Although there have been e orts to integrate Semantic Web technologies and Cognitive Science approaches, they are still relatively isolated. Herein we introduce a new ontology framework supporting representation and reasoning on complex cognitive information using Semantic Web tools. The framework consists of four parts: an event ontology for information about actions and events, an epistemic ontology containing facts about knowledge, belief, perception and communication, an ontology about future intentions and desires, and, nally, a deontic ontology for modeling obligations. The structure of the ontology framework is derived from the syntax of the Deontic Cognitive Event Calculus.
One of the most di cult problems of arti cial intelligence is the autonomous
action of arti cial agents. One of possible approaches to this problem is inspired
from common-sense psychology (CSP) and mentalistic models of human behavior.
Knowledge representation models are called mentalistic if they are related to
mental notions like knowledge, belief, goals etc. [
When an arti cial agent in such an environment communicates some information, either publicly or privately to another agent, this information cannot be taken at face value because it only represents a belief or can be even intentionally misleading. Such an account of communication requires mentalistic models distinguishing belief from knowledge.
An arti cial agent also needs to reason about the consequences of his own intended actions. On the top of that, reasoning about actions of other agents and predicted consequences of such actions is also required. In other words, models enabling such reasoning have to deal with past, current and future actions and generally with events, so exible handling of time is necessary.
Dealing with future events caused by actions of other agents requires
mentalistic models of their internal mental representations of future { their desires and
intentions. We call these future-involving mental states protential, based on the
term `protention', meaning the consciousness of future and coined by the
philosopher E. Husserl [
The last level of mentalistic model contains the representation of obligations.
These are limitations of behavior of the reasoning agent himself and of other
agents, which can be used for predicting their actions. Obligations require an
agent to act (or suspend action) when some de ned conditions of the environment
are observed. We call the model involving obligations deontic, based on similar
use of the term in the context of deontic logic [
Note that even if we use metalistic models and discus an agent's behavior in mentalist terms, this does not mean that such an agent would be required to have genuine mental states; accepting a thoroughly instrumentalist view of mental states of arti cial agents is su cient for our purposes.
We can now summarize the requirements for (or, directly, components of) a minimal model enabling arti cial agents' reasoning in the above de ned context: Model of events and actions Epistemic mentalistic model Protential mentalistic model
Deontic mentalistic model
As we have said above, there have only been very limited e orts to use
mentalistic approach in the Semantic Web context. However there has been a number
of approaches to model events in general, on the semantic web; for a most recent
overview see [
The Section 2 of the paper provides a general overview of the architecture of the proposed framework. Section 3 describes the Deontic Cognitive Event Calculus
Temporal Maintenance Perception Processor
Motoric Processor
SWRL Rule engine
DL reasoner (DCE C ). Section 4 describes the proposed ontology framework itself. Section 5 discusses open problems and possible future extensions of the framework. Finally, the last section provides some conclusions.
The overview of a hypothetical application architecture making use of the
proposed Deontic Cognitive Event Ontology (DCE O) is depicted in Fig. 1. We have
been inspired by the Soar cognitive architecture [
There are however some di erences. The core of our architecture consists in the Tbox of an OWL ontology (described in this paper) and a set of SWRL rules. These represent stable knowledge about the world. Information obtained from the environment is transformed by a component that is traditionally called Perception processor and, in our case, produces OWL statements to be included in the Abox of the ontology. On the top of the whole ontology, consisting of a Tbox, Abox and SWRL rules, there is a SWRL rule engine and a DL reasoner, which are responsible for inferential tasks. Another component traditionally called Motoric Processor monitors the current time and retrieves the statements representing actions that are to be performed at the given time point. The last component, which we call Temporal Maintenance, carries out auxiliary operations such as removing old Abox axioms from the data store, thus enabling the whole system to function e ciently. Namely, the growth of the number of axioms caused by continuous addition of new statements to the Abox combined with axiom production by the SWRL rule engine might be enormous; clearing of old data representing no-longer-useful knowledge would thus be necessary.
While the overall architecture is inspired by the standard Model Human
Processor, the internal structure of the ontology is based on the Deontic Cognitive
Event Calculus (DCE C ), which has been successfully used in a number of
realworld scenarios such as reasoning over a scene description [
S ::= OMbojmecetnjt AjgBeonotlejaSneljfF@luAengtenjtNjuAmcetriiocnType j Action v Event j action : Agent ActionType ! Action initially : Fluent ! Boolean holds : Fluent Moment ! Boolean happens : Event Moment ! Boolean clipped : Moment Fluent Moment ! Boolean f ::= initiates : Event Fluent Moment ! Boolean terminates : Event Fluent Moment ! Boolean prior : Moment Moment ! Boolean interval : Moment Boolean * : Agent ! Self payo : Agent ActionType
Moment ! Numeric t ::= x : S j c : S j f (t1; :::; tn)
p : Boolean j : j ^ j _ j ! j $ j 8x : S: j 9x : S: j
::= PB((aa;; tt;; )) jj KD((aa;; tt;; ho)ljdsC((ft;;t0)))jjSI((aa;;bt;; th;ap)pejnSs((aa;ctt;io)n(ja ; ); t0)) j
O(a; t; ; happens(action(a ; ); t0))
3. Deontic Cognitive Event Calculus (DCE C )
Deontic Cognitive Event Calculus (DCE C ) is a multi-sorted quanti ed modal
logic developed at Rensselaer Polytechnic Institute, which has a well-de ned
syntax and a proof calculus. Detailed information about multi-sorted rst order logic
(MSL) can be found in a book by M. Manzano [
The proof calculus is based on natural deduction [
DCE C is based on the Event Calculus (EC), which was rst introduced by
Kowalski and Sergot as a logic programming formalism for representing events and
their e ects [
DCE C adapts three sorts from EC : Event, Moment and Fluent. The sort Boolean is only used to capture truth values. The following elements of the signature f of the syntax of DCE C are adapted from EC: initially, holds, happens, clipped, initiates, terminates and prior. For the relation prior, which introduces an orders over time points, EC sometimes uses the simple symbol <.
We will now brie y describe the components of DCE C . A Fluent is anything the value of which can change over time. Typically it is a truth value of a proposition (e.g., \Peter is student"). A uent can be also a numerical value of a property that is subject to variation, e.g., temperature, but such a value can be easily transformed to a proposition (e.g., \The temperature is between 5 and 10 degrees Celsius.") so EC usually con nes its focus to propositional uents.
A Moment is a point in time. Points in time are ordered by the relation prior. The expression prior(t1; t2) means that the time point t1 precedes the time point t2 (e.g., t1 = 01/01/2015 precedes t2 = 01/01/2016). The term holds(f; t) says that uent f holds at a given time t (e.g., f = \Peter is student" holds at t = 01/01/2016). The expression initially(f ) indicates that the uent f holds from time 0.
Several signature members describe relations between events and uents. The general idea of EC is that events cause changes of truth values of uents. The expression happens(e; t) thus informs that event e happened in time t (e.g., e = \Peter concluded his studies" at t = 05/03/2016). The expression terminates(e; f; t) states that following event e uent f ceased to hold at time t (e.g., after e = \Peter concluded his studies" at t = 05/03/2016, proposition f = \Peter is a student" was no longer true). Similarly, the expression initiates(e; f; t) states that after event e uent f started to hold at time t (e.g., after e = \Peter was inaugurated" at t = 02/03/2011, proposition f = \Peter is a student" started to be true). The expression clipped(t1; f; t2) says that uent f is terminated between time t1 and time t2 (e.g., f = \Peter is a student" is terminated between t1 = 01/01/2016 and t2 = 01/01/2017).
DCE C introduces a mechanism to deal with epistemic information on the top of the event conceptualization of EC. DCE C has a classical monotonic view of the agents' knowledge of the world. The knowledge possessed by agents is considered to be unchanging, so if an agent knows at some time t, then the agent will continue to know for all time. On the other hand, an agents' beliefs can change as time passes. This marks a fundamental di erence in understanding knowledge vs. belief in DCE C .
The epistemic predicate C(t; ) indicates common knowledge (possessed by all agents) of at time t. The predicate K(a; t; ) says that agent a knows at time t. The predicate B(a; t; ) says that agent a believes in at time t. Finally, the predicate P(a; t; ) says that agent a perceives at time t.
DCE C also introduces tools for capturing the communication of agents. The predicate S(a; b; t; ) describes the communication of information from agent a to agent b at time t. Public communication of information at time t by agent a is denoted as S(a; t; ).
There is another set of predicates, which we may call behavioral: the predicate
D(a; t; holds(f; t0)) says that agent a desires that uent f would hold at time t0.
Similarly, to say that agent a at time t intends to perform an action of type
at time t0, we use the predicate I(a; t; happens(action(a; ); t0)). These predicates
are based on work by Goble [
when payoffAgent Object Payoff payoffValue Agent self communicationTo Communication who
when what payoffAction who
typeOfAction Interval dateTimeFrom dateTimeTo Event Action ActionType Belief
Perception who
Knowledge indendsActionType
Intention obligedToActionType who Obligation includesMoment initiatedBy
Termination terminatedBy Initiation
when Moment dateTimeMoment now
when Hold truthValue
holds/ofFluent-1 when
Fluent initialy what fluentRelationFrom fluentRelationTo
FluentRelation relationType when CommonKnowledge
Desire who
what indendsWhen
desiresWhen when when when what obligedWhen terminatesWhat when initiatesWhat prior clippedFrom clippedTo what what
Finally, the deontic predicate O(a; t; ; happens(action(a; ); t0)) should be
interpreted according to authors of DCE C as: \If it is the case that a at time t
believes then that is obligatory for a and this [obligation] is known by a."
The semantics of this predicate is based on a study by Castan~eda [
These predicates also require the introduction of the sort ActionTypes covering the general types of action, and of the function action(a; ) = b, expressing that for a given agent a, an action type produces a speci c action b. The operator is used to point out the reasoning agent himself among other agents in the universe of discourse. The operator payo (a; ; t) is used to evaluate an action of type performed by agent a at time t; the result of such an evaluation is of the Numerical sort.
The proposed Deontic Cognitive Event Ontology (DCEO) will be described in detail in this section. Fig. 3 depicts the general overview of the ontology, while Table 1 summarizes the object properties of the ontology and their features (domain, aD1 = EpistemicState t CommonKnowledge t ProtentialState t Communication t Obligation bR1 = Holds t FluentRelation cE.g. one event can terminate several uents. dD2 = EpistemicState t ProtentialState t Communication t Obligation eActions are generally considered to be performed by individual agents. fWe suppose that every action has only one ActionType. gWe may have several clipping information items for a single uent. hDtto.
iA communication can be aimed to several receiving agents. range, symmetry, re exivity, transitivity and functionality). Speci c axioms for each part of the ontology will be listed at the end of each respective section of the following text. The ontology can be downloaded from our website.4
The foundation of the ontology provides the class Moment, which consists of time points ordered by the object property prior. The property prior has the class Moment as both its domain and its range. We have introduced an additional data property dateTimeMoment, which explicitly speci es precise the date and time for those moments where it is known.
The function interval of the calculus has been rei ed to the class Interval of the ontology, and we introduced two similar data properties as for previous
class, dateTimeFrom and dateTimeTo, which specify the precise date and time boundaries of an interval. The object property includesMoment connects intervals (members of class Interval) with moments (members of class Moment) that belong to it.
The class Fluent represents propositional entities that are subject to change. It has a boolean data property initially, which corresponds to the function initially of DCEC . The class Hold represents the rei ed function holds of the calculus. The class Fluent is connected to the class Hold by the object property holds. The class Hold is in turn connected to the class Moment by the object property when, and it also has a data property truthValue. This whole structure allows us to represent that a uent holds a truth value at a given moment. The same uent can hold truth value true at one moment and the truth value false at another.
This structure then can become the content of an epistemic state, for example, an agent can believe that a uent (proposition) has the truth value false in the given moment. The data property truthValue is assumed to be boolean (truth values represented by 0 or 1) as in DCEC ; however, using this ontology design it may be easily adapted to dealing with various types of uncertainty.
The class Event and its subclass Action do not have any data type properties, and together with the class Agent they constitute the core of the event-focused section of the ontology. The class Agent has one speci c boolean data property, self. This corresponds to the function * and the sort Self of DCEC . The ontology in use may contain and describe number of di erent agents { only one of them is the reasoning agent himself and this property is used to represent this knowledge. Obviously only one agent can have this property true.
We have also introduced three general object properties: what, who and when. These represent various properties of DCEC with common ranges. The range of the object property what are classes Hold and FluentRelation: aside of epistemic states directed towards simple propositional uents (represented by the ontological structure around the class Hold described above) there may be epistemic states about relations between uents. The class FluentRelation has two related object properties fluentRelationFrom and fluentRelationTo, both connected to uents, and one data property fluentRelationType denoting the speci c type of relation between uents.
The range of the object property who is the class Agent. The property usually denotes the subject of an epistemic state or the active agent of a communication or of an action.
The range of the object property when is the class Moment, denoting the time of events, actions, epistemic or protential states throughout the ontology. Note that this property denotes the occasion of the actual occurrence of a protential state: the occasion when, e.g., the intention is mentally present. The protential state also has the second speci cation of time { the future time when something is intended (e.g., the agent now at time t1 intends to move to point X later at time t2). This second time speci cation is denoted by speci c separate object properties (see Sec. 4.3).
The classes Termination and Initiation represent the functions terminates and initiates of DCEC . They are connected by the object property when with the class Moment, and by the object properties terminatesWhat and initiatesWhat with the class Fluent, indicating what uent was terminated/initiated and when. The object properties terminatedBy and initiatedBy connect these classes with the class Event, representing the event that caused initiation or termination of the uent.
There are also four additional classes ful lling auxiliary functions. The class ActionType represents types of actions and provides classi cation for individual actions of the class Action. It is to be mainly used by higher ontological levels, to theoretize about possible or future actions. The class Object (superclass of class Agent) is a general class for all objects. Any other objects modeled by the ontology belong to this class.
The class Payoff represents the payo (using a numeric data property payoffValue) of an action represented by the class Action connected by the data property payoffAction. The rei cation of payo property to the class Payoff was necessary because the payo value is related to the Agent, Action and Moment. The same action of the same agent done at di erent moments may have di erent payo value. Similarly, the same action done at the same moment but di erent agents may have also di erent payo value.
The axioms of the event section of the ontology are the following:
Disjoint(Termination;Initiation;Moment;Fluent;Interval;Event;ActionType;Object;Payoff)
The core of the epistemic section of the ontology consists of three mutually similar classes, Knowledge, Belief and Perception, representing three di erent epistemic attitudes (there is also their superclass EpistemicState, which is for brevity not depicted in Fig. 3). They are connected to the knowing/believing/perceiving Agent using the property who. The content of the epistemic attitude is determined by the object property what connected to the classes Hold and FluentRelation. The time of knowing/believing/perceiving is determined by the property when connected to class Moment.
The class CommonKnowledge represents knowledge { property what { available, at a given time speci ed by property when, to all agents. It is similar to the epistemic states described above, but it obviously lacks the object property who, because it is know by all agents.
The class Communication represents the cases of information transfer between agents. It describes the communication of some content given by the (1) (2) (3) (4) (5) (6) (7)
The protential section of the ontology focuses on modeling the future. It currently only consists of two classes: Intention and Desire.
The class Intention represents an intention of an Agent (connected by the property who) to perform an action of the type determined by the class ActionType (connected by the object property intentionActionType). As mentioned above, there are two time speci cations: the object property when determines the time when the act of intention takes place, and the object property intendsWhen determines the time when the intended action should take place.
The class Desire is similar to Intention but it does not comprise any action of an agent himself. While agent may intend to do something himself (e.g. to pickup friend at 6pm), he does desire something to happen to him (e.g. desires to be picked up at 6pm by friend) . So in the case of intention the agents role is active, while in the case of desire the agent is passive. The Desire therefore only uses the object properties what, when, who and desiresWhen.
Both Intention and Desire are subclasses of the class ProtentialState (for brevity not depicted on Fig. 3).
The axioms of the protential section of the ontology are the following:
property what at some time de ned by the property when by an Agent described by the property who to another Agent { speci ed by the object property communicationTo.
The axioms of the epistemic section of the ontology are the following:
The deontic section of the ontology consists of a single class, Obligation. It is similar to class Intention, since obligation in this sense is structurally similar to intention. There is an individual (who) that at some point of time (when) thinks that his obligation is to do some action of an (ActionType), in a current or fu(8) (9) (10) (11) (12) (13) (14) ture time point (obligedWhen). Note that there are again two time speci cations similarly to the case of protential concepts.
The obligations are modeled `subjectively', so they are always related to an individual agent who is mentally aware of them as of a speci c behavioral binding applicable to his actions. The obligation is always an obligation to act somehow. Another important kind of obligation, that to abstain from an action, may be only modeled indirectly as obligation to a speci c passive action, e.g., \waiting at a given location".
The only axiom of the deontic section of the ontology is the following:
(15)
The DCE C calculus is based on the concept of a uent that has propositional character. Most foundational ontologies do not model propositional entities. It might be necessary to adapt the ontology so that it could somehow model propositional entities or even their internal structure. The relation between individual uents is now modeled using a FluentRelation, class but it might be insu cient.
In the real-world applications performing in the real time the speed of reasoning is of prime importance. The current expressiveness of the DCE O ontology is ALF C(D) . It is su ciently fast for the most of the reasoning tasks performed. However the complexity of the ontology may raise especially after adding the handling of internal structure of propositions. Such an increase of complexity could lead to signi cant reasoning-time overhead.
The protential section of the ontology currently contains the classes Intention and Desire. Both these classes concern future events and it might be possible to model those more precisely. Desires refer to foreseen future events that are positively evaluated but at the same time are not agent's own actions. Such events may be divided to those caused by the agent's actions (directly or indirectly) and those events that are desired but are out of the scope of being in uenced by the agent's actions.
Foreseen future events may also di er in terms of evaluation. There are positively evaluated events { desires { but there are also neutral events that are just foreseen as part of the probable future course of events { we may call them expectations { and also foreseen events that are negatively evaluated { fears. Such distinctions may be very useful for agent's understanding of his environment and for planing and realization of his actions.
This line of thoughts is connected to another possible extension of the ontology, one allowing to more precisely represent the plans of agent and courses of events that happen in his environment.
In the future the deontic section of the ontology may also be extended so as to enable the handling of more general concepts of obligation than the current \subjective" obligation adapted form DCE C does. It might be useful to think of \universal" obligations applicable to all agents, or of obligations applicable to an agent but unknown to him. Another improvement may be more direct handling of obligations to abstain from an action. This type of obligation is in some context more appropriate and important.
Solving the problem of an autonomous action of arti cial agents is indispensable in order to progress in many areas of arti cial intelligence. The research dealing with this problem in the context of the Semantic Web is very limited, and our project presents an e ort to bridge the gap between Semantic Web technologies and Cognitive Science.
We have presented an ontology framework based on the Deontic Cognitive Event Calculus (DCE C ) and inspired by other general cognitive architectures. We also presented a brief overview of our cognitive architecture that includes proposed ontology as its core and enables e ective cognitive reasoning on top of it.
The Deontic Cognitive Event Ontology (DCE O) described in this paper consists of four parts: event ontology, epistemic ontology, protential ontology and deontic ontology. The event ontology allows modeling of actions of arti cial agents occurring at speci c times or intervals. The epistemic ontology describes the mental states of these agents: belief, knowledge and perception, and their propositional content. It also enables the modeling of communication between agents and of common knowledge available to all agents. The proposed protential ontology models the agents' attitudes to future { their desires and intentions. This in turn in uences their autonomous actions. Finally, the deontic ontology aims at modeling obligations { the actions that the agent knows he is obliged to perform at a given situation.
The paper nally discusses some problems and limitations of the presented framework and proposes some possible future enhancements and extensions.
This work has been supported by the long-term institutional support of research activities by the Faculty of Informatics and Statistics, University of Economics, Prague.