Ontology Summit 2018 Synthesis
This page shows the open research problems and issues that were identified in the summit. This is an active page, and participants are encouraged to add to the page. Please focus on questions and problems. While parts of a solution can be mentioned, they cannot be the focus.
The following blogs are available for linking the problems below with the Summit Sessions as well as for recording discussions about the problems.
Meaning of Context
The word "context" comes from "con" meaning together and "text". The word "text", in turn, comes from "texere" meaning woven. So "text" is something woven, while "context" is something that is woven together (compare with the word "texture" which also comes from "texere").
The usual dictionary definition of context is "the circumstances that form the setting for an event, statement, or idea, and in terms of which it can be fully understood and assessed."
A Spectrum of Ways to Specify the Context of Knowledge
Ontologies are built using particular perspectives which may not be shared. Thus we need a way to qualify and specify what is meant by context and its associated perspective of an ontology, but really for any bit of knowledge no matter how formal it is. Another way of saying this is that we can also say something "about" an object or interest. In a simple formulation we may think of adding something about an object of attention as metadata about that object. We connect to ontologies as part of this view because a particular ontology which tells us something important about some domain which is the scope of an ontologies coverage. So an ontology might be used to express some background knowledge about some topic, data, object etc. But in turn there may be a context for an ontology by which we mean that we can say something about it, outside of what the ontology says itself. We note that different Summit speakers have expressed somewhat different views about what candidates for relevant context of an ontology might consist of or what the context on an ontology might be. But if we view context as a meta-level about something, call it the open level then we can ask what metadata is appropriate to discuss an object as well as what expressiveness may be required. Examples include the idea that a domain, like a medical illness could and should be modeled differently depending on the contextual view in which it is considered. Thus a view of illness might depend on a spatio-temporal coordinate (1st world vs. 3rd world), the thematic focus, a subjective perspective of agent (patient or therapist), some adopted level of granularity for the representation, and the intended application of the ontology. Both lightweight, metadata oriented methods using RDF to specify the context of RDF statements as well as the richer formal mechanism have been discussed as part of one topic of this Summit - an Open Knowledge Network (OKN). The OKN vision includes the development of knowledge graphs formed by extracting information from the WWW and the Semantic Well. A rich approach to developing open knowledge might specify a context using McCarthy's approach which treats contexts as formal objects over which one can quantify and express 1st-order properties. For example in the Cyc ontology time (temporal context) can modeled (and implemented) as a dimension of contexts using a particular interval semantics for temporary truth. Few ontologies use the CYC knowledge base of MTs to date which may be contrasted with lightweight methods using RDF to express some context in triples. One particular approach called RDF+++ uses triples to express source, time, confidence and providence information about an RDF fact. As part of the Summit Breakout we will explore the relative merits of lightweight vs. heavier ontological approaches to provide context. We will assemble some medical bio-science scenarios and associated data and investigate contextual issues within them. This may help identify one domain specific needs of context which might be useful as part of planned Open Knowledge Network effort. We intend to leverage existing work in OKN, Cyc's microtheries as well as prior attempts to document ontology scope and aboutness using metadata Gary Berg-Cross
Qualification of Context in Application Ontologies
Some aspects of context can be expressed as metadata about Application ontologies using formalisms as simple as RDF. For example, we can express the fact that, "the PATO ontology hasDomain Phenotypes" or that it is at a particular URL, or that it is version 1.2 etc.
But this may not be a rich enough view of context for many application ontologies. In these cases more background knowledge may be what needed to make application smart and for this some deeper ontologizing of the domain may be needed. An example might be better axiomization of what contexts in which certain objects or processes play particular roles or take on critical attribute. This view may help understand some discussion as part of the Summit that:
"The issue for the development and application of ontologies is that context is seldom explicitly stated." GaryBergCross
Context may be hinted but not explicit. Indeed, there is no wide agreement on the mechanism for specifying a context, and the context is often informally presented. CoryCasanave It is a challenge to make explicit the relevant contexts for ontology development and application. Most developers are not even consciously aware ofthe contexts they are working within. ToddSchneider
Ontology of Context
Since context may mean an indefinite number of things there is no definitive ontology of context. JackRing
But as PatHayes points out there may be a simple classification of contexts.
There are four kinds of context in linguistics:
- the text or discourse;
- the situation;
- common background knowledge; and
- the intentions of the participants. JohnSowa
There are contexts for agents and contexts for text. ToddSchneider
This is a fundamental research problem.
Consider the ontology of context versus the context of ontology. By this I (MikeBennett) mean, there are contextual statements in a given ontology, that make explicit the situation object contexts implicit in data models e.g. roles, temporality, occurrent (process/activity) etc. GaryBergCross suggests that it may be better to think of this as the internal context of an ontology than an "ontology of context."
Then there is the context in which the ontology is understood or deployed. (MikeBennett)
Context as Ontology
Closely related with the ontology of context is the issue of how one represents a context.
The relationship between an ontology and a context being used for interpreting the ontology is important and must be explicit and formal. Even if both the ontology and context are formally defined, it is not enough for the relationship between them to be defined by informal techniques such as the use of the same natural language terms in the ontology and context.
Upper ontologies are one approach to representing contexts.
The HTemp ontology (hierarchy of templates) that represents the second ontology of the NKRL (Narrative Knowledge Representation Language) system – the first ontology is the « standard » ontology of concepts, HClass, hierarchy of classes – represents a (may be limited, but effective) example of ontology of context. Each one of the (about 150) templates included in HTemp denotes in fact, in a formal way, a specific category of events, states, actions, behaviours… that can find a concrete instantiation in the real world. For example, a trip of John from Paris to London at a certain date is represented as an instance of the Template « autonomous displacement of a human being » where all the details of the trip – i.e., its context (character(s) involved, departure point, arrival point, date, modality like train or flight etc.) – are accurately registered. A recent description of the characteristics of NKRL can be found in my 2018 paper « Functional and Semantic Roles in a High-Level Knowledge Representation Language. Artificial Intelligence Review, doi: 10.1007/s10462-017-9571-5 ». Gian Piero ZARRI
Environment and Context
Environment and context are conflated in the systems world. Both are what outside a field or domain impinges on the central system. There should be a means for distinguishing them. JackRing JanetSinger
One may, of course, think of a system as part of a larger system which is a "system environment" in which the original system exists. But we may also distinguish environments as associated with surroundings or conditions in which a person, animal, or plant lives or operates. So it is often used to discuss systemetic relations in the evolved world.
Competency Questions as Context
Competency questions may provide some qualification of context, really scope, of an ontology. Looking at the competency questions a person can tell what an ontology is about. It provides a view of scope. GaryBergCross Competency questions are generally used as part of ontology engineering for checking that an ontology satisfies the requirements of interest the domain experts who are being used to develop the ontology. Competency questions constrain the design of an ontology. However, the competency questions are in the same context as the ontology, so it is unclear how they could specify the context in a formal manner. ToddSchneider
One view of the research problem is to clarify the role, if any, that competency questions can play in specifying a context. Another way, perhaps of saying this is, "how can competency questions become metadata to describe the 'aboutness' of an ontology?"
Context and Tarskian Truth
How do contexts relate to a Tarskian view of truth value? ToddSchneider
To some degree John Sowa has addresses this issue in his article on context http://www.jfsowa.com/ontology/contexts.htm : "To simplify metalevel reasoning, Tarski advocated a method of separating or stratifying the metalevels and the object level. If the object language L0 refers to entities in a universe of discourse D, the metalanguage L1 refers to the symbols of L0 and their relationships to D. The metalanguage is still first order, but its universe of discourse is enlarged from D to L0ÈD. The metametalanguage L2 is also first order, but its universe of discourse is L1ÈL0ÈD. To avoid paradoxes, Tarski insisted that no metalanguage Ln could refer to its own symbols, but it could refer to the symbols or the domain of any language Li where 0£i<n.
In short, metalevel reasoning is first-order reasoning about the way statements may be sorted into contexts. After the sorting has been done, the propositions in a context can be handled by the usual FOL rules. At every level of the Tarski hierarchy of metalanguages, the reasoning process is governed by first-order rules. But first-order reasoning in language Ln has the effect of higher-order or modal reasoning for every language below n. At every level n, the model theory that justifies the reasoning in Ln is a conventional first-order Tarskian theory, since the nature of the objects in the domain Dn is irrelevant to the rules that apply to Ln." (inserted by GaryBergCross)
What is a sufficient level of context description for operational purposes? ToddSchneider This may be based on requirements and thus as part of ontological engineering would be part of the competency question development with domain experts. GaryBergCross What 'aspects' of context need to be made explicit in order to facilitate interoperability? ToddSchneider Specific data interoperability may be handled by making sure that the semantics of the targeting data which will interoperate are part of a common conceptual model that has been formalized in a compatible fashion. GaryBergCross
Scope and Context
What is the relationship between scope and context?
Use cases and competency questions may give some idea of the intended scope. See the competency question section for more detail of how scope is related to context. One can also explicitly state what is outside of scope because it is not handled as part of competency. Making competency questions part of the metadata of an ontology is a research question and it is not clear how such an effort might be general and useful.
Time as Context
There are several ways to introduce time to an ontology: 3D, 3D+1, 4D. The importance of the temporal dimension for domain like medicine has been well recognized and there are numerous models including the Time Event Ontology (TEO). The RDF++ effort explicitly addresses capturing time in an RDF form to document RDF facts.
Simple question, how do acronyms represent context? ToddSchneider Simple answer, poorly since they aren't good at formal semantics. You can ask what role they may play since efforts to extract knowledge from text have to resolve similar terms. That's a well studied area although difficult since language may be used imprecisely.
Relevance of Linguistics
How much of the work done in linguistics relative to context is relevant to the development or uses of ontologies (in information systems)? ToddSchneider
The challenge is drawing the big lessons from these fields and integrating them in a general (rough) framework. JanetSinger
Machine learning to extract information from text is discussed as part of OKN. GaryBergCross
This is not exactly the same as evaluation of an ontology (as in Ontology Summit 2013) which is focused on self-consistency and satisfaction of requirements (as specified by use cases and competency questions). These are important and essential for ontology development. However, it is undeniable that a great many ontologies have been developed that use state of the art development processes and that satisfy requirements and fitness criteria, yet are not being used.
How do (implicit) context(s) impact the (potential) uses of an ontology? ToddSchneider
Is the 'efficacy' of an ontology equivalent to the entailments that can be drawn from it? ToddSchneider
You might ask what data interoperability can it support or how easy is it to integrate with another ontology. But the questions may be indefinitely extended. GaryBergCross
If you think of ontologies as nesting/holonic, with levels of abstraction, an 'upper' ontology may be decomposed into different contexts in sub-ontologies. Or one may think of related ontologies such as in the medical domain which include an anatomy ontology like the Common Anatomy Reference Ontology (CARO). CARO is related to other biomedical ontologies. They may share a super structure that has been harmonized and includes a concept of "Biological Entity." CARO includes relations to critical external ontology terms from Basic Formal Ontology (BFO).GaryBergCross
However, in cases of independently developed sub-ontologies they may seemingly contradict each other. Is this okay? Do they really conflict at more detailed levels? BobbinTeegarden
Situations allow seemingly contradictory statements so long as they are in different situations. In one situation a cat is alive and in another situation the same cat is dead.
Figure-ground relationships have been explored/discussed in art (Escher is an example). Will it be necessary to describe the context for both is a question raised by TerryLongstreth
Figure-ground phenomena can be discussed in terms of such things as sensory fusion and perceptual framing which in part involves the use of background knowledge. So for example if an image stimulus evokes one feature in an perceived image it might further evoke the idea of "women's image." (Following DOLCE and the Semantic Sensor Net an observation can be seen as invoking first a quale in an observer’s mind. ) What gets evoked varies from person to person based in there experience and thus is hard to generalize. Ontologies like DOLCE are grounded on the idea that properties do not exist independently, but depend on other entities, which are the so-called features of interest. So there may be many ways of interpreting an image. The eyes of a women's face might be such a feature that attracts attention.
Contexts and Actions
Ontologies are used in systems, be they people communicating with natural language or computer systems that are interoperating. Context is necessary for the elements of an ontology to be interpreted by a system, but once the elements have been interpreted, some action is taken by the system. How did the context influence the action? (Q issue from Gary - a formal ontology may be processed by a "system" using logic and not need further interpreatation. That is one of its advantages. It may be that human reasoning may take into account more context and we can imagine smart systems using other ontologies for more background knowledge, but this is not yet the general case of how ontologies are used, say in medicine.)
What has clarified for me is the relationship between context, as something that impacts an interpretation, and the interpretation "act". Any such act is "closed" in the set of "text" that impacted the interpretation. CoryCasanave Can one have a context without an agent or class of agents? ToddSchneider
One part of a context for action would be the idea of what action is afforded. This idea is hightlighted in Rome, Erich, and Georg Dorffner. Towards affordance-based robot control. Springer-Verlag Berlin Heidelberg, 2008. A core idea is cited below. Note the connection to Sowa's idea that intentions are central to context. But this also starts to address this very difficult idea of who a cognitive agent perceives and interacts with the environment. "Perceiving affordances in the environment means perception as filtered through the individual capabilities for physical action and through the current goals or intentions, thereby coupling perception and action deep down in the control architecture and providing an action-oriented interpretation of percepts in real time. Moreover, affordances provide on a high granularity level a basis for agent interaction and for learning or adapting context-dependent, goal-directed action."