Actions

Ontolog Forum

Revision as of 16:05, 31 March 2018 by imported>KennethBaclawski
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

ToddSchneider: Does context exist independent of an agent?

thomas loertsch: pfps: https://s3.amazonaws.com/ontologforum/OntologySummit2018/OpenKnowledgeNetwork/OKN-Session-2--GaryBergCross_20180328.pdf

pfps: Thanks thomas.

pfps: Also, does anyone have any good information on what is happening in the OKN effort? I tried (a lot) after the last call, but didn't get anywhere.

pfps: In particular, there is the claim just now "OKN uses lightweight standard vocabulary-like tools like Schema.org and a general representation for web-available information like a knowledge graph.". Where is information on how this was determined?

pfps: Is this an official proposal to OKN? How does one send in a proposal? How does one find out the current proposals?

Gary Berg-Cross: The proposal seems to be to identify 4 things about an RDF fact - space, time, providence and confidence.

ToddSchneider: Vinh, on slide #4, what does 'probability' refer to?

thomas loertsch: Google finds nothing on "Triple++" or "Triples++".

Gary Berg-Cross: Some background to this - Vinh defines a singleton property as a property instance representing one specific relationship between two particular entities under one specific context. Singleton properties can be viewed as instances of generic properties whose extensions contain a set of entity pairs. In order to represent the uniqueness for a singleton property, we assign a unique URI for each singleton property.

thomas loertsch: Gary Berg-Cross: but having one ID per triple sounds more like named graphs to me.

Gary Berg-Cross: This approach uses some singleton property relations like has source and extractedOn.

ToddSchneider: Vinh, could you provide a definition of 'singleton property' as a binary relation?

RaviSharma: @vinh -In Heathcare as we are dealing with real objects or concepts, please describe what is the purpose of reification, is it to create databases or populating with real data?

John Sowa: Do we have time? We still have two more talks.

BobbinTeegarden: Allegrograph is actually a quintstore, and one of the 'extras' is an id for the triple.

Gary Berg-Cross: @John, the talk on Healthcare is basically 2 slides I am told...

John Sowa: Since Vinh has a second talk, could we postpone questions to the end of her second talk.

Patrick Stingley: I was unable to find documentation for Triple++ Is there a reference for this data structure?

Gary Berg-Cross: I was unable to find documentation for Triple++ Is there a reference for this data structure?

anonymous: I don't know if I would say "hack". I would say that named graphs were intentionally left with undefined semantics so that they could meet many different use cases.

DouglasRMiles: Can some links be posted in here?

DouglasRMiles: From Vinh's last few slides

RaviSharma: @Gary - kindly place the link.

John Sowa: I have to leave at one. I had hoped to hear more about Cyc and its current status.

DouglasRMiles: JS: re-re-scheduled for 1 :\

Gary Berg-Cross: Some links Online Community Discussion Forum

ckg-community@googlegroups.com

Join the group to receive the latest updates at https://groups.google.com/forum/#!forum/ckg-community/join

Gary Berg-Cross: NITRD Big Data Interagency Working Group: 3rd Workshop on an Open Knowledge Network (2017). https://www.nitrd.gov/nitrdgroups/index.php?title=Open_Knowledge_Network

2.OKN White paper: http://ichs.ucsf.edu/wp-content/uploads/2017/08/OKN-White-Paper..docx

RaviSharma: @Amit - what type of complexity in relationships for more complex entities?

RaviSharma: Notes: Amit Sheth Described evolving nature of complex relationships and will send relevant papers and examples if we send an email.

Vinh: @Todd: probability may refer to the certainty of the triple extracted from text

John Sowa: I have to leave.

Vinh: @Thomas: having one ID per triple does not have to be named graph

Vinh: the singleton property can play the role of the identifier within a triple

anonymous: One ID per triple is more like reification.

ToddSchneider: Vinh, Thank you (for the explanation about probability). Can you provide a mathematically flavored definition of 'singleton property' (e.g., as a binary relation)? thomas loertsch: Vinh, I find that singleton properties break something very essential: the shared vocabulary for predicates/links. The one argument for singleton properties that addressed a real issue was that they work within the triple model. But if you switch to quads anyway I see no reason for singleton properties anymore.

anonymous: thomas loertsch: You would want to associate provenance with the graphs? I.e. all triples with common provenance go into a single named graph and then you make triples with the name of the graph as object?

thomas loertsch: @anonymous: probably something like that, yes

RaviSharma: @amit Sheth - I tried on linked in as well as in your CV, your email is not accessible, kindly paste it here or send me a test email at drravisharma@gmail.com

Vinh: amit@knoesis.org

RaviSharma: @Vinh - thanks. please also make it available in his Bio or CV.

Vinh: @thomas: many datasets implementing and evaluating named graph, singleton property, reification, n-ary

MikeBennett: GSIB = Globally Systemically Important Bank (banks identified by G20 as requiring a specific new reporting regime from Basel called BCB239). In case you were wondering.

Vinh: @thomas: the ckg-community is going to discuss these approaches and how they address the practical issues in the community

Vinh: http://knoesis.org/amit

thomas loertsch: @vinh: thanks for providing the input! I wasn't aware of the ckg-community until today and just joined the googlegroup looking forward to discuss this more!

Vinh: @thomas: hope to see more discussions like this soon

AndrewDougherty: Beginner question, is assuming local transform correctness guarantees global transform correctness (if that assumption was being made) tantamount to assuming model-theoretic compactness? Simple yes or no will suffice.

AndrewDougherty: K ty!

RaviSharma: @Charles, for the heavyweight obvious from graph complexity, how do you filter relevant links?

DouglasRMiles: @Charles: I think OKN's data can be loaded into CycL Mts but not the other way around right?

Kalpa Gunaratna: one thought on modeling KGs that are easy to query and support question answering. We lose easiness of understanding triples when they are represented using n-ary, reification, and/or singleton property. That is the main reason why publicly available or industrial KGs have their own way of representing context information or do not represent them at all. It just make them hard to process in the applications. Good example is question answering. I think this problem needs more though put into it. May be we need to rethink RDF model or triple representation. We also have to think, are we going to patch RDF model and not going to address its short comings?

DouglasRMiles: @Andrew .. to followup on Charles answer.. you can see CycL proof nodes as each being "compact" when it comes to compactness

AndrewDougherty: @Charles: Are the prerequisites for understanding the formal verification explicitly defined in the Cyc KB? So for instance, would an ideal NLU algorithm be required to add concepts to the KB to understand this presentation and the QnA?

Vinh: @Kalpa, question answering with contexts is one of the main goals in the OKN visions

DouglasRMiles: (BTW when we say OKN in this context we are just meaning Triples)

RaviSharma: Notes: Charles described the fact that there is API etc., that allow you to study a particular entity and its relevant relationships, say for a company kerberos or its identity etc.

Vinh: the CKG community will gather and form the W3C interest group for addressing the representation of meta triples, making it easier for applications to use

DouglasRMiles: How can Cycorp get the rest of the world to use CycL ?

DouglasRMiles: (clarification to my last question.. right now the Semweb folks *are* the greatest needers of CycL .. how do we get them signed up?)

DouglasRMiles: I don't think they "don't want CycL"

Cory Casanave: The modal/contextual relationships described seem inherently higher-order, yet the logic is described as FOL. Are first-order limitations required?

anonymous: Why would there exist a single model of provenance that would be pragmatic for all uses?

Kalpa Gunaratna: @Vinh and others: QA is a nightmare when I think of processing RDF based triples. It is making commercial products or applications hard to get any intelligence. This is a must to give more thoughts on this. Not to just have yet another representation. They need to support better processing capabilities.

anonymous: No, I am saying that you are begging the question of whether there _is_ an appropriate global compromise.

Kalpa Gunaratna: I meant when context is integrated into triples.

Vinh: @Kalpa, that is the reason why the OKN initiative is started

Vinh: and the W3C working group may be the initial step towards that effort

pfps: Again, where can current information about OKN be found?

thomas loertsch: I think we just need a basic but flexible (or flexible but basic) mechanism to capture any sort of context so we can interchange any sort of context information, from provenance on the triple level to anything else. If one actually needs that kind of provenance granularity then you'd better optimize for it by adding another column to your triple/quad/forever store. But that shouldn't be reflected by the basic formalization.

Gary Berg-Cross: @PfPS again you probably know of NITRD Big Data Interagency Working Group: 3rd Workshop on an Open Knowledge Network (2017). https://www.nitrd.gov/nitrdgroups/index.php?title=Open_Knowledge_Network

anonymous: thomas loertsch: +1

DouglasRMiles: We need to get the W3C signed up to uses contextual logic

Vinh: @Douglas: +1

Cory Casanave: RDF: Can context (including time) be represented with fine-grain RDF graphs, the quad store?

anonymous: Is that an open meeting?

anonymous: How does one register?

Vinh: @pfps: I will pass the ckg-community@ to other OKN organizers

Vinh: so that the updates will be sent to this group

EvanWallace: There was some discussion on the need for contextualizing Sem Web content by time or other contexts, in one of the breakouts at the Sem Tech Symposium in Dayton.

anonymous: Never mind, I guess.

EvanWallace: A practical solution to this would really improve the utility of Sem Web content.

AndrewDougherty: A few questions: can you please link to the recording? What symposium is being linked to? ++Question about OpenCyc

pfps: Yeah, I know of the three workshops, but getting information on how OKN is progressing beyond them appears to be impossible.

pfps: And there does not appear to be any way to participate in the development or discussion of OKN.

anonymous: pfps: Exactly.

Vinh: for now, it is up to the community to pick it up

thomas loertsch: Where are the video recordings posted?

thomas loertsch: Ah, never mind, just found out myself...

anonymous: Is NITRD managing the OKN effort? NSF? Certain individuals? If the last, is there a list of them, or contact info? Somewhere?

Vinh: for the development and discussion of OKN, please join the ckg-community@googlegroups

Vinh: no, they don't (I got this confirmed by Chaitan Baru)

Vinh: https://groups.google.com/forum/#!forum/ckg-community/join

DouglasRMiles: I am going to complain a bit right now...

pfps: But how can the community get together to push OKN if there is no way to have discussions with the people who should be leading?

DouglasRMiles: Vinh / OKNs research is exactly the kind of research that needs to have RDF/OWL not impose any impedance. The value and beauty of CycL was that it sought to remove any such impedance to the logicians.

pfps: How will joining ckg-community@googlegroups help connect to OKN? Is that an official (or even well-known) part of OKN?

anonymous: DouglasRMiles: Cyc is lower impedance than W3C's suite?

Vinh: we will organize further discussions

anonymous: To whom?

thomas loertsch: @pfps: it seems that's exactly what Vinh is saying

DouglasRMiles: @anonymous: well lets say we have Engine Implmentors, Graph Curators and Logicians

DouglasRMiles: RDF/OWL was designed to make engine implementable in 200 lines of code

Vinh: @peter: we would connect this group with the OKN leaders

DouglasRMiles: at the expense of everything else

anonymous: DouglasRMiles: No, let's not. We don't have any of those roles at any of the institutions at which I have ever worked. We have people with roles that other disciplines recognize. How is a librarian, familiar with Dublin Core and Schema.org, going to fare with Cyc? How is a web developer who understands Drupal or the like going to enjoy Cyc?

DouglasRMiles: CycL engine takes 2 man years to create (Is that about right Charles?)

DouglasRMiles: ok so lets use the usual role name: Programmer, User

DouglasRMiles: @anonymous RDF/OWL programmer works for 200 minutes then he/she (the programmer) spends the next 2 years explaining RDF/OWL. User twists Programmers arm during such time programmer cries in pain and blames W3C.

anonymous: So if we stack the vaporware CycL engine you propose on top of the non-existent database, we're already ahead of the game?

anonymous: Except we have to make the rest of the world agree.

DouglasRMiles: ah.. CycL is not vapourware.. but you point out a real horrible problem here

pfps: The last information I have on the leaders of OKN is that there are three, Chaitan Baru, RV Guha, and Andrew Moore. See ichs.ucsf.edu/wp-content/uploads/2017/08/OKN-Next-Steps-1.docx

DouglasRMiles: Cyc/CycL is so incredibly well designed and written cycorp has incentive only to keep the world in the dark

anonymous: Yyyyeah. That must be the reason.

DouglasRMiles: Projects like OpenCyc, LarkC where govts only paying Cycorp money if they share. So unless there is enough pressure form the world (and money to pay) cycorp cant really spend all of its time trying to give away pearls. Also they cant afford when there are worthy takers to teach how to use cyc,

Vinh: @pfps: I am in touch with Guha and Chaitan

DouglasRMiles: @anonymous many things taught about Cyc are the things people are interesting and not about what Cyc was really

Vinh: @pfps: Guha served on my dissertation committee on the CKG

DouglasRMiles: @anonymous or at least not what some people believe to be cyc's value.. for instance cyc's value to me is the performance it has over arbitrary logical deduction and how Microtheories are partitioned

DouglasRMiles: So my value in Cyc is how it give people like OKN ideal stack (thus the value to cyc for me is not about KB of cyc) so what i meant in the RDF/OWL vs CycL for users/programmers

pfps: @Vinh: how would you connect to the leaders? Do you have some private link to them?

Vinh: nope, by email

anonymous: Chaitan's email is cbaru at nsf dot gov

pfps: @Vinh: I've been in touch with Chaitan, but I have been unable to get information on how OKN is progressing.

anonymous: I have Yolanda Gil's somewhere

DouglasRMiles: is i and (at least 5 others i know that have never worked at cycorp) is that we can write CycL because we can write modal propositional logics.. and only have to go back and recorrect our logical errors 2 to 3 times to get it correct

Vinh: @peter: I asked him that question, he said it is up to the community to pick it up and move forward, that's why we created this community

DouglasRMiles: yet with RDF/OWL we cant even get it on paper

anonymous: Yeah, that sounds like a really appealing feature for someone who wants to integrate data from Facebook with census data.

Vinh: @peter: he's in the mailing list as well, so he will update us if he has any

anonymous: Well, good luck. CycL is certainly intellectually deep.

DouglasRMiles: with RDF/OWL i don't know how I'd even begin to integrate data from Facebook with census data.. despite that being my dayjob for 2 years! yet in CycL takes me exactly a week. i mean my dayjob was to make company software OWL/RDF driven. i finally left after feeling like i just embezzled 2 years salary. yet with CycL.. i knew exactly how to make the company's data work, and left so i could do that without the guilt of taking their money anymore. oops he/she left but i still feel like an answer is deserved. So for example per context the company had.. I almost always contain that view from a DataMicrotheory and ensure i am ontological correct by creating a VocabularyMicrotheory. (i have no such guidance in OWL/RDF) Then I try to ontologize what the company "wants" to actually be their vocabulary. And that to becomes VocabularyMicrotheory .. but not i am on my way to writing some rule to create transforms of that data. i mean i try to see if i can at least have a pretension of semantics between rules that relate the datamicrotheory to a potential ApplicationContext. i don't know how much of this process i made up or how much i read.

pfps: @Vinh: but the ckg resources are for promoting ckg to OKN, what if I don't believe that ckg is a good idea for OKN?

DouglasRMiles: we need Charles Klein to find us doc on how Mts should be set up. (i know i haven't read about Mts .. i just tried to figure out what i was looking at and doing)

Vinh: @peter: we had a meeting among the OKN organizers (with Guha, Yolanda, Chaitan, etc.), we discussed the implementation of a prototype

Vinh: and in the future it will be improved

Vinh: therefore, I signed up for this prototyping during the meeting

AndrewDougherty: Hi, I still can't seem to find the link to the recording, I would appreciate it if someone could paste it here, thanks!

Vinh: I will get the OKN community discussion and feedback for further improvements

Vinh: @Andrew: the link will be available later I think

AndrewDougherty: Vinh: Thank you!

DouglasRMiles joined the Contextualized Knowledge Graph Community Discussion Forum if CKG is going to import YAGO-SUMO .. I suppose it expects to be as expressive as CycL (at least needs to be) my hope is that CKG will be able to handle SUMO https://github.com/ontologyportal/sumo/blob/master/Mid-level-ontology.kif i don't mean sizewise but semantically with its time contextualizations and such

(=>
  (and
     (successorAttribute ?ATTR1 ?ATTR2)
     (holdsDuring ?TIME1 (property ?ENTITY ?ATTR2)))
  (exists (?TIME2)
     (and
        (temporalPart ?TIME2 (PastFn ?TIME1))
        (holdsDuring ?TIME2 (property ?ENTITY ?ATTR1)))))

DavidJWhitten: How's your LogicMOO/CommonLogic in Prolog doing?

DouglasRMiles: It's doing good, I am trying to come up with the UI that the assertions can be edited. It is an age old problem. Facet Editor that pops up salient properties. +? or a KE Textarea populated with Microtheory contents?

DavidJWhitten: Yeah. For a KB it is particularly problematic as some people want to have separate predicates for different orders of arguments. MotherOf vs isMother etc.

DouglasRMiles: Should it look like Protege 4+ or like Protege 3.0

DavidJWhitten: I have really only used the Cyc Browser and one at Stanford over 10 years ago.

Vinh: @Douglas: I'm looking at the SUMO, will keep you updated!

DavidJWhitten: I think Cycorp added predicates for marking if a predicate was used for inference or for editing didn't they?

DouglasRMiles: Right, so from the Cyc Browser if you click a TestQuery.. it brings up the ideal page for editing the query parameters (and for starting and and debugging a query)

DavidJWhitten: Do you like the utility of the TestQuery, and want to change the presentation, or do you see it as lacking ?

DavidJWhitten: SUMO still uses attributes, but Cycorp has decided not to do so, right?

DavidJWhitten: Have you ever heard Cycorp explain why they philosophically aren't necessary ?

DouglasRMiles: So TestQuery isn't lacking .. but that is only one of 100+ idiomatic editing facilities that are needed Well attributes can be encapsulated in isa hierarchies (is what i understood) i think if attributes normally being temporally fluent whereas isa's being durations normally for the time the object exists in other words if something is transitioning between having a main color or red to blue.. I'd think that be be an attribute rather than a (isa xyz123 BRedColoredThing) (isa xyz123 BlueColoredThing) (isa xyz123 RedColoredThing) to a (isa xyz123 BlueColoredThing) I understand it make cycorp code slightly simpler since they already had to code such a transformation anyways and semantically it is still even legal (isa xyz123 (CollectionWithKappaFn (?Inst) (attribute ?Inst BlueColor)) well probably more like: (isa xyz123 (CollectionWithKappaFn (?Inst) (mainColorOfObject ?Inst BlueColor))) Why i like Cyc is that there is no one representation I *have* to use I mean we can reintroduce attribute/2 and expect it to work in the same way (equal M (MotherOf C)) is equivalent to (isMother C M)) having lots of ways to say the same things is such a good thing (though it does of course in the sense of some systems create lots of utility rules) (though it does of course in the inspire creation of lots of utility rules ..and makes the engine a bit more code) but at least reduces the bottle necks of users trying to figure out how to prematurely optimize for SPARQL i be just as frustrated if Cycorp, Logicmoo, W3C or OKN made me trying to figure out how to change or redesign my semantics rather than letting me use the semantics i designed since it is hard enough for me to cognitively design the logical interactions for an application anyways So for example in Logicmoo.. I designed it to where every time a person drinks water they are a little less thirsty representing the frame that water gets drank has a few properties.. since as the time it occurs.. the drinker.. the water source etc. well i should say most of the time i am representing fluent and situation calculus over frames not the entire frame description is required in preconditions i need to transition between frame-speak and ontological speak in at least in at least a comfortable manner

DavidJWhitten: makes sense.

DouglasRMiles: So UI-wise i suppose i keep re-deciding richly marked up KE text-like is going to be the easiest for people but of course I have to have tooltips and autocompletion Seeing if i can use SWANK/SLIME over a web repl top provide those tips and autocompletes for the CommonLogic and also of course make what is on the page (what user just typed or was already on the page) clickable SWOOP did the best in this case with 13-Figure9-1.png where Concise format can also just plain be typed on top of opops have to run for a few

pfps: @vinh: is there information on this meeting with the OKN leaders/

Vinh: @peter: let me check if there is any note taken