From OntologPSMW

(Difference between revisions)
Jump to: navigation, search
(Discussion)
(Attendees)
 
Line 85: Line 85:
 
* [[TomTinsley|Tom Tinsley]]
 
* [[TomTinsley|Tom Tinsley]]
 
* [[VickyHailey|Vicky Hailey]]
 
* [[VickyHailey|Vicky Hailey]]
 +
* Ƴ
  
 
== Discussion ==
 
== Discussion ==

Latest revision as of 17:56, 4 June 2020

[ ]
    (1)
Session Standards for Knowledge Graphs
Duration 1.5 hour
Date/Time 03 June 2020 16:00 GMT
9:00am PDT/12:00pm EDT
5:00pm BST/6:00pm CEST
Convener Ram D. Sriram and Ken Baclawski
Track What

Contents

[edit] Ontology Summit 2020 Standards for Knowledge Graphs     (2)

Knowledge graphs, closely related to ontologies and semantic networks, have emerged in the last few years to be an important semantic technology and research area. As structured representations of semantic knowledge that are stored in a graph, KGs are lightweight versions of semantic networks that scale to massive datasets such as the entire World Wide Web. Industry has devoted a great deal of effort to the development of knowledge graphs, and they are now critical to the functions of intelligent virtual assistants such as Siri and Alexa. Some of the research communities where KGs are relevant are Ontologies, Big Data, Linked Data, Open Knowledge Network, Artificial Intelligence, Deep Learning, and many others.     (2A)

We have witnessed considerable activity in the development of knowledge graphs and networks. In order for these advancements to be widely used, we will need standards to ensure performance, conformity, and safety in applications. Further, standards play a very important role in innovation. Our panel will discuss various questions related to standards and knowledge graphs/networks – Why do we need standards? What is the state of the art? What are the future directions? What should various organizations be doing?     (2B1)

[edit] Conference Call Information     (2C)

[edit] Attendees     (2D)

[edit] Discussion     (2E)

[12:00] Ravi Sharma: Ken are we expecting Ram?     (2E1)

[12:02] Lisa Carnahan: Ram is at a Fellowship ceremony from the Solid Modeling Association.     (2E2)

[12:06] Ravi Sharma: Lisa, Barry and Micheal - welcome     (2E3)

[12:12] Ravi Sharma: Lisa - A small set of representatives from domain design and develop standards, but is adoption driven by industry need or by agency push or both?     (2E4)

[12:13] Todd Schneider: What are any existing standards that may be applicable to/for knowledge graphs?     (2E5)

[12:18] Ravi Sharma: Lisa - there is one more perspective, where do KGs or Ontologies fit in the standards canvas across board (of standards)?     (2E6)

[12:29] Ravi Sharma: Barry - Hub would also have to include basic entities and relations of every domain that is represented in Spokes?     (2E7)

[12:32] Michael Gruninger: @Barry - the creation of one ontology for the Web was not the raison d'etre for ISO 21838. The role of ISO 21838 is to standardize the criteria for what constitutes a top level ontology and to standardize the axiomatizations of existing top level ontologies that satisfy these criteria     (2E8)

[12:33] Ravi Sharma: Barry - The general attributes are like domains, and we say we do not want domain entities?     (2E9)

[12:35] Mike Bennett: Is the TLO definition about 'Things' or 'Data about'? These are not the same thing.     (2E10)

[12:37] Vicky Hailey: I thought the ontology community rejected the ISO standard?     (2E11)

[12:39] Mike Bennett: If the TLO definition is intended to cover data, then there is potentially data for any concept as conceptualized by some entity or organization, not only things that exist in reality. For example events and situations to avoid.     (2E12)

[12:41] David Eddy: Does LOD do anything with operational, commercial behind-the-firewall systems?     (2E13)

[12:44] Janet Singer: Are the two spatial metaphors of Top vs lower level and Hub and spoke ontologies both used in ISO 21838?     (2E14)

[12:44] Mike Bennett: Needless to say data may also represent falsehoods, works of fiction, mistakes, mischaracterizations and so on.     (2E15)

[12:48] Alex Shkotin: @Barry, how to get slides?     (2E16)

[12:48] Bobbin Teegarden: @Barry space (where) and time (when) are two of the aspects of a context. how do you relate those to the other aspects of context: who what where when why how...?     (2E17)

[12:48] Ravi Sharma: Barry - How is spatio-temporal aspect compatible with OMG standards especially space and time , calendars and SBVR?     (2E18)

[12:51] Janet Singer: @Vicky see Michael Grueninger's response above. There does seem to be lack of clarity re that     (2E19)

[12:54] Ravi Sharma: Michael - what is your view on TLO, should it contain top level entities of the domains that they connect to or just metadata like attributes defined in BFO?     (2E20)

[12:57] Mike Bennett: @Michael what about different ontologies that are consistent but that represent different, incompatible conceptualizations of the problem domain?     (2E21)

[12:57] Douglas R Miles: Queued up question for Michael: What is your preferred axiom editor program?     (2E22)

[12:58] Douglas R Miles: (Also interested in my question being answered by others)     (2E23)

[12:59] David Eddy: How does one automate the extraction, discovery of ontologies from existing systems?     (2E24)

[12:59] Mike Bennett: +1 for 'semantic requirements' - I still see people taking a bunch of words and putting axiomatizations around them.     (2E25)

[13:00] Douglas R Miles: Do we have a knowledge engineering environment (KEE) that works with the standards?     (2E26)

[13:01] Ravi Sharma: Micheal - Is there a standard for semantics? if ontology standard is going to depend on semantics?     (2E27)

[13:04] David Eddy: see: "Philadelphia's Philosopher Mechanics: A History of the Franklin Institute 1824 - 1865" by Bruce Sinclair for a story of what it takes to actually adopt effective standards     (2E28)

[13:04] Michael Gruninger: @Mike Bennett: as a community, we need to do a better job at identifying the different possible ontological choices in a conceptualization, so that we can distinguish superficial incompatibilities from deeper ontological incompatibilities     (2E29)

[13:06] Mike Bennett: @Michael I heartily agree. I wish FIBO had been more motivated in this direction.     (2E30)

[13:06] Janet Singer: @Mike, @Michael Should it be semantic requirements? Or pragmatic requirements that a given ontology allows an agent to make the distinctions in a domain that would lead an observer to judge their behavior/actions as intelligent (Mark Musen used that criterion in the CS520 KG seminar yesterday)     (2E31)

[13:07] Douglas R Miles: Since Barry: is answering questions.. Does he have a preferred knowledge engineering environment (KEE) that allows him to make axioms and ontologies work together?     (2E32)

[13:07] Mike Bennett: Perhaps IAOA ISTC can publish a set of evaluation criteria that potential industry standard ontologies can document conformance to.     (2E33)

[13:07] Michael Gruninger: @Douglas R Miles: we definitely need better environments for editing axioms. In particular, we need an integrated development environment for ontologies in different languages.     (2E34)

[13:07] Pete Rivett: a (live) rabbit is not an "object"     (2E35)

[13:07] Douglas R Miles: (also that was my same question to @Michael)     (2E36)

[13:08] Douglas R Miles: Is there anything as good as Protege Pre-3.4 ?     (2E37)

[13:08] Cory Casanave: Question for the presenters: When using an upper ontology it is common to find commitments incompatible with the domain needs; is there ways we can improve the layering of commitments to allow for improved interoperability while retaining the advantages of a hub?     (2E38)

[13:08] Mike Bennett: @Pete I think they mean object other than in the UML / OO sense of that word.     (2E39)

[13:08] Douglas R Miles: (3.4 or earlier)     (2E40)

[13:10] Bobbin Teegarden: @Barry I think what you're saying is that the hub and spoke model is tree shaped (more like a taxonomy) and that work has to be done to create the relationships across the tree to get to the graph shape of an ontology. BUT in my experience, it is very difficult to get from the tree to a real a representative graph. Was that the intent of your process?     (2E41)

[13:11] Michael Gruninger: @Cory Casanave: In some conceptualizations, we can characterize the possible choices. For example, for time, there are three possibilities -- timepoints, time intervals, and a combination of timepoints and intervals. For other conceptualizations, we do not have such a complete characterization     (2E42)

[13:11] Mike Bennett: By 'Army' I presume Barry means the US Army (your standard reminder that the US is just one country and ISO is an international standards org)     (2E43)

[13:11] Douglas R Miles: Yes, I am trying to find a IDE for mixing axioms and ontologies.. one that can have a analysis tool plugged in     (2E44)

[13:13] PeteRivett: no @Mike Bennett I meant the regular meaning of "object" as in an inanimate thing     (2E45)

[13:13] Ravi Sharma: The summits I have attended since 2008 and the Forum exists longer, thanks to Peter Yim, yet we are yet not able to integrate or interoperate different ontologies using a common standard? I know they can interoperate but not necessarily using a standard?     (2E46)

[13:14] Mike Bennett: @Pete then what would an animate thing be?     (2E47)

[13:16] PeteRivett: an agent?     (2E48)

[13:16] Mike Bennett: If a rabbit is an agent the who is the patient?     (2E49)

[13:16] Evan Wallace: @Pete: In BFO, an Object is essentially any discrete material entity, so in BFO a Rabbit would be an Object.     (2E50)

[13:18] Pete Rivett: a great illustration of why a single TLO is a bad idea IMO     (2E51)

[13:19] Mike Bennett: @Pete your questions are a good illustration of why you need a TLO.     (2E52)

[13:19] Evan Wallace: OBI which is a spoke of BFO has the notion organism which appears to be a sibling of Object. Note sure how the OBO folk would sort out where Rabbit should be attached, given this.     (2E53)

[13:19] Douglas R Miles: Colore is a great example     (2E54)

[13:19] Mike Bennett: Oops     (2E55)

[13:19] Paul Tyson: Do physicists recognize "discrete material entities" (objects)? I thought they only recognized statistical measurements of particle/energy density? Discrete implies a hard border of some sort.     (2E56)

[13:20] Damion Dooley: I'll ask why organism isn't under object. I sit on the OBI curation call so can fetch an answer pretty quick on that one!     (2E57)

[13:21] Douglas R Miles: TimeInterval you'd think should be a subclass of Event     (2E58)

[13:21] Douglas R Miles: (but sure, they might need to be fully disjoint)     (2E59)

[13:22] Janet Singer: Wouldn't validation always be against competency requirements (why) while other requirements would be a matter of verification (how)?     (2E60)

[13:23] David Eddy: "No"... excellent     (2E61)

[13:23] Mike Bennett: Will there be tome for questions other than from Ravi?     (2E62)

[13:24] David Eddy: I knew I was in trouble when the JPL folks said: "Its not rocket science..."     (2E63)

[13:24] Janet Singer: Benchmarks for ontology validation do seem to be the key standard needed     (2E64)

[13:24] Evan Wallace: @Paul Tyson - I was using my own words to describe the Object notion. The BFO 2020 elucidation reads: an object is a material entity which manifests causal unity & is of a type, instances of which, are maximal relative to the sort of causal unity manifested.     (2E65)

[13:25] Douglas R Miles: "Colore is a great example" = of where the units that are ontologically under some ontological Collections should beat least _able_ to fall under someones ontology     (2E66)

[13:25] Ravi Sharma: Paul - yes they do, at least for large matter entities, but at Quantum fields levels they do obey FD or BE statistics and also challenges in absolute localization.     (2E67)

[13:25] Douglas R Miles: be at*     (2E68)

[13:27] Ravi Sharma: Ken - thanks for asking Qs while zoom was down.     (2E69)

[13:27] Michael Gruninger: @Janet Singer: verification considers whether the ontology can answer the competency questions. Validation considers whether we have the "right" competency questions     (2E70)

[13:30] Todd Schneider: Cory, possibly via reference ontologies.     (2E71)

[13:30] Douglas R Miles: "right" competency questions = can be a set of test questions for a Inference engine?     (2E72)

[13:31] Janet Singer: @Michael yes, and there is no point in ensuring something does the thing right if you haven't ensured it will do the right thing (meet the purpose/need). Engineering often loses sight of that     (2E73)

[13:32] Douglas R Miles: "Is a Loan a set of assertions around a Situation"?     (2E74)

[13:32] Todd Schneider: A reference ontology makes some ontological commitments but is 'incomplete', in that additional constraints (aka commitments) can be added.     (2E75)

[13:32] Ravi Sharma: Michael - thanks for answering.     (2E76)

[13:33] Mike Bennett: A loan is an exchange of commitments.     (2E77)

[13:34] Ravi Sharma: Loan is a contract between two entities to commit to future transactions?     (2E78)

[13:34] Mike Bennett: (that is, any transaction is an exchange of commitments; a loan is a kind of transaction in which one commitment is discharged right away, the other is a commitment to a set of interest payments and principal repayments)     (2E79)

[13:35] Ravi Sharma: Mike - yes there could be multiple not only two!     (2E80)

[13:35] Todd Schneider: An ontology represents a collection of distinctions.     (2E81)

[13:35] Janet Singer: How much discussion about ontologies is about ensuring that needs are met, rather than assuming that ontologies are of course worth developing and then getting lost in the weeds of languages, representations, logics, worldviews, etc     (2E82)

[13:35] Douglas R Miles: Is there a tool that lets us use BFO inside an IDE?     (2E83)

[13:35] Bobbin Teegarden: @MG Isn't a 'domain' a context (with a point of view) and thus relationships and the configuration of relationships are relative to the context of that domain...?     (2E84)

[13:35] Mike Bennett: So using REA as a mid-level ontology, framed within a TLO, lets you frame all things that are kinds of agreement, kinds of transaction etc., in a single unambiguous framework.     (2E85)

[13:35] Ravi Sharma: Mike - yes better explained.     (2E86)

[13:36] David Eddy: @Douglas... as you start crossing boundaries, things go wonky.     (2E87)

[13:37] David Eddy: Industry is just about beginning to be able to lift their head (infant)... dreams of driving a Ferrari, are decades away.     (2E88)

[13:38] Janet Singer: Todd Yes, and distinctions that an agent needs to make to operate in a domain helps operationalize ontology evaluation     (2E89)

[13:39] Ravi Sharma: Ken thanks for a good session     (2E90)

[13:39] David Eddy: example... we "lost" a software request to Army in 1987... we had Assembler routines... AI was required. A working prototype was never achieved.     (2E91)

[13:39] David Eddy: working code was not as good as AI     (2E92)

[13:40] Ravi Sharma: Lisa - thanks     (2E94)

[13:41] Mike Bennett: Words are way more contextual than concepts.     (2E95)

[13:42] Bobbin Teegarden: @MikeB Yes!     (2E96)

[13:42] Leia Dickerson: 100% agree with Janet     (2E97)

[13:43] Gary Berg-Cross: One would hope that the competency questions about "male" and "female" would expose this problem of clothing products.     (2E98)

[13:44] Douglas R Miles: (ClothingTypeByAttributeFn Male) (ClothingTypeByAttributeFn Blue)     (2E99)

[13:45] Douglas R Miles: (I like to think that context is a FooFn)     (2E100)

[13:45] Ravi Sharma: On the other hand the TLO will be general enough that Any Hub type can connect to it? What value can we derive from it, except notional interconnectivity of a thing to everything?     (2E101)

[13:46] Gary Berg-Cross: Adversarial questions is needed as part of KE and conceptual analysis.     (2E102)

[13:46] Mike Bennett: @Gary great insight. Need a methodology.     (2E103)

[13:47] Douglas R Miles: as in: (instance (ClothingTypeByAttributeFn "Male") ClothingClassification)     (2E104)

[13:48] Gary Berg-Cross: KEs should think like Socrates at times.     (2E105)

[13:48] Mike Bennett: Yes, got to have Socratic dialog.     (2E106)

[13:48] Douglas R Miles: Argumentation Logic     (2E107)

[13:50] Bobbin Teegarden: Maps across contextual ontologies work, not sure how you get away from then to move forward without them (and not just start over)...     (2E108)

[13:51] Gary Berg-Cross: Methodologies for the hub the spokes and domains...and their connections.     (2E109)

[13:53] Douglas R Miles: In the spirit of John Sowa, @Barry can the CYC ontology fit under BFO and not be lossy ?     (2E110)

[13:54] Douglas R Miles: CYC UpperLevel*     (2E111)

[13:56] Douglas R Miles: (someone else can ask Barry)     (2E112)

[13:57] Bobbin Teegarden: The key difference between a taxonomy and an ontology is the shape: taxonomy's are tree shaped (generally) and ontologies are graph shaped. A generalization, I know, but it is hard to get cleanly from a tree to a graph...     (2E113)

[13:58] Mike Bennett: @Bobbin that is a good description of one of the 4 concepts I described for which the word Taxonomy may be applied.     (2E114)

[13:59] Ravi Sharma: The ability of Search using KGs assumes Logic (reasoning) and therefore Ontology?     (2E115)

[14:01] Gary Berg-Cross: KGs may readily adopt the male female product concept while the problems with that commitment are more likely to be seen as a problem as part of a rationalizes system of organized knowledge that KE and ontological engineering produces.     (2E116)

[14:03] Bobbin Teegarden: But aren't upper ontologies to be used as upper structures, but not drill into deeper domain details?     (2E117)

[14:04] Ravi Sharma: Ram mentioned QEDT for BFO     (2E118)

[14:09] Ravi Sharma: Ravi asked Q on KG whether one needs logic (reasoning) only namely ontology only or also needs AI XAI or ML for getting meaningful response from KGs?     (2E119)

[14:10] Ravi Sharma: Cory - nice to see you these days, thanks.     (2E120)

[edit] Resources     (2F)

[edit] Previous Meetings     (2G)


[edit] Next Meetings     (2H)