WordNet and some links from each definition to SUMO concepts. Clicking through to Artifact reveals some very interesting work including some inferences such as “artifacts are the result of making” and “plans express artifacts”. Looks pretty good. Then I noticed at the top of the page the definition of artifact: “A CorpuscularObject that is the product of a Making.”. Corpuscular! The only other time I’ve seen that word has been when reading about Isaac Newton. Clicking on CorpuscularObject was when the disappointment started to set in. Each page has a link to a graph of the sub/super class relationships for each concept. The graph for CorpuscularObject looks like this:
- SelfConnectedObject
- CorpuscularObject
- ContentBearingObject
- OrganicObject
- Artifact
- Nest
- CorpuscularObject
Can you spot the deliberate mistake? Why is nest expressed at this level? I did wonder that perhaps it was some meaning of the word that I wasn’t familiar with but no, a nest is “Any structure which is created by nonhuman Animals for the purpose of giving birth to their offspring.” Browsing SUMO turns up lots of these kind of inconsistencies.
Yes, I understand it’s not finished and it’s only “suggested”, however it strikes me that there’s a certain arrogance in what these grand, all encompassing ontologies set out to achieve. They seek to lock down and define all essential aspects of the world in a central and dictatorial manner. Yet the world doesn’t work like that. Meanings change from time to time, place to place, mind to mind. When I think of a nest, I rarely imagine sticks and twigs, instead I picture my son surrounded by beanbags, nestled down, reading a book. My humble advice is to keep it small. Start with something manageable and grow it just enough. Let different parts of it evolve at different rates just like language does. Let small pockets of expertise grow their own ontologies, then figure out how to work with them. Identify the common terms and use those as stepping stones between domains. Decentralise and diversify. The result will be a richer, relevant and more useful web of knowledge about the world.
Heh, well put. I entirely agree with your general conclusions, though do reckon these things have their uses. Things like Sowa’s top-level categories could be very handy when modelling real-world stuff (assuming you have fat dictionary at hand).The SUO & stuff like Cyc could potentially be very useful when trying to discover connections between things – the problem is that they are presented as if the were the only universal ontology. If you imagine them to be one of many (i.e. don’t take ‘em too seriously), then hooks into them could be useful alongside more local ontologies.It’s a pain that a lot of people seem to think the SemWeb approach is about a single, global vocab – as you suggest, the real strength here is in the flexibility and support for decentralisation and diversity.
I agree they could have their uses. But the scale of the task is such that the ontology will be ripe for obsoletion by the time it is completed.I think the real growth area in the coming years will be in ontology mediation – converting between and interfacing with diverse ontologies. The innovation of OWL was to make terms addressable thereby enabling authors to make equivilence statements with reference to other ontologies.
There are some other top-level ontologies, like DOLCE or BFO, which are much more consistent than SUMO (oh well, at least DOLCE is, I didn’t take a deeper look at BFO yet).The point is, that with such top-level ontologies you can enable (or at least, help with) automatic ontology merging – and there it becomes very interesting in the Semantic Web context.
I think that top-level ontologies will become important for such things as automatic ontology mapping and data integration. And there are some top-level-ontologies, where much more thought is put into than in SUMO, just take a look at DOLCE or BFO (although I admit not to have studied BFO yet myself).
Folks, I’m puzzled by this article, which states that there is an “inconsistency” in SUMO. That’s a fairly strong word, which in a logical system would normally be taken to mean “contradiction”. Of course, no contradiction exists in the example mentioned. Now, one can reasonably make a case that “Nest” does not belong directly under “CorpuscularArtifact”. Indeed, now that you’ve pointed this out, I’m inclined to put it under “Artifact” at least. But to dismiss all of SUMO by one example from the MId Level Ontology (MILO), is a bit of a strech. I’d be eager to know of any other definitions that you find problematic, but please lets not jump to conclusions of exaggerating either their influence on the entire enterprise, or even exaggerating the degree of concern about any particular infelicity.
Hi Adam,Thanks for taking the time to comment. I should have been more careful when writing “Browsing SUMO turns up lots of these kind of inconsistencies.”: I didn’t intend to imply a logical inconsistency. I could have better described them as being incongruous.Taking another look, I’m not sure that simply moving “nest” to be under “Artifact” actually solves the problem. Using the SUMO browser to explore Artifact a little suggests to me that Artifact is a subclass of IntentionalProcess which is “A Process that has a specific purpose for the CognitiveAgent who performs it.”. SUMO defines a CognitiveAgent as “A SentientAgent with responsibilities and the ability to reason, deliberate, make plans, etc. This is essentially the legal/ethical notion of a person.” There are some examples of creatures that might be CognitiveAgents, but it looks to me as if most nest-building creatures would be excluded. Perhaps there needs to be some distinction made between objects constructed through instinct and objects constructed through planning. I would then place nests, beaver lodges, bee hives, termite mounds and the like as instances of the former class. What do you think?Ian
Hi Ian, Ah yes, now I see why Nest is just a CorpuscularObject. My colleague Ian Niles created the concept, and it looks like he did indeed put it in the right place. So, do you agree, and do you still see the definition as incongruous? If so, why? The general point you mention about upper ontologies is important. Where would the “right hooks” come from for many little bottom up ontologies, if not from an upper ontology? There are many reasons for using an upper ontology, but you’ve mentioned one important one.Adam
Adam,It’s incongruous because its level of specificity is greater than the other three terms at that level of the ontology. Where would “chrysalis” be placed?One solution would be to introduce a new “ConstructedObject” term as a peer of “ContentBearingObject” and “OrganicObject”. Then “Artifact”, “Nest”, “Chrysalis” etc are made descendants of “ConstructedObject”.Ian
Ian, I think it would be good to explore what is the practical impact of having terms that subjectively are at different levels of specificity inherit from the same class. There’s no practical impact on the logical conclusions that can be drawn, as far as I can see. There is a potential impact on human understanding of the ontology, but I believe that is addressed in this case by having “Nest” exist in the MId-Level Ontology, rather than SUMO. Are there other negative impacts that you see?Adam
Ian, A few further thoughts on the specific concepts here. I’m not sure what the benefit of having ConstructedObject would be. Any creature would then properly be a ConstructedObject because of the biological process of reproduction. One problem that many ontologies have, I believe, is a result of not having formal rules that define the terms. The problem is a proliferation of classes that are mostly undefined, and unneccessary. If we allow constructed objects which are the result of any non-agentive process, then we admit every object into that class, since every physical object owes its existence to some process. We might then consider restricting this notion to be AnimalConstructedObject. That might be argued, but still we’d be left with a definition that was only that an animal made the object. That wouldn’t have much impact on the logical conclusions that might be drawn. I think one would be better off adding a rule to the concept of Nest that simply states that it was the result of a Process where the agent is an Animal. The class hierarchy is an important representational feature, but is overloaded by ontologies that are constructed in limited languages that have only class membership and relation typing at their disposal. Once one has available a more powerful logical language, a more detailed analysis and implementation is possible. So, I think the moral here is not to get hung up on looking at just the class structure of SUMO, since it’s not merely a taxonomy.