the foundation ontology as a basis for semantic interoperability patrick cassidy

1

The Foundation Ontology as a Basis for Semantic Interoperability

Patrick CassidyMICRA, Inc., Plainfield, NJ

[email protected]

Version modified from the OIC-2008 presentation

Outline • Accurate general Semantic Interoperability requires that different systems

represent their knowledge using a common Foundation Ontology (FO). The common Foundation Ontology provides a common set of concept representations that can be used to logically describe the intended meanings of any of the more complex concepts not in the FO.

• Basing the common FO on the full set of fundamental conceptual elements (sometimes called “semantic primitives”) provides a means to limit the need for agreement; viewing the FO as a means to translate among different representations allows complete freedom for locally optimal representation.

• A common foundation ontology that can be widely adopted should be as small as possible, to maximize agreement

• To make usage easier, utilities are necessary; a good Natural Language interface may be required

• A collaborative project to develop an FO according to these principles, by a consortium of about 100 participants, is feasible and limited only by the need for funding of the participants

2

The Problem of Independent Applications • Different groups develop their own databases, terminologies, and ontologies. Local

communities want to do their own thing, not be forced to conform.• When there is a need to communicate information among independently developed

databases or applications, for automatic use (without human intervention) information needs to be communicated and interpreted accurately – i.e., systems need to interoperate at the semantic level.

• The ability to create special data structures without reference to a common vocabulary standard for information transfer within a community having specific needs cannot scale to the general situation where information must be in a form interpretable by any other system that can use it.

• The absence of a common vocabulary for data elements makes accurate interoperation impossible.

• Semantic Interoperation requires a common standard of meaning; if any undefined terms are used in the description of a term or concept, accurate interpretation is impossible. Only an agreed basic vocabulary will support accurate interpretation within a community. The FO supplies the basic “vocabulary” for logical representation of many domains, allowing translation of any domain ontology into any other.

3

A Language is More Than a Grammar• The term “language” has been used to mean qualitatively different things,

such as computer “languages” like FORTRAN, C, or JAVA. • A language in the sense relevant to the interoperability problem must have

both a grammar and a vocabulary.• Ontology “languages” such as OWL or CL consist primarily of a grammar,

with a minimal semantics defining the logical operations.• Regardless of how widely any grammar such as OWL is used, accurate

communication will be impossible without some agreement on a vocabulary.• The FO tactic makes that agreement easier by requiring agreement only on

the basic concepts used to describe all others, leaving application developers free to invent an unlimited number of terms and concepts for local use without any contact with other groups, and still have their information accurately interpreted by any other system using the basic vocabulary.

• The goal of the FO project is to find agreement on that limited vocabulary of basic ontology elements, and demonstrate that this is adequate to support general interoperability.

4

A Solution for The Problem• Locally developed applications can use small, specialized ontologies,

idiosyncratic ontologies, or no ontology at all and still perform their work perfectly, and share information using local agreements for the meaning of the data.

• BUT When local applications need to share complex information with many other systems, a common, expressive standard of meaning (i.e. a common language) is essential for communication.

• The Solution – a common Foundation Ontology to provide a standard for Content to complement the existing standards for Format and basic-level Semantics (such as OWL or FOL).

• There is a widespread assumption that getting some broad agreement on a common Foundation Ontology is impossible. This assumption is largely based on the ambiguity of language, and a simplified view of the function of the Foundation Ontology. There is no technical, social, or psychological barrier – what has been missing is a proper interpretation of the function of the FO, and adequate funding.

5

Overview of the FO Project(1) The goal is to find a means to translate assertions in one ontology language (grammar

+ vocabulary) into another ontology language. The translations will use axioms having elements in common between the two ontologies.

(2) there are some ontology elements whose intended meanings cannot be expressed solely as an FOL combination of other ontology elements. These are called the "primitives“ in this discussion.

(3) for any given group of domain ontologies, there will necessarily be some set of such primitives that will be sufficient to logically specify by FOL combination the intended meanings of the non-primitive elements of all of the other ontologies in the study. These constructed meanings will not necessarily be complete descriptions of the intended real-world referents; they will be sufficient to perform the computations desired for the applications supported by the domain ontologies.

(4) To *accurately* translate logical assertions among those domain ontologies, the most parsimonious tactic (and probably the fastest) would be to identify the primitives in common among those domain ontologies, include them in an FO, and use them to create translations of assertions between the domain ontologies. Those translations will use "bridging axioms" to convert assertions from the form in one ontology to the form in another ontology.

6

Overview of the FO Project (2)(5) To minimize the changes in the FO as new domain ontologies are linked to

(mapped to or logically expressed by) the FO, it is advisable to try to identify as many of the possible primitives as can be identified, at the earliest stages of testing of the FO. This should reduce the number of new primitives that need to be created as new domain ontologies are linked to the FO. Since the test has never been done, we do not know whether or how quickly the need for new primitives will drops for each new domain. That can only be determined by testing the FO process. It is possible that new primitives will need to be continually added; even so, this method promises to be the most effective to achieve the maximum and most accurate semantic interoperability that is possible at any given time.

7

Overview of the FO Project (3)(6) As possible inventories of primitives that should be included in a *starting* FO, to

aim for the broadest coverage as quickly as possible, I suggest using the senses associated with the Longman dictionary defining vocabulary - 2148 words, and probably over 4000 senses. Longman has been tested for its ability to linguistically define all other words in the dictionary, but whether there could be a similar small inventory of primitive ontology elements that can combine to specify *all* other ontology elements is unknown and may be impossible. The more relevant question is whether a set of primitive ontology elements can be found that will not need *significant* supplementation as new domains are linked to the FO; if little supplementation is needed, the FO should be stable enough for most practical tasks requiring semantic interoperability. This question can only be answered by testing multiple domain ontologies versus some common FO.

(7) Other possible sources of essential primitives could be the 3000 most frequent Chinese characters (covering 98.9% of modern text) and the 4000 most common signs of AMESLAN. But these symbols have not been tested as a "defining vocabulary".

8

Overview of the FO Project (4)(8) The fastest method to test the FO hypothesis is to

create a consortium of multiple diverse groups of ontology developers and potential users (and relevant standards groups), to (1) conduct a carefully organized study to agree on a common FO containing all of the primitives necessary to translate assertions among their domain ontologies; (2) test the utility of the FP-linked ontologies in practical applications, and (3) test the ability of the FO to support accurate semantic interoperability among those applications.

9

The Principle of Semantic Primitives• For any given group of domain ontologies, it is possible to identify

some set of basic ontology elements (the Foundation Ontology, or FO) that can be combined to form the more complex ontology elements in the domain ontologies. Those basic ontology elements can be viewed as representing the “semantic primitives” for that group of ontologies. These primitive elements can be used to translate information from its form in any one of those ontologies to its form in any of the others. Any logical contradictions among the linked ontologies can be recognized and represented.

• As the number of ontologies linked increases, the number of new primitive elements required to link those ontologies will decrease. At some point the FO will be stable enough to serve as a reliable standard of meaning for accurate semantic interoperability.

10

What are the “intended meanings”?The intended meaning of an ontology element reflects

two criteria:• The meaning that the creator of the ontology element intends to

capture by the logical specification: ideally, this will be unambiguously described by the linguistic documentation as well as reflected in the logic.

• The behavior of the programs that use that ontology element must correctly reflect the behavior (insofar as it is affected by that ontology element) that the ontologist and programmer both intend for that program.– Any change in an ontology that affects the logical specification of an ontology

element must not change either the meaning as intended by the ontologist, or the behavior of the ontology-based application, unless that change is understood and accepted by the ontologist and programmer.

11

Mapping versus Translation

• Automated mapping without a common foundation ontology is too inaccurate for mission-critical automated decisions

• Semiautomated mapping without a common interlingua ontology is too expensive - order of n2 effort; however, mapping to a common ontology reduces the effort to integrate multiple ontologies, including those initially developed without reference to the common ontology.

• Domain ontologies developed from the start by using the common foundation Ontology to describe the domain terms will be automatically translatable into each other, with no need for post-hoc mapping to any other ontology.

12

Mapping and Translation• Whenever careful mappings are created between different

ontologies, the elements appearing in the mappings will be either primitive elements or elements than can be expressed as an FOL composition of primitives. Identifying the common elements in such mappings could provide one method for identifying the fundamental primitives that are required for describing the meanings of the domain elements. Those primitives can then be used for translation among many ontologies.

• One ontology mapping project (“COLORE”) may provide a source of candidates for inclusion in the FO.– See: http://ontolog.cim3.net/file/work/OOR-Ontolog-Panel/2009-02-

19_OOR-Development-II/colore-ontolog-oor--MichaelGruninger_20090219.pdf

13

The FO and COLORE

• The FO as described here could be the “given ontology” that is referenced in one of the COLORE use cases:

COLORE example use case:• retrieving ontologies with respect to particular

relationships e.g. all ontologies that are interpretable by a given ontology

14

15

The Translation Tactic:Everybody Gets Everything They

WantBy supporting Translation among different local

knowledge representations• Nobody has to stop doing anything they want to,

they can do it exactly the way they want to do it• When applications need to communicate, the

developers only need to learn the common defining language (or collaborate with someone who already knows it) and map to it

• Learning and using the common language of the foundation ontology is time-consuming, but can be made easier by utility programs – commercial and open-source, and a Natural Language interface.

Alternatives to a Common Foundation Ontology? Mapping post-hoc vs. ab initio

• The relations between types in two different ontologies may be: – Synonymy (same intended meaning)– Specialization (one type may be a subtype of the

other)• The added constraint(s) by which the specialized

type differs from the parent type must be specified, and that specification may require adding new types or relations

– Overlap (there may be parts of the meaning of one type similar, and other parts different)

16

Difficulties with Mapping Ontologies Developed Completely Independently

– Representations often combine fundamental components of meaning in different ways

– Elements of different ontologies may overlap, rather than map directly or be in a hierarchical relation

• The areas of overlap and non-overlap may require creation of new types or relations, more basic than the composite types.

• Dissecting the components of each overlapping representations requires human-level intelligence

• Creation of new basic types and relations requires human-level intelligence; cannot be done automatically

– The documentation rarely has sufficient information even for a human to resolve the ambiguities

– Mapping legacy ontologies to a common Foundation Ontology will reduce the efort from order of n2 to n.

17

Benefits of Mapping to a Common FO

• Ontologies or Database Schemas newly created using the basic “conceptual vocabulary” of the FO will be automatically interoperable from their creation.

• Retroactive integration of ontologies or DBs can benefit from mapping (semiautomated or by hand) to a common FO.

• The effort for mapping to a common FO is order of n, rather than order of n2 for mapping between domain ontologies

18

More Difficulties with Mapping• Discovering the relations requires human inspection,

though some automated methods can assist by suggesting relations, derived from text. It is not possible to automatically specifying the meaning of a relation by axioms expressing the inferences derivable from that relation holding.

• Unless the original developers of two ontologies are available for consultation, the documentation will often be inadequate to resolve ambiguity of meaning of the elements

• Creating a mapping may take more time than creating a new common ontology for the two applications

19

Difficulties with Mapping Ontologies (continued)

• If two ontologies are accurately mapped (e.g. manually or semiautomatically), then the result will be in effect a common merged ontology. This can be used for interoperability, BUT:– (1) The net effect will be the same as having started with a

common ontology, but the merger will be useful only for the ontologies mapped, and not for others. Mapping to additional ontologies will be increasingly complex.

– (2) The effort to create this kind of mapping will only be cost-effective for the most important problems.

– (3) The cost of such a process on multiple ontologies will be much greater than the cost of defining the domain ontologies by use of a common FO, unless the domain ontologies are very simple (few semantic relations and an accurate inheritance hierarchy)

20

How Can Incompatible Theories be Included?• The basic concepts that are required to specify meanings are generally

agreed on. Differing viewpoints will usually be expressible in terms of a common vocabulary, and assertions in different terminologies or syntaxes will be directly translatable into each other.

• One example: the often mentioned ‘incompatibility’ between 3-D (endurantism) and 4-D (perdurantism) views of objects in time. The actual assertions of each viewpoint are accurately translatable into assertions in the other viewpoint:

• Pat Hayes (email to UOM-forum Aug. 8, 2009): “But, for the record, I reach the conclusion from the observation that

anything that can be said in a 4D ontological framework can be transcribed into a 3D framework based on the continuant/occurrent distinction, and vice versa. The differences between them, I have concluded, are really nothing more than a matter of notational choice. “

21

3D-4D Translation Axioms (Pat Hayes)From Pat Hayes: (forall (x (t Time) P)(iff (P x t)(P (x during t)) ))“Think of this as a 'bridging' axiom, part of a translation specification, if you like”. . . And later;• There are a variety of notational options in combining a simple timeless assertion with a temporal

parameter. One is to treat the time as a context, in effect attaching it to the entire sentence (or in IKL, proposition): (ist t (P x y)) (ist t (that (P x y)))

• another is as an extra relational argument, giving the 'fluent' style which goes naturally with continuants: (P x y t)

• and a third is to connect it to the object(s) being related, the relation then being naturally understood as a relation between time-slices:

(P (x at t)(y at t))• But in fact, these are really all just notational variations on a single theme. They amount to choosing

where in the parse tree of the simple _expression_ to attach the parameter, is all. If we simply FORGET the philosophy for a second, then we can treat this as an arbitrary conventional choice, and think of them as all meaning exactly the same thing, and therefore equivalent.

22

3D-4D Translation Axioms (COSMO)• {PH isanInstanceOf Object}• {PH4D isanInstanceOf Object4D}• {t1 isanInstanceOf TimePoint}• {t2 isanInstanceOf TimePoint}• {t1t2 isanInstanceOf TimeInterval}• {t1t2 hasStartingTimePoint t1}• {t1t2 hasEndingTimePoint t2} • {PH4D isTheWholeLife4dVersionOf PH}• {PHt1t2 isaTimeSliceOf PH4D from t1 to t2}• ;; If we included a ‘during’ similar to the one Pat Hayes uses, it might look like:• {(PH during t1t2) isIdenticalTo PHt1t2};; The bridging axiom for a specific assertion would be:• {{PH isLocatedAt IHMC from t1 to t2} iff {PHt1t2 isLocatedAt IHMC}}• ;; And, redundantly, given the above:

• {{PH isLocatedAt IHMC from t1 to t2} iff {(PH during t1t2} isLocatedAt IHMC}• • The above explicitly has a 4D entity PH4D as TheWholeLife4dVersionOf the ‘dimension neutral’ object PH.

NOTE: The bridging axioms can be generalized by using row variables.

23

Bridging Axioms in General

• More detail for bridging axioms for various scenarios translating different styles of representation were presented in:

IKRIS Scenarios Inter-Theory (ISIT)– Jerry Hobbs with the KRIS Scenarios Working Group– http://nrrc.mitre.org/NRRC/Docs_Data/ikris/ISIT_spec.pdf

• Mirrored at:

http://micra.com/COSMO/HobbesEtalBridgingAxioms.pdf24

How Can Incompatible Theories be Included?(continued)

• When representation of genuinely logically incompatible theories, not merely different viewpoints, are desirable in the FO or in some extension, the theories can be represented as theories using the defining elements of the FO. The assertions in theories are not themselves directly part of the ontological commitment of the FO, and describing incompatible theories does not make the FO itself inconsistent.

25

Representation of Incompatible Theories Does not Make the FO Self-contradictory

• A logical contradiction in the FO would have some pair of statements of the form:– (P ?x) and (not (P ?x))

• But theories are represented in the FO as separate contexts:– (isTrueIn (P ?x) Theory1) and (isTrueIn (not (P ?x))

Theory2)• Logically contradictory theories can be described in

the FO but not asserted to be true in the FO itself.

26

Similar Approaches• H. Wache, T. Vogele, U. Visser, H. Stuckenschmidt, G. Schuster, H. Neumann, and

S. Hübner, "Ontology-based Integration of Information -- a Survey of Existing Approaches," Proceedings of the IJCAI-Workshop Ontologies and Information Sharing, Seattle, WA: 2001, pp. 108-117Accessed at: http://www.let.uu.nl/~Paola.Monachesi/personal/papers/wache.pdf

• H. Wache, "Towards Rule-Based Context Transformation in Mediators," in Proceedings of the International Workshop on Engineering Federated Information Systems (EFIS), 1999, pp. 107-122.

http://citeseer.ist.psu.edu/cache/papers/cs/9658/http:zSzzSzwww.informatik.uni-bremen.dezSz~wachezSzPaperszSzefis-99-wache.pdf/wache99towards.pdf

27

Will Translation Among Logically Incompatible Ontologies Always be Possible?

• Not necessarily. BFO is (for example) a single-inheritance ontology, and it is possible that trying to translate assertions from multiple-inheritance ontologies would cause a logical contradiction.

• This might be avoided if the single-inheritance axioms of BFO are only used during the development of the classes of BFO-dependent ontologies, and not during data (instance) entry or query time .

• There may or may not be workarounds for other cases of logical incompatibility. The Foundation Ontology (FO – see below) project would have to determine whether there are practical workarounds for true irreconcilable inconsistencies.

• Groups that develop ontologies too inconsistent with the FO to use the translation mechanism , may develop a special FO for their own community with whom they must interoperate.

28

Integration of Diverse Information• Multiple diverse views of the same information

will always be present• Integration requires a method to translate from

one terminology and format to others• Overlap of meanings requires dissection of

complex meanings into component primitives• For integrating multiple diverse views, a

foundation ontology having representations of all of the primitives is required; a common syntax and bare logical functions are insufficient to resolve ambiguity and meaning overlap.

29

Integration of Knowledge SourcesVia Semantic Interoperability

• Representation of knowledge using a logical-based ontology allows automated inferences using multiple data sources – “connecting the dots” rapidly and accurately, based on rules created by the domain experts

• Automated reasoning that is reliable enough to be trusted to make important decisions without human intervention requires accurate information.

• Information transferred from other systems can be used reliably only if the information is interpreted accurately. 99% accuracy is insufficient.

• Accurate automated interpretation requires a common foundation ontology among information sources.

30

Why is 99% Accuracy Insufficient?The number of inferences deduced in the course of

proving a test theorem can be greater than 10,000. (See: Owen L. Astrachan and Mark E. Stickel, Caching and Lemmaizing in Model Elimination Theorem Provers http://www.cs.duke.edu/~ola/papers/cade92.pdf.

If the likelihood of error in each step is as low as 1%, the chance of reaching a correct conclusion is 0.9910000 = 2 -44

If the number of inference steps in solving a problem is 68, there is a 50-50 chance of arriving at the correct conclusion with 99% accuracy in translation.

31

Can a Foundation Ontology be Generated Automatically by Extraction From Text?

• Semantic relations are the heart of the meaning of an ontology.

• Some relation labels among elements can be extracted with modest accuracy (<50%) from text;

• BUT: The logical implications of relations, which are the basis for the meanings, are extremely difficult to specify without human-level understanding

• Therefore an automatically generated ontology will be of low quality and need human editing to be useful for accurate reasoning.

32

Is there any Benefit to Extract an FO Automatically From Text?

• Automatic ontology generation may be helpful in new and specialized domains, where no existing relevant ontology covers the field; BUT– (1) there are already several FOs from which to draw

candidate basic concept representations; – (2) if there are in fact a limited number of primitives,

the effort at hand-crafting an FO will not be prohibitive, since it will only have to be done once, with low maintenance cost.

• Therefore there is no benefit in attempting to extract an FO automatically, even if that process helps with some domain ontologies.

33

Foundation Ontology• Generically, a Foundation Ontology is an ontology containing

logical representations of the most general (abstract) entities (types, relations) that are used in constructing more specialized or domain-specific representations. Existing examples are OpenCyc, SUMO, BFO, DOLCE, ISO15926 and others.

• For practical convenience, more specific extensions can be maintained to avoid unnecessary recreation of existing ontology elements; these extensions can form a hierarchy of ontologies (logical theories)

• If logically inconsistent ontologies are included in the set of reference ontologies, they may be represented as a lattice of theories.

• An FO used to support interoperability of any set of domain ontologies will have all of the basic concepts required to represent

34

35

What A Common Foundation Ontology Isn’t

≠ A controlled vocabularyEach community can choose its own words to refer to concepts, and map those to the FO

≠ A mandated standardUsers can use any common ontology or none, as their own needs dictate. A common FO is required only for accurate communication among multiple independent applications.

≠ A Restriction on expressivenessAn individual user can use any local application with any language or technology. What must be expressed using the FO is only that information that needs to be shared with other communities.

Primitive Concept RepresentationsA used here, a primitive ontology element is an

ontology element whose intended meaning cannot be represented as a FOL combination of other elements in the FO. An FO that is intended to function to translate elements from one ontology to another should have all of those primitives that are used to represent the ontelms in either of the ontologies; the primitives may be in the FO itself, or in some mid-level or domain-level ontology used by both communicating ontologies.

36

Are There a Fixed number of Primitives?• It is possible that there may be no limit to the number of primitive

ontology elements required to construct other ontology elements in all other domains. For the FO principle to serve for translating multiple ontologies, it is only necessary that all of the primitive elements required to construct all those ontologies are represented in the FO or in some extension common to the communicating ontologies.

• Therefore it is not necessary that there be a fixed number of primitive ontelms in order for the FO tactic to support accurate interoperability.

• However, evidence from linguistic experience suggests that the number of primitives required for broad applicability of an FO may be small. This evidence may provide some participants with additional motivation to explore the FO tactic. Primitives are discussed further below.

37

Are There a Fixed number of Primitives?Arguments from communication

• The number of primitive concepts people use internally for their own thinking cannot be easily determined.

• What is important for interoperability is the number of primitives used for communication.

• Accurate communication depends on agents using symbols whose meaning is understood by all – this in turn depends on the meanings being associated with common perceptual experiences. The number of such distinguishable common experiences is limited.

• By age 18, most people can understand definitions of new terms based on the fundamental concepts they have already learned.

• This is the basis for the use of a limited defining vocabulary in dictionaries like Longman’s.

38

The Chemistry Analogy(not to be taken seriously)

• Primitive Concepts can be viewed as the atoms of thought – the smallest units of information manipulated in human thinking and automated reasoning

• Complex concepts are like molecules, composed of two or more primitives.

• Natural language is like cooking – composing useful mixtures of molecules to create an interesting medley of concepts. Linguistic expressions are “food for thought”.

39

How Is Semantic Interoperability Achieved by a Common Foundation Ontology? (overview)• The elements of domain ontologies or databases are represented

as First-Order-Logic (FOL) combinations of ontology elements (types, relations, axioms, functions – for short, “ontelms”) already present in the Foundation Ontology.

• When information is to be communicated between systems using different domain ontologies, each system communicates, in addition to the data, the logical descriptions (axioms) for ontelms not already in the Foundation Ontology (or public extensions) that are required to understand the meanings of the data.

• Each system, able to interpret both FOL and the ontelms used to describe the meanings, will be able to produce the same inferences from the same data, when both use the same or a functionally equivalent FOL inferencing engine.

• If the reasoning used in each local system is restricted to logical inference on the represented knowledge, interoperability will be optimal. Local procedures may be created for efficiency purposes, provided that they use the knowledge in ways compatible with the logical meaning .

40

Semantic Interoperability via an FO (more detail: 1)The goal: An FO that can support the goal of “broad, general, accurate

semantic interoperability” can be viewed as: a system of agreed data structures and programs that allow *any* local group using this common system to place information **on any topic** on the internet or some other public place, or to transmit it directly to another system, and have the information interpreted in the sense intended by its creators, regardless of whether the transmitting and receiving systems have any prior contact. Proper interpretation requires that both transmitting and receiving systems reach the same inferences from the same data, and have the same real-world referents for each term.

Any system that has more relations, will of course be able to reach additional inferences, but these will not be logically contradictory to the inferences that the less complete system reaches.

A system that has more data may also reach additional inference, but the inferences should not be logically contradictory to the inferences reached by the less-informed system, unless the additional data itself is contradictory to that in the less-informed system.

41

Semantic Interoperability via an FO (more detail: 2)

The FO would be used in this manner:1. The ontelms in the FO all have a meaning agreed to by the participants,

and the logical specifications and linguistic documentation is unambiguous enough to satisfy all participants that they agree on the intended meanings,

2. The ontelms in domain ontologies or upper ontologies are identical to or logically specified as FOL combinations of ontelms in the FO (or in extensions of the FO).

3. The computations performed with ontology-specified data in applications (other than simple input-output, or computations not affecting data communicated among applications) are performed either (a) using an agreed common implementation of FOL; or (b) the procedural code that is part of some element in the FO. Thus the calculations performed on data in communicating systems should be identical, and produce identical inferences.

42


4. When any two programs that want to interoperate and have separately developed domain ontologies need to communicate, then in addition to the data that is to be transmitted, the transmitting system must send all the logical descriptions of the domain elements needed to describe the data that are not already in the FO or in some extension used in common between those two applications. There then needs to be an integrating program that (on the receiving side) takes the new descriptions of previously unknown elements, and integrates them into the local ontology, to arrive at an ad-hoc (temporary) merged ontology that is sufficient to properly interpret the data communicated. The merger should be accurate because all of the new ontology descriptions use only FO elements in FOL combinations, and the FOL implementation is common among all communicating systems.

43

5. Any application that can properly interpret elements of the FO should be able to properly and consistently interpret elements described as FOL combinations of those elements.

6. Therefore the computations performed by all applications using the FO should all arrive at the same inferences from the same data. That is all one can demand for programs that are intended to be interoperable.

7. If any procedural code is used locally that manipulates the data other than for input output or presentation, there may be a risk of misinterpretation. The local programmers need to be aware of the risk, and avoid misuse of the data so as to change its intended meaning.

44


8. For information not transmitted to other systems, of course local systems have complete freedom to use them as they consider optimal. It is only the information transmitted to other systems that has to be interpretable by means of the FO specification.

9. Recall that the FO will be able to have procedural code labeled as functions. Any systems that require procedural code for proper interpretation of transmitted data, that is not adequately mimicked by FOL, can add it as a primitive function to the FO or to a domain extension ontology used within some community.

10. The FO, in order to accommodate newly mapped systems that require new primitives, should have an expeditious procedure for rapidly adding new primitives, after review by the technical committed agrees that the new element is not FOL specifiable using existing FO ontelms, and is not redundant or logically contradictory to the existing FO.

45


Potential issues (1): There is one potential problem in he manner o f using newly specified

ontology elements required to interpret transmitted information. I may not always be possible to recognize when the intended meanings of elements in separately developed domain ontologies are identical. Since the FO allows alternate structures to represent the same meanings, but has translation axioms among them , the various alternatives can in principle be calculated and compared for identity. But unless the system can develop some normal form into which all elements can be converted, identical meanings may not always be recognized as such. It will have to be investigated by the FO consortium whether it is possible to develop a normal form for the FO, or if not, whether failure to recognize identity would have significant negative effects.

46


Potential issues (2):An additional issue is whether newly added axioms could change the

interpretations of existing FO ontelms. To minimize that potential, it would seem important to try to identify all axioms necessary to specify the intended meanings of the FO primitives as fully as possible at the earliest stage, so that few if any need to be added after the initial shake-down period of a few years. Additions of new subtypes or relations that are only conservative extensions of the FO may not be problematic in the same way.

For stability, it is important that the intended meanings of FO ontelms remains constant, so that the logical interpretations of elements does not change over time. When elements representing new concepts are needed, they are added to the FO or some extension. Extensions should not create any logical conflict with the FO.

Systems that have local data or local ontology extensions may derive additional inferences from the same data, but these should not be logically contradictory to those derivable by other systems.

47


Other Advantages of the Common FO• The FOL rules that can be created within a domain ontology

can represent not only the data combining operations that are performed in procedural programming, but can also implement checking for consistency and accuracy of the input and results.

• If some procedural code is nevertheless required for local data processing, the use of an FO will still reduce the number of data elements that need to be interpreted carefully and processed according to the common interpretation; as a result, the chance of inadvertently using an interpretation different from that used by others will be reduced.

• Procedural code that is useful for more than a few local uses, if representable as a function, may be included in the FO or a mid-level extension as a new primitive element.

48

49

The Integrating Function of the Foundation Ontology

Obligation Duty

GenericObligation

SameAs

SameAs

Foundation Ontology

Domain Ontology 1Domain Ontology 2

50

Foundation Ontology (FO)

Provides defining concepts to specify conceptual message Content

The Ontology for Integrating Databases

GeneralKnowledge

Knowledge Base

uses FO For Definitions

Data Collection Interfaces

Database Translating Interfaces

Customers InventoryPatientData

CommercialProducts

Regulations Intelligence

51

Common Foundation Ontology providesdefining concepts to specify content of

messages passed among modules

The Foundation Ontology for Integrating Applications

AnalysisSupport

Case-Based Reasoning

LinguisticInformationExtraction

Probabilistic Reasoning

SpatialReasoning

Information Store(s)Use FO forDefinitions

Interfaces

InformationRetrieval

Sensorsand

Robotics

Task Control: SelectProcesses To Solve Current Problem. Iterate to more specific problems.

InterfaceTo user

52

In Multi-Agent Architectures Like Cougaar, the Ontology Can Be the Backbone of The Communication

Language (http://www.cougaar.org/)

Uses Ontology

Uses Ontology

Special Agent Communication Protocols?• Certain agents in a multi-agent architecture may only need to

communicate with a small number of other agents (a point emphasized by John Sowa on the Ontolog forum Jan 2010). These communications may take advantage of protocols much simpler than the FO for accurately sharing information among them. If any one of these agents can express its information according to the FO, this local community can also interoperate accurately with the wider world of FO users.

• Such specialized protocols could be used for independently developed agents. The FO could include a means to make such protocols publicly accessible.

• Such specialized communications protocols may reduce the overall effort to create certain multi-agent systems that can communicate with other FO-aware systems.

53

Global Semantic Integration; Combined use of Local Interfaces and Protocols with FO for Global Communication

Research;

Commercial Ontologies and Databases

Government Agencies:InteragencyProtocols

Academic

Small businesses

Commercial Organizations Communicating Privately

Foundation Ontology provides the grammar and vocabulary for communicating among local ontologies or DBs

Small Businesses:May have special internal or Local

interfaces to other businesses

Individuals orCommunities

General Public

Visualization Tools

Reasoning Tools

Manufacturing,Inventory

Personnel,Internals

Sales,Advertising

Single Large Corporation; Local interfaces not using the FO

The Evolutionary Tactic• The combination of a common standard of

communication among modules, and a modular design for applications, allows incremental evolutionary improvements in function. Improved modules can be created by totally separate groups, and plugged in to the whole system.

• The same standard, used to communicate information among multiple applications, similarly allows incremental improvement in the function of multiple applications (or agents) to achieve different goals.

• Separate applications or modules can communicate to form a “Society of Mind” (Minsky) approaching human levels of information processing ability. Communication among the modules uses the FO as a common language to express meanings.

55

The Foundation Ontology . . .. . . is not required to be used in toto in every application; individual

applications will only use as much as is needed to support the reasoning for that application. Redundancy will not cause computational inefficiency in the applications, as an application only needs to use one of the alternative views. A utility should be included with a common FO, to extract only the needed parts. This tactic is used in the NIEM, where IEPD’s may use only a small number of the total elements.

. . . is required when separately created ontologies, applications, or databases need to transfer information. The FO supports translation of data from one local terminology into the other by having a complete inventory of primitive elements into which complex domain entity representations can be analyzed.

. . . Will not break existing applications or databases if used only for translating data transferred from one system to another.

56

The Foundation Ontology . . .. . . Is intended to be internally Logically consistent.Alternative theories can be represented using the

same set of primitive ontology elements.If (A) is an assertion expressible using ontelms of

the FO, and an alternative assertion (not A) is also desired for some application, then both (A) and (not A) can be moved to an extension of the FO as alternative theories; both (A) and (not A) are describable (can be expressed) using only elements in the FO.

Example: ontology has Theory1 and Theory2 ontology may assert: (holdsIn A Theory1) and not(holdsin A Theory2)

57

What Does it Mean to “Specify the meaning of a term”? (simplified example)

“The birth mother of a person is a woman who has given birth to that person”

{{?Mother isTheBirthMotherOf ?Child} impliesThat (ThereExists {((exactly one) ?Event) and ((exactly one) ?Date) and ((some) ?Location)} suchThat {{?Event isa BirthEvent} and {?Event occurredOn ?Date} and {?Event occurredAt ?Location} and {?Mother is (The Mother in ?Event)} and {?Child is (The Baby in ?Event)} and {(The BirthDate of ?Child) is ?Date} and {(The BirthPlace of ?Child) is ?Location}})}

Are All Composite Ontelms Defined by Necessary and Sufficient Conditions?

No. Most types in a domain ontology will only be logically specified by necessary conditions. This leaves some ambiguity in the logical description, but the intended meanings can be made as specific and unambiguous as is required for proper use of the ontelms in applications that use them.

59

Meanings for the Foundation Ontology• Whether meanings are interpreted intensionally (as equivalent to their

ontological representations) or extensionally (by use of verification procedures), the ontology itself serves to construct the meanings used by the computer for reasoning and deciding.

• Evidence that database meanings have been properly interpreted will require human evaluation of the correctness of inferences. The goal is for every data element to be used in computer programs in a manner that is consistent with the intended meaning. If a program uses procedural code for reasoning, rather than FOL inference on ontology instances, then the meaning must be clear to the programmer.

• Evidence that text meanings have been properly represented can be obtained from (1) question-answering or (2) conversation (the Turing test).

• For robotic systems, recognizing objects and object types, performing actions and recognizing when actions have been performed will be additional tests (“procedural Semantics”, as described by Woods).

60

Meaning: Procedural Semantics

Meaning and Links: William A. Woods, AI Magazine 28(4) Winter 2007

– "In this theory the meaning of a noun is a procedure for recognizing or generating instances, the meaning of a proposition is a procedure for determining if it is true or false, and the meaning of an action is the ability to do the action or to tell if it has been done."

61

Are Ontology Meanings Stable?• For any given ontology, the sum of all inferences for any data

input will always be the same. If the FO is changed, however, there is a potential for a program using the FO to change its output for the same data.

• The pragmatic criterion for “meaning” to remain constant is: the same data will give the same output for all programs using the FO. If a change to the FO does not change the behavior of existing programs using it, then the relevant meanings of the ontology elements representing those data elements is considered unchanged.

• Thus, the FO can change without changing the relevant meanings of the ontology elements used in programs.

• Conservative additions of ontology elements to the FO may not cause any changes to existing program behavior.

62

Primitive Concepts• Primitives: the most basic units of thought (such as the part-of relation)

that are used in combination to create more complex units of thought (such as an Automobile).

• A Primitive is a concept or ontelm that cannot have its meaning specified solely by use of some FOL combination of other independently described primitives

• No consensus on how many primitives there are• The COSMO project aims to provide an estimate of the upper limit (if any)

on necessary primitives• Focus on primitives is useful to provide a starting target for a foundation

ontology of minimum size; it is not a technical requirement for a functioning foundation ontology

• Focus on primitives can provide a useful paradigm for those who expect a limited number; it can be ignored by participants in an FO development project who expect an unlimited number of primitives; the FO only needs to include all the primitives used by the domain ontologies and applications that want to communicate by means of the FO at any given time.

63

How Many Primitives?• Wierczbicka’s “universal core” contained 60 primitives

common among multiple languages (see Cliff Goddard Bad Arguments Against Semantic Primitives, in Theoretical Linguistics, Vol. 24 (1998), Available at: http://www.une.edu.au/bcss/linguistics/nsm/pdfs/bad-arguments5.pdf)

• The Longman Dictionary of Contemporary English (LDOCE) uses 2148 words to define its over 64000 terms.

• Cheng-Ming Guo analyzed the Longman defining vocabulary (Ph.D. Thesis, 1989) and determined that there are 1433 actual “basic” words (representing 3200 word senses) that can be used, recursively, to define all of the words in the Longman dictionary

64

How Many Primitives? (continued)• The Japanese Toyo Kanji contain 1850 characters – those required to

be learned by completion of secondary education. Some basic words are represented phonetically, not as characters.

• In Chinese, knowing 3000 to 4000 characters qualifies one as “literate” (able to read a newspaper). In modern Chinese text, the first 3000 characters cover 98.94% of text and the first 4000 cover 99.68%.

• Sign language (AMESLAN) dictionaries contain from 2000 to 5000 signs.

• The first representation of the Longman defining vocabulary plus associated basic concepts in COSMO will contain at least 7000 types and 700 relations, but probably fewer than 10000 total elements (in progress). Many of these may not be primitive.

• Doug Lenat speculates, from experience with Cyc, that as many as 15,000 primitive concept representations may be needed to serve as a “Conceptual Defining Vocabulary” (personal communication).

65

The Lesson From Linguistic PrimitivesThe experience from the Longman defining vocabulary,

the Chinese inventory of commonly used characters, and the number of sign-language signs shows that:

• When there is a reason to minimize the number of symbols used to communicate, it is possible to find a small set of a few thousand that can be combined to describe any other useful concept.– Whether this principle will also hold when the

symbols have the precise meanings of ontology elements needs to be tested.

66

What Makes a Concept Primitive?• In the computer context, the most basic primitives are those that cannot be

specified solely by FOL combinations of pre-existing ontelms, but must include some procedural code in order to be used properly by applications. Calculable arithmetic functions are examples.

• If two ontelms can only be represented by mutual reference (direct or transitive) to each other, they are considered as co-primitive (if either one is a primitive), or co-specified (if inside an extension to the core FO)

• If the meaning of an ontelm can only be described by reference to example instances, rather than by necessary conditions, it is considered as primitive. Implied reference to experienced instances is characteristic of terms referring to feelings or emotions.

• If the meaning of a new element is expressed solely by asserting that it is disjoint with some primitive element(s), it is a primitive.

• If a new relation has no logical inferences derived from an assertion in which it relates other ontelms, it is primitive. This means that its meaning is available only in the documentation or comments interpretable by humans. Its meaning in applications will be determined by its usage.

• Other criteria may be also be identified, and can be adopted by the consortium developing the common FO.

67

Types of PrimitivesProcedural• For logics such as FOL the primitives are the procedural implementation of the logical symbols.• For semantic interoperability, the implementations must be identical, e.g. open-world versus closed-

world.• Although logicians can manipulate the symbols consistently, programs need procedural code to carry out

the transformations implied by the logical symbols and their traditional interpretation.• A practical foundation ontology needs to include, as well as the implementation of the FOL symbols all of

the procedural code that cannot be represented by the logical symbols of the FOL (or other logic) that is agreed to as the basis for semantic interoperability among the community that intends to be interoperable. Those procedures will be among the primitives agreed to, which then need to be included in the foundation ontology.

Instance-based (Perceptual, Intuitive)• *Other primitives* that are not procedurally encoded will depend on human interpretation – all

programmers need to understand the intended meaning and, when using those concepts in a program, need to use them in a way that does not generate logical contradictions. Such primitives may include representations of emotions or other mental objects that can only be fully understood by exposure to instances, i.e. depend on individual perceptions as well as on the expectation that other people, especially computer programmers, have closely analogous mental states induced by those perceptions. To the extent that a programmer’s understanding of such primitives may differ from the understanding of other programmers, there is a risk that programs using data in such categories will be inconsistent with each other.

• At some point, computers ma be able to test assertions about types by retrieving information from the Web about well-known instances of those types. This provides some potential “grounding” of meanings independent of human interpretation. 68

Procedural Primitives• To assure that information is interpreted

consistently among applications, wherever procedures are required to interpret transmitted information, the procedures must have the same computational effect (even if coded in different languages).

• This may require the communication of procedures composed of combinations of more basic procedures. Some common procedural language will be required.

69

How Many Primitives? (continued)Are there an infinite number of “microsenses” (Wittgenstein) among the

primitive concepts?• In any given “language Game” (Wittgenstein) each word has only one or a

few senses• Creating texts with Definitions and Descriptions is one form of “language

game”. The need to convey precise information to a general audience without opportunity for feedback (as in written text) forces the use of a small number of widely understood senses.

• The words of the linguistic defining vocabulary (as in Longman’s) are therefore likely to have only one or a few senses that are required for the word-definition “language game”.

• Guo’s work provides evidence that this is true; fewer than 2 senses per word are required for the Longman definitions.

• Given a line and the Peano axioms, an infinite number of line segments of different lengths can be defined. The use of a large number of composable concepts does not at all imply that the number of primitives must be large.

70

How Many Primitives? (continued)Will there be specialized primitives required for specialized fields?Perhaps. One may visualize, for example, abstract mathematical objects that have a

certain relation between them, unlike any relation between objects in the physical world. Such newly defined relations may also be primitives, and there may be a large number of them. But math primitives differ from primitives intended to represent real-world things.

But in general, these primitive relations will only be needed for communication **within** that specialized field or sub-branch of some field, and can be maintained in the specialized ontology extensions used for those specialized fields.

Since they are not needed or useful for communication with other fields, they do not have to be added to the foundation ontology itself. The foundation ontology needed for accurate communication among different fields will still be stable. The primitives need to be in the FO only if they are needed to represent information that is transmitted to agents not using the same extension ontology.

The main concern of the Foundation Ontology is that the number of primitives required for communication among diverse applications be as stable as possible.

71

72

How many words are needed to understand a text?

Estimates of minimal vocabulary sizes needed for academic purposes start at a low of 5,000 words for reading authentic texts (Laufer, 1997) and range up to 10,000 words for reading university textbooks (Hazenberg & Hulstijn, 1996). Nation (2001) argues that at least 97% of the vocabulary of a text need to be known to gain adequate understanding of the text. To read literary texts extensively with understanding and relative ease, 98% of the words of the texts need to be known (Hirsh & Nation, 1992; Hu & Nation, 2000). Native English speaking children consider a vocabulary load of 2 unknown words per 100 words (98% known) difficult reading (Carver, 1994). 2 words per 100 words translates into roughly 1 unknown for every 5 lines of text read

The COSMO Project• Motivated by an absence of a widely accepted Foundation

Ontology that can serve as a standard of meaning• The COSMO ontology is intended to serve as a test

ontology to investigate the Foundation Ontology principle, to demonstrate how an FO with all of the primitives ontelms required to specify some set of domain ontologies will support accurate semantic interoperability among those domain ontologies. These efforts with the COSMO will also test the size of the required inventory of primitive concept representations.

• COSMO was initiated in in 2005 [13] as a project of the Ontology and Taxonomy Coordinating Working Group (ONTACWG) , a working group of the Federal Semantic Interoperability Community of Practice.

• Since then it has been continued by Patrick Cassidy

73

The COSMO Project (continued)• Since late 2007, the objective has been to create an

initial version that includes representations of all of the words in the Longman Defining Vocabulary, as well as other ontelms that are considered as basic for describing everyday things.

• This version will be tested to determine if it contains all of the primitives needed to represent terms in specialized fields.

• The number of new primitives required for each increment of new representations will indicate whether there is an asymptotic limit to the number of primitives required to represent all fields.

• This criterion of sufficiency is probabilistic.74

The COSMO Project (continued) Attaching Linguistic labels for NLU

WordNet Synset Assignments Cannot Be one-to-oneExample: verb ‘move’ sense 2 in WordNet: 2. (60) move, displace - (cause to move, both in a

concrete and in an abstract sense; 'Move those boxes into the corner, please'; 'I'm moving my money to another bank'; 'The director moved more responsibilities onto his new assistant')

Includes: (1) physical motion = ‘Translocation’(2) Transfer of money = ‘MoneyTransfer’(3) Transfer of Responsibility = ‘AssigningaResponsibility’

75

Principle for Labeling COSMO Elements with English Labels:

• Use individual words in their most common senses• Otherwise use word combinations to make the meaning

as clear as possible without a paragraph.• The documentation elaborates on what the label hints atSteven Weinberg: “If words are to have any value to us, we

ought to respect the way that they have been used historically, and we ought especially to preserve distinctions that prevent the meanings of words from merging with the meanings of other words.” – from Dreams of a Final Theory

76

What’s New in the COSMO?

• A little less than half of the ontelms in COSMO are not also present in OpenCyc or SUMO

• BUT the goal is to make it as small as possible while still having all of the semantic primitives needed to describe entities in any domain

• Keeping it small will make it easier for multiple developers to agree on the structure, and make it easier to learn and to use

• “A theory should be as simple as possible, but no simpler” -- Einstein

77

COSMO Phasing

• Phase 1 will develop an OWL ontology with representations of all of the Longman defining words. (est. completion second quarter of 2010)

• Phase 2 will test that version for adequacy in specifying meanings of at least 1000 specialized terms

• Phase 3 will convert the OWL version to a Common-Logic compatible version

• Phase 4 will develop a Natural Language interface to the ontology to make use easier

78

Open Source, Open Method

• To serve as a widely used standard, any ontology needs input from many different developers and users with differing views and preferences. COSMO is fully open to input from any source, provided that it is logically consistent with existing content.

• If funding becomes available for a collaborative development of a Common Foundation Ontology by a similarly open method, that project will supersede COSMO.

79

Multiple Viewpoints• An important function of a Foundation Ontology is to serve

as a means to translate other, specialized knowledge representations into each other. It may be, but does not have to be, used as the only top-level ontology.

• Different ways to represent the same entity can be accommodated, provided that they are logically consistent and can be translated into each other. Everything that anyone feels is necessary can be included – if not in the core of semantic primitives, then in an extension representing alternative theories.

• A given application may use only a small part of the COSMO, extracted as needed for its own purposes; therefore redundant alternative representations will not reduce the computational efficiency of applications

80

Criterion for Evaluation• The question to be determined is whether new primitives,

beyond the starting inventory representing the Longman vocabulary, are required to represent knowledge in specialized domains, and if so, how many?

• The rate of increase of the number of ontelms in the COSMO for each increment (e.g. of 1000 term representations) will provide evidence whether there is a limit (an asymptote) in the number of terms required to represent many other fields.

• If no asymptote is suggested, a small rate of increase may still allow use of a common Foundation Ontology as a means of semantic interoperability, but with more careful attention to versioning.

• When mature, the need to add new primitives should rarely occur

81

COSMO: Current Status

OWL version (January 2010)• Types (classes): 6750• Relations (OWL properties): 770• Restrictions: 1466• Longman Terms remaining to be represented: 300 (out of 2148)

82

Summary of COSMO Effort to Date• Representation of the Longman defining

vocabulary in OWL will likely require fewer than 10,000 ontology elements.

• Some of those are not primitive elements, and can be specified as combinations of other elements. They are included because they represent common concepts and will ease the creation of a Natural Language Interface.

• Planned addition of rules in a CL-compliant format will increase the number of elements, by at least the number of relations

83

Why Hasn’t a common FO Been Adopted Yet?

• Existing Upper Ontologies are complex theories developed primarily by small closed teams and presented to the world for adoption.

• They:– Are complex and hard to understand, and therefore time-

consuming to use properly– Each use only one of several alternative ways to represent

concepts, unsatisfactory to many who have already adopted other formalisms

– Have no publicly accessible open-source applications that demonstrate their utility, therefore not providing the incentive required to motivate the effort to use them

84

How Can a Common FO Development Process Avoid Those Problems?

• Provide adequate funding to include at least 100 participants with substantial time (>20% full-time) to ensure that a wide variety of views are included.

• Keep the ontology and associated programs full open-source and invite comments from any interested party

• Include all identifiable primitive concepts required by the participants• Include all desired logically consistent ways of expressing concepts, with

translation axioms among them• Develop a subontology extraction procedure to eliminate unnecessary

complexity when applications are developed• Develop a Natural Language Interface and other utilities to make the

ontology easy to understand and use• Develop at least three open-source applications including a basic

language-understanding program, to demonstrate the utility.

85

The Consortium Process for Developing a Common FO

• Has the initial goal of creating an open-source Foundation Ontology that is freely usable and developed by a process open to input from any source, evaluated by the technical committee of the consortium

• Has the additional goal of creating a user community broadly representative of diverse views and approaches to developing logic-based ontologies, that can maintain, test, and evolve a common FO.

• The main goal is to providing free open-source examples of both individual applications using the FO and demonstrations of how the FO enables accurate semantic interoperability among the independently developed applications.

86

The Consortium Process (Contd.)• Will require at least 100 participants spending

substantive (>20%) time, to assure adequately broad representation of different views on fundamental ontology issues, and suitability for creating representations used in many different applications.

• Will require a commitment by the participants to test and critique the initial FO for several years.

• Anticipates a need for continued maintenance, as new applications may reveal unanticipated need for new primitives.

87

The Consortium Process (Contd)• Will have as the core effort finding a means to relate

different views for representing basic ontology elements, and translate the different representations into each other, so that any ontology developed using one representation can be translated into the vocabulary and format of any other mapped to the common FO.

• Will develop utilities to make the FO easier to use (preferably with a NL interface), and example applications demonstrating how it can be used to enhance performance of applications.

88

FO Development by the Consortium • The development method will be agreed to by the participants

in preliminary discussions. One possibility is below (see also continuing slides) – but a different method may be chosen– Start with all foundation ontologies (“upper ontologies”)

used by the participants, plus the root concepts in the hierarchies of the domain ontologies (and database schemas) and all of the relations used in the ontologies or databases of the participants

– Determine which of those are identical in intended meaning, and whether those that appear to be identical can be organized in a common ontology, allowing alternate means of expressing the same intended meaning, and providing translations among the alternate expressions. This common set of concepts forms the starting core.

89

FO Development by the Consortium (contd.)

– For elements that are not identical in meaning, determine whether they can be added to the base common FO without creating logical inconsistencies.

– If logical inconsistencies are detected, determine if the creators/users of the inconsistent ontologies can agree on a consistent representation adequate for their purposes; if so, add that to the FO; if not, determine how to describe their differing elements using the common FO, and add those elements as parts of mutually inconsistent extensions to the FO.

90


– For those elements (types and relations) that are not identical in intended meaning, determine if they can be logically specified (described or expressed) as combinations of the elements already in the FO. If not, they can be added to the FO as additional primitives, if so they can be added to a mid-level or domain extension – or retained in the FO if there are no objections.

• The resulting FO will be the first version, to be tested for its ability to support interoperability.

• This version will be refined as experience by the consortium using the FO in practical applications demonstrates the need for modification.

91


• In addition to the translation axioms that convert different ways of expressing the same concept into each other, it will be necessary to develop a translation program that can take information in one local form and translate it into the other form.– This will require a means to recognize when

independently developed logical specifications of local elements (types, relations) are identical

92

FO Utilities and Applications• Utilities likely to be required for wide adoption:

– An extraction utility that can extract out only that part of the FO that is needed for a particular application

• A similar principle is used in the NIEM information exchange system: Information Exchange Packet Descriptions (IEPDs) use only a part of the entire NIEM vocabulary to form messages used for specific purposes. A utility assists creation of such IEPDs.

• For the FO, the utility would start with the domain ontology extension(s) required for a particular application, and extract only the parts of the FO needed for reasoning with those elements.

93

FO Utilities and Applications (contd.)• A Natural Language Interface (NLI) is required that can

accept a NL description of a domain concept, using the English defining vocabulary for the ontology , and:

• Tell whether that concept is already represented• If not, create a logical representation of the needed

concept, and a text description of the logical representation, for verification by the user.

• The NLI can be structured in modules (see next slide) to allow incremental evolution of an increasingly effective assistant for understanding and using the FO. The NLI can be developed by the efforts of multiple groups, each focusing on one or a few modules.

94

95

Foundation Ontology:Provides defining concepts to specify

conceptual message Content

Architecture for the Natural Language Interface to the FO

Parsing Disambiguation LearningEntity

ExtractionMetaphoricReasoning

Knowledge Base for the NLI: Includes specialized

knowledge of the structureOf the Foundation Ontology

Interfaces to Text Input and Output

WordExperts

NLP Understandingand generation

TC

FO Utilities and Applications (contd.)• Possible Example Applications:

– Database Integration, demonstrating federated search with reasoning using elements from more than one RDB.

– Natural Language Understanding, limited to conversational capability with a 6-year old native English speaker, and including the capability of the NL interface, allowing users to query the ontology.

– A biomedical application, possibly demonstrating integration of patient medical information with other attributes of the patients, outcomes, and functions of the providers.

96

The Consortium ProcessWill commence after preliminary discussion results in:• A clear statement of the objective• A timetable and measures of progress• A procedure for rapid resolution of disputes

(anticipating that consensus will often not be possible, in which case quick voting will be required)

• Will not aim for a formal standard, but a de facto open-source standard accepted for use within the consortium community, and freely available to the public.

97

Toward the Future• The potential for widespread agreement on a common Foundation

Ontology presents an opportunity to develop a tool that can substantially accelerate progress in developing intelligent applications by allowing multiple processes or applications to communicate accurately. This will allow rapid evolution of any application that consists of multiple modules.

• Development of new computational techniques for information processing will be accelerated by allowing more effective reuse of routines that use the same standard of meaning.

• Adequate funding is required to allow development of a widely acceptable foundation ontology by representatives of a broad range of perspectives. At minimum, funding for at least 100 participants may require over 10 million dollars per year for 1 to 3 years, to develop the FO as well as the suggested utilities and applications, and to demostrate how the FO supports interoperability.

98

END

• COSMO ontology:– http://micra.com/COSMO– Email: [email protected]

• Acknowledgement:– Refinement and clarification of details of the

project suggested here have benefited from many discussions with the members of the Ontolog forum.

99

Semantic Integration: References

• IKRIS Scenarios Inter-Theory (ISIT)– Jerry Hobbs with the KRIS Scenarios Working Group– http://nrrc.mitre.org/NRRC/Docs_Data/ikris/ISIT_spec.pdf

• T• Y• Y• Y• Y

100

Popular Semantic Web domain ontologies (Swoogle, July 2005)

Ontology prefix Namespace URI # of Docs. Populated

rdf http://www.w3.org/1999/02/22-rdf-syntax-ns# 382K

rdfs http://www.w3.org/2000/01/rdf-schema# 82K

owl http://www.w3.org/2002/07/owl# 64K

daml http://www.w3.org/2001/03/daml+oil# 5K

dc http://purl.org/dc/elements/1.1/ 250Krss http://purl.org/rss/1.0/ 165K

admin http://webns.net/mvcb/ 130K

sy http://purl.org/rss/1.0/modules/syndication/ 90K

foaf http://xmlns.com/foaf/0.1/ 77K

cc http://web.resource.org/cc/ 74Kcontent http://purl.org/rss/1.0/modules/content/ 60K

trackback http://madskills.com/public/xml/rss/module/trackback 56K

iw http://inferenceweb.stanford.edu/2004/05/iw.owl# 47K

bio http://purl.org/vocab/bio/0.1/ 35K

geo http://www.w3.org/2003/01/geo/wgs84 pos# 25K

vCard http://www.w3.org/2001/vcard-rdf/3.0# 20K101

History of COSMO• Started by including all common terms in OpenCyc and

SUMO (+MLO)• Added parents of the common terms• Added terms from DOLCE and BFO not in either• Added types and relations to map database tables to

the ontology• At revision 589 started supplementing with missing

Longman defining terms; beginning statistics:– 3659 types (OWL classes)– 362 relations (OWL properties)– 414 restrictions• Current revision (851) 6803 types, 770 relations:Added 3150 types and 400 relations to COSMO while adding

representations of 1300 Longman defining words

102

Skepticism• “We cannot get everyone to agree on a single

foundation ontology”– We don’t need everyone, just a self-sustaining community

• “We don’t need another foundation ontology”– The fact that none has gained a critical mass of users

demonstrates that we do need another one, but one that is constructed by a very wide community of users.

– The COSMO is not the common FO, but is being used to demonstrate that a common Foundation Ontology is technically feasible, if funding is available.

• There is no limited ‘conceptual defining vocabulary’– Implies an unlimited number of primitive concepts; this is

susceptible to experimental refutation, and the COSMO project is designed to test this question

103

Distracting Terminology Issues

• Concept: a unit of thought or of automated information processing – not necessarily an abstract mental object. Ontologies are composed of ontology elements (“ontelms”) that represent such entities: see next slide.

• Definition: A description of the meaning – not necessary and sufficient conditions; to specify the meaning of (in words or logic)

• Meaning: an interpretation that approximates human-level understanding (see later slide)

• Understanding : conversion to a logical representation of the meaning

104

Consistent Data Acquisition Up-Front ?

Lynn Vogel: It’s also the fact that, in many cases, if you structure data up front, it will take you a bit longer to collect it. You could argue that once you’ve collected it, you have this fabulous treasure trove of structured data that can advance the science of what we do. But there’s an overhead for the individual clinicians who are collecting the data. They’re already under enormous pressure regarding the amount of time they spend with their patients. If you say, “Oh, by the way, for every patient that you see now, we’re adding 15 minutes so you can structure your data so that we all will be smarter,” that’s a pretty hard sell

• Lynn Vogel is vice president and CIO of The University of Texas M. D. Anderson Cancer Center. In addition, he holds a faculty appointment at The University of Texas in Bioinformatics and Computational Biology

• (from Price Waterhouse Cooper report on Semantic Technology 2009)

105

Words, Concepts, Ontelms

• Words are not Concepts. The elements in an ontology (types, relations, functions, axioms, instances) are neither “concepts” nor words, but language-independent logical structures. The meanings of the ontology elements do not change, but the words used to refer to them may change rapidly and vary with user.

• To avoid distracting terminology discussion, these are referred to as “ontelms” (ontology elements) in this presentation.

106

109

<owl:Class rdf:ID="ConceptualWork"> <rdfs:subClassOf rdf:resource="#AbstractSymbolicObject"/> <rdfs:comment>In COSMO a 'ConceptualWork' (a MentalObject) is classified as an AbstractSymbolicObject, since such works are always created in symbols, though the symbols may have information content – the 'meaning'. COSMO differs somewhat from the Cyc description in that we consider Codes to be included, but have a different usage of the term 'Code'. Cyc: OPENCYC 1: MAY 23, 2002 The collection of abstract works which are the deliberate creations of one or more individuals working in concert, have instantiations [#$instantiationOfCW] which are #$InformationBearingThings, and associated #$AbstractInformationStructures. This is a specialization of #$DevisedPracticeOrWork [q.v.]. For works with propositional content ; see the more specific collection, #$PropositionalConceptualWork (PCW). Positive examples include: #$MobyDickNovel (as opposed to any instances of #$BookCopy such that (#$instantiationOfCW #$MobyDickNovel BOOK_COPY)), Beethoven's 9th Symphony (as opposed to any performance of this symphony or any copy of its score). Negative examples include: games (performances are not IBTs), awards (they do not have associated #$AbstractInformationStructures), paintings (not abstract), customs (not deliberate creations), natural languages (not a deliberate creation), and codes (their uses, not instantiations, are IBTs). </rdfs:comment> </owl:Class>

Guo’s Longman Analysis• Guo, Cheng-ming (1989) Constructing a machine-

tractable dictionary from "Longman Dictionary of Contemporary English" (Ph. D. Thesis), New Mexico State University.

• Guo, Cheng-ming (editor) Machine Tractable Dictionaries: Design and Construction, Ablex Publishing Co., Norwood NJ (1995)

• Yorick Wilks, Brian Slator, and Louise Guthrie, Electric

Words: Dictionaries, Computers, and Meanings, MIT Press, Cambridge Mass (1996).

110

Words, Concepts, Representations

• Words are not Concepts• Concept: a unit of thought or reasoning

– (from Random House Webster)– 1. a general notion or idea; conception.– 2. an idea of something formed by mentally combining all its

characteristics or particulars; a construct.– 3. a directly conceived or intuited object of thought.

• In an ontology a “concept” is only that which is represented by the elements of the ontology (types, relations, instances, rules, functions). These are the things that are manipulated by a reasoning system

• The “representandum”• Words are not representanda.

111

Words Label Concepts

• Ambiguity: the same word labels multiple concepts

• Synonymy: more than one word labels the same concept

• Context-sensitive usage: the same word in different contexts can label different concepts

• An ontology organizes representations of concepts – mapping to words is a different task.

112

113

Concepts vs. WordsMathematical Theory / | \ / \ \ / | \ / \ | \ \ | \ /

Axioms:(Every TypicalCat has ((exactly 4) Legs))

(Every House has ((atLeast 1) Door))

HouseCat

Siamese

OntologicalTheory (Meanings) Terminology

“Siamese Cat”

“Residential House”

“Haus”

“chat siamois”“Siamesische Katze”

“House”

“maison”

“Siamese feline”“Siamese”

“дом”

シャム猫

114

Contexts

• Q: Isn’t context important?

• A: Very. Existing ontologies have modules, contexts, or similar mechanisms (“microtheories”). More elaborate contextual reasoning may be necessary.

Some Primitives fromWierzbicka

• I, YOU, SOMEONE, SOMETHING, THIS,• THE SAME, THINK, WANT, KNOW, SAY, DO,

HAPPEN, GOOD, BAD, WHEN/TIME,• WHERE/PLACE, BECAUSE, NOT, MAYBE, LIKE,

KIND OF, PART OF.

115

Meaning via Human Interpretation• Nirenburg and Raskin: Ontological Semantics MIT

Press, 2004– “Meaning should be studied and represented”– Meaning needs to be “anchored in extralinguistic reality”

but the “verificationist premise” of Procedural Semantics is not shared

– In Ontological Semantics meaning is intensional. “… meaning is a statement in the Text-Meaning Representation (TMR) language ”

– “The connection between the outside world … and Ontological Semantics … is carried out through the mediation of the human acquirer of the static knowledge sources.”

116

Longman Definitions: “obligation”• See: http://www.ldoceonline.com/• Obligation: a moral or legal duty to do something• Duty: something that you have to do because it is

morally or legally right • Have to: if you have to do something, you must do it

because it is necessary or because someone makes you do it.

• Must: to have to do something because it is necessary or important, or because of a law or order

• Necessary: something that is necessary is what you need to have or need to do

117

COSMO: “obligation”• <owl:Class rdf:ID="Obligation">• <rdfs:comment>A MentalObject that refers to some FutureSituation that the Agent having the Obligation may cause to happen or may refrain from doing; if the Agent does not

perform an Action to cause the FutureSituation to occur, then some negative consequence is likely to be incurred for failure to perform the Obligation. . The type of negative consequence (legal punishment, social condemnation, eternal damnation, pangs of conscience, being grounded by one's parents) will be characteristic of different types of Obligation.. Each Obligation is assigned by some Authority, which could be a person‘s own conscience (reflecting learned social mores), or the mores of the community. In the case of a Debt, the Authority may be the person owing the debt and the person to whom the debt is owed, if the debt arises from some agreement or transaction.

• An Obligation may be created in an ObligationCreatingEvent (which see). The notion of 'Obligation' is too primitive to be easily described by simple relations. In essence, an 'obligation' is a relation of an Agent to an Event that is derived from a belief

about what kind of behavior is best in a situation. The exact formalization of this notion is still incomplete as of 0.49. See also 'ResponsibilitySituation' for a closely related concept. Linguistically an Obligation is expressed in several ways:• 'Tom has an obligation to do X'• 'Tom is obliged to do X'• 'Tom has a duty to do X'• 'Doing X is Tom's (obligation/duty).'• 'Tom ought to do X'• 'Tom must do X'• 'Tom should do X'• 'Tom is responsible for doing X'• Similar phrases may be used to express an action that is not an Obligation, but is a prerequisite for some desired situation: 'In order to get into college, Tom must get good grades.'• The linguistic analyzer must recognize the discourse relations that distinguish obligations from prerequisites. The type 'Obligation' in COSMO represents only true Obligations.• • Each instance of Obligation will represent an Action that the agent with the Obligation is obliged to perform or refrain from. When expressed linguistically, that action will be

prefaced by the word 'to', e.g. 'to drive no faster than 60 miles per hour'.• • Cyc: A collection of microtheories; a subcollection of #$SupposedToBeMicrotheory. Each instance of the collection #$Obligation is a microtheory which contains assertions

describing what some agent (the #$obligatedAgents) is obliged to do, or make true, for one or more other agents, possibly including society in general. An obligation is the most general case of some agent owing something to another. Obligations may be undertaken in conjunction with various kinds of #$Agreements. Unlike an agreement, however, an obligation need not have a second known party (though some do). An obligation can exist and be understood without identifying another particular agent as the 'holder' of the obligation - and that may be true, even if the beneficiary (#$obligationOwedTo) can be identified. For example, assuming that parents have an obligation to care for their children, it is not clear with whom a parent has 'agreed' to take care of his or her child. Some common ways to incur an obligation are through social transactions (e.g. family duties, friendship, favors) or through financial transactions (e.g. a #$PaymentObligation). In addition, obligations may be imposed on those who are subject to one or more instances of #$CodeOfConduct, e.g. #$SportsRulesOf-BoxingSportsEvent or #$OfficeCodeOfConductMt.

• • Corresponds to senses 2 and 3 and part of sense 1 of 'obligation' and sense 2 of 'duty' in WordNet:• NOTE that sense 2 is a state, and linguistically would be expressed by a phrase like 'under an obligation', rather than the word 'obligation' itself. Sense 3 should be a subtype,but is

not yet represented. • 1. (14) duty, responsibility, obligation - (the social force that binds you to the courses of action demanded by that force; 'we must instill a sense of duty in our children'; 'every right

implies a responsibility; every opportunity, an obligation; every possession, a duty'- John D. Rockefeller Jr)• 2. obligation - (the state of being obligated to do or pay something; 'he is under an obligation to finish the job')• 3. obligation, indebtedness - (a personal relation in which one is indebted for a service or favor) </rdfs:comment>

• (continued . . . .)118

COSMO: “obligation” (continued)

<guid>bd58bfd0-9c29-11b1-9dad-c379636f7270</guid> <rdfs:subClassOf rdf:resource="#Responsibility"/> <rdfs:subClassOf rdf:resource="#RulesForConduct"/> <rdfs:subClassOf rdf:resource="#NeedMicrotheory"/> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="#wasAssignedByAuthority"/> <owl:someValuesFrom rdf:resource="#Authority"/> </owl:Restriction> </rdfs:subClassOf> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="#refersToExternalEntity"/> <owl:someValuesFrom rdf:resource="#FutureSituation"/> </owl:Restriction> </rdfs:subClassOf> <wordnet>obligation</wordnet> <wnsense>obligation1n</wnsense> <wnsense>obligation2n</wnsense> <wnsense>obligation3n</wnsense> <wordnet>duty</wordnet> <wnsense>duty2n</wnsense> </owl:Class>

119

120

Specialists Will Want to Use Specialized Terms in Definitions

• The “Controlled Defining Vocabulary” is infinitely expandable.

• Probably, at least three levels will emerge:– the basic irreducible defining vocabulary– the general defining vocabulary, having common

terms which are defined (linguistically) by use of the basic vocabulary, and specified logically within the FO or some extension as FOL combinations of the ontelms in the FO.

– specialized defining vocabularies, containing terms of interest to specific domains, definable by use of the general defining vocabulary

121

Definition Acceptance HierarchyExecutable Specification: Methods, Sequence, States

Axiomatic Ontology: Quasi-2nd Order, Function Terms

OpenCycSUMO DOLCE

Restricted FOL: OWL

Taxonomy/Thesaurus/Terminologyaccepts

accepts

is used in

Economic Benefits of Semantic InteroperabilityThe practical significance of semantic interoperability has been measured by

several studies that estimate the cost (in lost efficiency) due to lack of semantic interoperability. One study[1], focusing on the lost efficiency in the communication of healthcare information, estimated that US$77.8 billion per year could be saved by implementing an effective interoperability standard in that area. Other studies, of the construction industry[2] and of the automobile manufacturing supply chain[3], estimate costs of over US$10 billion per year due to lack of semantic interoperability in those industries. In total these numbers can be extrapolated to indicate that well over US$100 billion per year is lost because of the lack of a widely used semantic interoperability standard in the US alone.

[1] Jan Walker, Eric Pan, Douglas Johnston, Julia Adler-Milstein, David W. Bates and Blackford Middleton, The Value of Healthcare Information Exchange and Interoperability Health Affairs, 19 January 2005

[2] http://www.bfrl.nist.gov/oae/publications/gcrs/04867.pdf[3] http://www.nist.gov/director/prog-ofc/report99-1.pdf

122

Zipfean Distribution of Characters?

Characters % of Text• 640 80• 1569 94• 3234 99.2• 4562 99.84• 5659 99.968First 80% of text requires fewer Chinese characters

than each succeeding 80% of each remainder.123

Zipfean Distribution of Characters?

Characters % of Text• 1153 90• 3050 99• 4926 99.9• 6223 99.99

124

the foundation ontology as a basis for semantic interoperability patrick cassidy

Documents

common foundation ontology

common vocabulary standard

common standard

ontology languages

fo tactic

semantic interoperation

accurate interpretation

domain ontology