An OWL Ontology for Biographical Knowledge. Representing Time-Dependent Factual Knowledge Hans-Ulrich Krieger and Thierry Declerck German Research Center for Artificial Intelligence (DFKI), Universität des Saarlandes, Allgemeine Linguistik krieger@dfki.de, declerck@dfki.de Abstract Representing time-dependent information has become increasingly important for reasoning and querying services defined on top of RDF and OWL. In particular, addressing this task properly is vital for practical applications such as modern biographical information systems, but also for the Semantic Web/Web 2.0/Social Web in general. Extending binary relation instances with temporal information often translates into a massive proliferation of useless container objects when trying to keep the underlying RDF model. In this paper, we argue for directly extending RDF triples with further arguments in order to easily represent time-dependent factual knowledge and to allow for practical forms of reasoning. We also report on a freely available lightweight OWL ontology for representing biographical knowledge that models entities of interest via a tri-partite structure of the pairwise disjoint classes Abstract, Object, and Happening. Even though the ontology was manually developed utilizing the Protégé ontology editor, and thus sticking to the triple model of RDF, the meta-modelling facilities allowed us to cross-classify all properties as being either synchronic or diachronic. When viewing the temporal arguments as “extra” arguments that only apply to relation instances, universal biographical knowledge from the ontology can still be described as if there is no time. Keywords: OWL biography ontology, representation of time-dependent information, practical temporal reasoning. 1 Synchronic and Diachronic Relations cg isChairman vf Linguistics and philosophy make a distinction between syn- cg holdsAt [????-??-??, 2003-07-??] chronic and diachronic relations in order to characterize cg isChairman gsk statements whose truth values do or do not change over cg holdsAt [2005-01-01, ????-??-??] time. Synchronic relations, such as dateOfBirth, are rela- However, the association between the original statements tions whose instances stay constant over time, thus there is and their temporal extents get lost in the resulting RDF no direct need to attach a temporal extent to them. Con- graph: sider, e.g., the natural language sentence: cg isChairman vf @ [????-??-??, 2003-07-??] Tony Blair was born on May 6, 1953. cg isChairman vf @ [2005-01-01, ????-??-??] cg isChairman gsk @ [????-??-??, 2003-07-??] Assuming a RDF-based N-triple representation (Carothers cg isChairman gsk @ [2005-01-01,????-??-??] and Seaborne, 2014), an information extraction system might yield the following set of triples: as the second and third association are not supported by the above natural language quotation. tb rdf:type Person tb hasName ”Tony Blair” 2 Approaches for Representing tb dateOfBirth ”1953-05-06”ˆˆxsd:date Time-Dependent Knowledge Since there is only one unique date of birth, this works per- fectly well and properly capture the intended meaning. Several well-known proposals have been presented in the literature in order to equip (binary) relation instances with Diachronic relationships, however, vary with time, i.e., time or other kinds of information. The individual rewrit- their truth value do change over time. Representation ing schemas are depicted in Figure 1; see Welty and Fikes frameworks such as OWL that are geared towards unary (2006) and Krieger (2014) for a closer overview: and binary relations can not be extended directly by further (temporal) arguments. Consider the following biographical 1. directly equip the relation instance with addi- information: tional/temporal arguments (Krieger, 2012); Christopher Gent was Vodafone’s chairman until 2. apply a meta-logical predicate as used in the situation July 2003. Later, Chris became the chairman of calculus (McCarthy and Hayes, 1969); GlaxoSmithKline with effect from January 1st, 2005. 3. reify the original relation à la RDF, turning the prop- From these two sentences, the information extraction sys- erty into a class (Manola and Miller, 2004); tem might discover the following underspecified time- 4. employ a fact identifier à la YAGO, implicitly leading dependent facts: to quads (Hoffart et al., 2011); cg isChairman vf @ [????-??-??, 2003-07-??] 5. wrap the range arguments in an object, called N-ary cg isChairman gsk @ [2005-01-01, ????-??-??] relation encoding by W3C (Hayes and Welty, 2006); Applying the synchronic representation schema for dateOf- 6. encode a perdurantist/4D view in OWL (Welty and Birth from above would give us: Fikes, 2006); 101 approach rewriting schema 1 marriedTo(p, p0 ) 7−→ marriedTo(p, p0 , s, e) holds(marriedTo(p, p0 ), t) 7−→ ∃f . holds(f, t) ∧ 2 type(f, Fluent) ∧ subject(f, p) ∧ predicate(f, marriedTo) ∧ object(f, p0 ) marriedTo(p, p0 , s, e) 7−→ ∃e . type(e, MarriedToEvent) ∧ 3 person1(e, p) ∧ person2(e, p0 ) ∧ starts(e, s) ∧ ends(e, e) 4 marriedTo(p, p0 , s, e) 7−→ ∃i . i := marriedTo(p, p0 ) ∧ starts(i, s) ∧ ends(i, e) marriedTo(p, p0 , s, e) 7−→ ∃o . marriedTo(p, o) ∧ 5 type(o, PersonTime) ∧ person(o, p0 ) ∧ starts(o, s) ∧ ends(o, e) marriedTo(p, p0 , s, e) 7−→ ∃t, t0 . marriedTo(t, t0 ) ∧ 6 type(t, TimeSlice) ∧ hasTimeSlice(p, t) ∧ type(t0 , TimeSlice) ∧ hasTimeSlice(p0 , t0 ) ∧ starts(t, s) ∧ ends(t, e) ∧ starts(t0 , s) ∧ ends(t0 , e) marriedTo(p, p0 , s, e) 7−→ ∃t, t0 . marriedTo(t, t0 ) ∧ 7 type(t, Person) ∧ hasTimeSlice(p, t) ∧ type(t0 , Person) ∧ hasTimeSlice(p0 , t0 ) ∧ starts(p, s) ∧ ends(p, e) ∧ starts(p0 , s) ∧ ends(p0 , e) marriedTo(p, p0 , s, e) 7−→ marriedTo s e(p, p0 ) ∧ 8 marriedTo s e v marriedTo ∧ starts(marriedTo s e, s) ∧ ends(marriedTo s e, e) 9 marriedTo(p, p0 , s, e) 7−→ marriedTo(p, p0 ) ∧ starts(p0 , s) ∧ ends(p0 , e) Figure 1: Different ways of representing the atemporal statement (the “fluent”) marriedTo(p, p0 ) between two people p and p0 , being true for the time period t = [s, e]. “7−→” should be read as rewrite to. The last representation schema only works if the original property (here: marriedTo) is inverse functional for all relation instances (which needs not to be the case). 7. interpret the original entities as time slices (Krieger, from their “parts” (i.e., from information that is accessible 2008); through properties from the new individual)—this “trick” 8. encode the temporal extent through new synthetic reminds us of constructing perfect hash functions over com- properties (Gangemi, 2011); plex objects, as known from computer science. 9. use relation composition applied to the second argu- Approach 1 is pursued in the temporal database commu- ment which does not work in general, but only if orig- nity under the heading valid time (Snodgrass, 2000). The inal relation is inverse functional. measurements in Krieger (2012) and Krieger (2014) have shown that this approach easily outperforms all other ap- 2.1 Discussion proaches during querying and reasoning (computation of The above approaches are in a certain sense semantically the deductive closure) in the time domain by several orders equivalent in that we can rewrite one approach to another of magnitude. In some cases, this divergency can make one without losing any information. It is worth noting that a difference between doable and intractable applications. all approaches invalidate standard OWL reasoning, even Consequently, we think the time now is ripe for allowing n- though they can be implemented within the RDF frame- ary relations, or as Schmolze (1989) once put it in the early work, and thus at least explicitly stated information can days of KL-ONE “... the advantages for allowing direct be queried by, e.g., SPARQL engines. Nevertheless, the representation of n-ary relations far outweigh the reasons non-temporal entailment rules for RDFS (Hayes, 2004) and for the restriction.” OWL Horst/OWL 2 RL (ter Horst, 2005; Motik et al., 2012) can be adjusted, so that rule-based reasoners that go beyond 2.2 Tuples vs. Triples: Representation & Reasoning symbol matching, such as Jena (Reynolds, 2006) or HFC We would like to make our preference towards a direct rep- (Krieger, 2013), are still able to perform extended entail- resentation of additional (temporal) arguments more clear ments under these new encoding schemas. by looking at concrete examples. Consider the Wikipedia Most of the above approaches require to rewrite the orig- entry for Tony Blair which says he married Cherie Booth inal ontology, sometimes by turning relations into classes. on 29th March 1980 (today = 2015-05-08), leading to the With the exception of approach 1, all approaches require quintuple representation: to introduce one or even two brand-new individuals per tony blair marriedTo cherie booth time-dependent fact (see Figure 1). As a consequence, ”1980-03-29”ˆˆxsd:date ”2015-05-08”ˆˆxsd:date reasoning and querying with such representations is ex- tremely complex, expensive, and error-prone. Further- A meaning-preserving triple representation which adheres more, the representation schemas 2–7 bear the potential of to a W3C best practice recommendation, called N-ary re- a non-terminating closure computation in case the newly lation encoding (see rewrite schema 5 in Figure 1) would introduced individuals are viewed as existentially quanti- instead result in five triples, a new individual :ppt, a new fied, i.e., anonymous logic variables (RDF: blank nodes). type ValuePlusTime, and the three “accessor” properties Luckily, this last danger can often be avoided by generat- hasValue, starts, and ends: ing unique URI names that are deterministically generated tony blair marriedTo :ppt 102 :ppt rdf:type nary:ValuePlusTime @action :ppt nary:hasValue cherie booth ?s = Max2 ?s1 ?s2 :ppt nary:starts ”1980-03-29”ˆˆxsd:date ?e = Min2 ?e1 ?e2 :ppt nary:ends ”2015-05-08”ˆˆxsd:date The additional left hand side four-place relation Intersec- Such a representation has a three times larger memory foot- tionNotEmpty from the @test section of the rule simply print, a slightly more complex structure, and is a bit harder checks whether the two temporal intervals [s1 , e1 ] and to read. However, as indicated above, the new individual [s2 , e2 ] have a non-empty intersection, indicated by xxx be- (in our example blank node :ppt) might turn out to be prob- low: lematic during entailment reasoning (no longer guaranteed p(x,y) to terminate). |xxx————| p(x,z) Now let us focus not only on the representation of (static) |——xxx| knowledge, but on the (dynamic) derivation of new knowl- · · · ———s1 —s2 –e1 ———e2 ————. t edge through entailment rules in order to see how much worse a (recommended) triple representation becomes. If this is the case, we mark the subject bound to ?x being Consider the following entailment schema for functional of type owl:Nothing (same as for the original rule), but this diachronic datatype properties (in Section 3.3, we will look type assignment now only holds for the overlapping obser- at the corresponding entailment schema for functional di- vation time, given by the maximum of the starting times (= achronic object properties). The original non-temporal ?s) and the minimum of the ending times (= ?e), as com- schema looks like this (we use the rule syntax of HFC puted in the @action section of the rule. (Krieger, 2013) in the examples below): The above natural extension of the non-temporal rule, how- ?p rdf:type owl:FunctionalProperty ever, turns into an awfully looking and terribly inefficient ?p rdf:type owl:DatatypeProperty rule when being couched in a triple-based setting: ?x ?p ?y ?p rdf:type owl:FunctionalProperty ?x ?p ?z ?p rdf:type time:DiachronicProperty → ?p rdf:type owl:DatatypeProperty ?x rdf:type owl:Nothing ?x ?p ?blank1 @test ?blank1 rdf:type nary:ValuePlusTime ?y != ?z ?blank1 nary:hasValue ?y Such a rule schema is useful, e.g., for detecting contradic- ?blank1 nary:starts ?start1 tory birth dates for one and the same person (famous ex- ?blank1 nary:ends ?end1 ample: Louis Armstrong; right: August 4, 1901, wrongly ?x ?p ?blank2 claimed by him: July 4, 1900). Such a schema matches, for ?blank2 rdf:type nary:ValuePlusTime instance, ?blank2 nary:hasValue ?z louis armstrong dateOfBirth ”1901-08-04”ˆˆxsd:date ?blank2 nary:starts ?start2 louis armstrong dateOfBirth ”1900-07-04”ˆˆxsd:date ?blank2 nary:ends ?end2 and binds louis armstrong to ?x, dateOfBirth to ?p, ”1901- → ?x rdf:type ?new 08-04”ˆˆxsd:date to ?y, and ”1900-07-04”ˆˆxsd:date to ?z. ?new rdf:type nary:ValuePlusTime Having found problematic cases is signaled by assigning ?new nary:hasValue owl:Nothing the “bottom” type owl:Nothing to the subject element of the ?new nary:starts ?start triple bound to the logical variable ?x (= Louis Armstrong) ?new nary:ends ?end on the right hand side of the rule. @test Adding time to this rule schema makes it applicable to ?y != ?z other functional relations such as hasSalary which do IntersectionNotEmpty ?start1 ?end1 ?start2 ?end2 change over time, as indicated by the property character- @action istics time:DiachronicProperty in the rule below. Extending ?start = Max2 ?start1 ?start2 the rule schema is quite easy by equipping the fourth and ?end = Min2 ?end1 ?end2 fifth left hand side clauses with a temporal extent (things ?new = MakeUri owl:Nothing ?start ?end that have been added are underlined): Note how the relevant input information is hidden in the ?p rdf:type owl:FunctionalProperty two container individuals bound to ?blank1 and ?blank2 ?p rdf:type time:DiachronicProperty and how the output is wrapped in a brand-new individual ?p rdf:type owl:DatatypeProperty ?new, generated by MakeUri from the @action section. ?x ?p ?y ?s1 ?e1 ?x ?p ?z ?s2 ?e2 2.3 Limitations → Several points are worth mentioning here. Firstly, we are ?x rdf:type owl:Nothing ?s ?e not dealing here with duration time in order to resolve ex- @test pressions like Monday or 20 days against valid time. This ?y != ?z needs to be handled by a richer temporal ontology and tem- IntersectionNotEmpty ?s1 ?e1 ?s2 ?e2 poral arithmetic. 103 Secondly, temporal quantification, such as four hours every bio:Person ≡ pol:Person week, needs to be addressed by a richer temporal inventory. or constrain the domain and range of potentially underspec- Thirdly, even though underspecified time is handled by our ified properties, e.g., implementation through wildcards in the XSD dateTime > v ∀op:hasHolder . bio:Agent format (e.g., year missing in Over New Year’s Eve, I have The property hasHolder from the opinion ontology (prefix visited the Eiffel Tower), we do not focus on this here. op) is a good example of a property for which only the do- The solution requires to make certain rule tests sensitive to main has been specified, viz., op:Opinion: the fact that underspecified time is only partially ordered. These tests then return true, false, or don’t-know, whereas > v ∀op:hasHolder− . op:Opinion only true indicates that the test has succeeded, leading to However, hasHolder consciously lacks its range, since this the instantiation of the right hand side of the rule. information should only be added when several ontologies Fourthly, coalescing temporal information (i.e., building are brought together. larger intervals from intervals with overlapping parts) The above axioms together with the two terminological ax- should be addressed in custom rules and should not be re- ioms from the biography (prefix bio) and the politics (prefix garded as part of the extended RDFS/OWL rule set, since pol) ontologies this functionality depends on the (semantic) nature of pred- bio:Person v bio:Agent icates and the assumption whether temporal intervals are pol:Journalist v pol:Person convex (i.e., contain no “holes”) or not. guarantee to draw legal inferences, such as journalists are And finally, certain temporal inferences such as p(~x, s, t) holders of opinions, even though the interface axiom above entails p(~x, s0 , t0 ) in case s ≤ s0 ≤ t0 ≤ t should not be constrain holders of opinions to be of type bio:Agent. handled in the below rules, since termination of the com- putation of the deductive closure is no longer guaranteed. TMO has been assembled from 16 sub-ontologies, some of Such information can only be obtained on the query level. them also dealing with the representation of biographical knowledge, others describing concepts that can be found in 3 Ontology for Biographical Knowledge politics and sociology. Especially the opinion ontology can We already indicated that we favor approach 1 as it is the be used to model provenance information, important for bi- most perspicuous of the nine approaches presented above, ographical knowledge; for instance, information about the: shows the best memory and runtime footprint, and always • holder of the opinion: hasHolder; guarantees a terminating closure computation for extended • source from which the info was taken: extractedFrom; RDFS (Hayes, 2004) and OWL (ter Horst, 2005) entail- • time when the opinion was published: utteredAt; ment, as shown in Krieger (2012). • trustworthiness of the holder: holdersTrust; In the introduction, we argued that axiomatic knowledge • polarity of the opinion: hasPolarity. about classes (TBox) and properties (RBox) does not need to have a notion of time—this is universal knowledge which The TMO ontology suite is freely available for aca- we assume to be static. For instance, we do not assume that demic research and to other sites upon request (see http://www.dfki.de/lt/onto/). Parts of the taxonomic structure the subtype relationship between two classes only holds for some period of time or that an URI should be regarded as a of the biography ontology is depicted in Figure 2. property at time t and as a class at a different time t0 (even 3.1 Overall Guidelines though this would be possible). The assertional knowledge TMO, and thus the biography ontology, implements several of an ontology (ABox), i.e., the set of relation instances, “guidelines” that we have found useful in many projects however, is what we equip with time (see the various ap- which have dealt with the representation of time-dependent proaches for the marriedTo example in Figure 1), as this is knowledge (some of the arguments have already been pre- knowledge that has undergone a temporal change. sented): In this section, we present the schema (the TBox and 1. model the TBox and RBox axioms of an ontology as if the RBox) of an ontology that we had developed origi- there is no time, since the ontology schema is regarded nally for the TAKE project (http://take.dfki.de) and that was to be immutable; consequence: standard ontology ed- used in the KOM PARSE project (http://komparse.dfki.de) itors, such as Protégé can be used for this task. to represent biographical information about celebrities 2. cross-classify all properties as being either synchronic (Adolphs et al., 2010). This ontology has been or diachronic; advantage: these property characteris- reused and extended in the EU projects MONNET tics can be used, amongst other things, as distinguish- (http://cordis.europa.eu/fp7/ict/language-technologies) and ing marks in entailment rules (see examples). T REND M INER (http://www.trendminer-project.eu). This bi- ography ontology is now part of a larger set of indepen- 3. populate the ABox of an ontology with extended rela- dently developed ontologies (called TMO, for T REND - tion instances, i.e., with quintuples whose fourth and M INER O NTOLOGIES) which are interlinked to one an- fifth argument encode the temporal extent of the pre- other through the use of interface axioms (Krieger and ceding atemporal statement (the triple). Declerck, 2014). These interface axioms either relates 4. extend the RDFS/OWL entailment rules by a tempo- classes (TBox) and properties (RBox) from different sub- ral dimension; example: use XSD’s date or dateTime ontologies through the use of description logic axiom con- format to implement an interval-based calendar time structors, e.g., (used by the examples in this paper). 104 Figure 2: The class subsumption hierarchy of the biography ontology. Note the two subclasses time:DiachronicProperty and time:SynchronicProperty of class rdf:Property that are used to cross-classify (i.e., to type) the properties of the biography ontology; see Figure 4. 3.2 Tri-Partite Structure pening is basedOn or leadsTo entities (i.e., either abstract The biography ontology assumes a tri-partite structure, things, further happenings, or concrete objects), thus these defining a most general class Entity, having pairwise dis- properties can be used to encode pre- and post-conditions of joint subclasses Abstract, Happening, and Object. TMO is a happening. An instance of this class also involves Agents a lightweight ontology that consists of 146 classes and 80 and happensAt a Location. Situations help to “terminate” properties, and is of expressivity SHIN (D), according to the decomposition of a Happening. The other subclass the Ontology metrics pane of Protégé, version 4.3.0. A par- Event can be used to model simple unordered processes, tial view of the three subclasses and properties linking them as it comes with three relational properties of its own, viz., is given in Figure 3. startsWith, continuesWith, and endsWith, all mapping to Happening (see Figure 2). 3.2.1 Abstract Ontological categories that do not fit into Happening or Ob- 3.2.3 Object Objects are “physical” things and mostly deal with Agents ject are regarded to be of type Abstract, thus this class is a kind of “remainder” category. Abstract things can be used (an exhaustive disjoint partition between Person, Group, to describe literal concepts, e.g., activities, academic de- and political State) and other categories that we think are grees, ideas, inventions, the life, or personal, professional, relevant for biographical information, e.g., Location, ma- and social roles. An abstraction manifestsIn real-world hap- terial Property, or WorkAndProduct. A Person isAwareOf penings, whereas the outcome of a happening leadsTo vir- a Happening: (s)he “owns” it, can be part of it, or learns tually everything (= Entity). For example: a specific mili- about a happening. As isAwareOf is a diachronic property, tary activity (the invasion of Poland) manifested in World awareness of a happening might even turn into oblivion. War II. The outcome of WW-II has led to military inven- 3.3 Practical Temporal Reasoning tions (Abstract), has led to the Cold War (Happening), and has led to the building of 86 U2 aircrafts (Object). For a larger non-trivial example, let us again turn our at- tention to the marriage of Tony Blair and Cherie Booth. 3.2.2 Happening marriedTo is at the same time a symmetric, a diachronic, Happenings are things that “happen” or “unfold” and are a functional, and an object property (see the Types pane at disjointly categorized as being either static atomic Situa- the bottom of Figure 4). tions or dynamic decomposable Events. They come with a We mentioned that we have cross-classified every property (possibly underspecified) startDate and endDate. A hap- from the biography ontology as being either synchronic 105 why we introduced another property divorcedFrom, be- ing the temporal disjoint object property to marriedTo (see the owl:disjointObjectProperty pane in Figure 4). As the Economist article does not specify the date of marriage, we better opt for a moment in time, when Blair and Booth were definitely married (actually a day: start = end). Luckily, the right hand side sameAs inference from above, together with another extended OWL entailment rule, called rdfp11 (ter Horst, 2005), makes sure that even tony blair marriedTo cherie blair ”1980-03-29”ˆˆxsd:date ”2015-05-08”ˆˆxsd:date Figure 3: Properties of the biography ontology which relate is a valid entailment, exactly what we expect. the three disjoint classes Happening, Object, and Abstract. The solid blue triangle on the right side should indicate sub- 3.4 Temporal Arguments as Extra Arguments classes of the class Abstract, such as Achievement. So far, our approach has argued for a direct encoding of the or diachronic and have already discussed the temporal ex- temporal extent through two further arguments, turning a tension of the entailment rule for functional diachronic binary relation, such as marriedTo ⊆ Person×Person into a datatype properties in Section 2.2. Let us now focus on the quaternary one: marriedTo ⊆ Person×Person×date×date. complementary rule for functional diachronic object prop- Given the original relation signature, the non-temporal en- erties which is applicable to the marriedTo relation: tailment rule schema for symmetric binary relations from ter Horst (2005) thus leads to the following instantiation: ?p rdf:type owl:FunctionalProperty ?p rdf:type time:DiachronicProperty marriedTo(p, p0 ) → marriedTo(p0 , p) ?p rdf:type owl:ObjectProperty as symmetric relations swap their domain and range argu- ?x ?p ?y ?s1 ?e1 ments (p, p0 being two people). ?x ?p ?z ?s2 ?e2 Now, if we add time (b = begin; e = end), we obtain:1 → marriedTo(i; j, b, e) → marriedTo(j, b, e; i) ?y owl:sameAs ?z @test Clearly, something has gone wrong here because sym- IntersectionNotEmpty ?s1 ?e1 ?s2 ?e2 metric relations assume the same number of arguments in domain and range position. One solution would be to redu- Here, as in the former example, the additional left hand side plicate the starting and ending points, so we would end up test IntersectionNotEmpty checks whether the two tempo- in sexternary relation: ral intervals [s1 , e1 ] and [s2 , e2 ] have a non-empty intersec- tion. Assuming that a person is not married to more than marriedTo(i, b, e; j, b, e) → marriedTo(j, b, e; i, b, e) one partner at the same time, such a rule is able to iden- This is not an appealing solution as the structures become tify individuals/URIs bound to ?y and ?z for two properly larger, and rules and queries are harder to formulate, read, overlapping observations through the use of owl:sameAs. debug, and process. What we would like to see is some- Consider again the Wikipedia entry for the marriage of thing like: Tony Blair and Cherie Booth that we used in the example marriedTo(i; j; b, e) → marriedTo(j; i; b, e) from Section 2.2: whereas the second semicolon should indicate that the ad- tony blair marriedTo cherie booth ditional temporal arguments are extra arguments, belonging ”1980-03-29”ˆˆxsd:date ”2015-05-08”ˆˆxsd:date to the relation instance as such (a kind of relation instance and furthermore assume that the Economist article The annotation, not possible in OWL). Thus with this idea in loneliness of Tony Blair from December 2014 mentioned mind, we can still keep the idea of having only binary re- that Cherie Blair is Blair’s wife (quintuple again): lations, without introducing any new identifier (contrary to tony blair marriedTo cherie blair the rewrite schemas 2–7 from Figure 1). ”2014-12-20”ˆˆxsd:date ”2014-12-20”ˆˆxsd:date Nevertheless, we are not arguing against arbitrary n-ary re- Now it is safe to assume that Cherie Booth and Cherie Blair lations as we are convinced that many binary relations in are in fact the same person, according to the successful ap- today’s ontologies are ignoring additional arguments (e.g., plication of the above temporal entailment rule: properties oriented towards ditransitive verbs or having ad- ditional modifiers/adjuncts) or come along with unsatisfac- cherie booth owl:sameAs cherie blair tory means to encode the additional arguments (relation It is worth noting that sameAs statements will not be composition, by taking the object of a binary relation in- equipped with a temporal extent—commonsense dictates stance into account). The current biography ontology, for that once we do identify individuals, they will never fall instance, poorly models the property obtains as a relation apart. At every moment in time, we never know how long a 1 For better readability, we separate the domain and range ar- person is married to his/her partner in advance. That is guments from one another by using a semicolon. 106 Figure 4: The property subsumption hierarchy of the biography ontology. between people and (academic) degrees. In order to ob- In case more (e.g., time) and especially optional arguments tain the educational organization where the degree was ob- are investigated, our verdict concerning the different ap- tained, we employ relation composition at the moment, us- proaches might turn into a different direction, so the rep- ing an additional property obtainedAt between degree and resentation format needs to be updated (in the best case) or education: changed (in the worst case). Consider the following exam- obtainedAt ◦ obtains ⊆ Person × EducationalOrganization ple, taken from (Davidson, 1967, p. 83): This way of representing the additional argument is related Jones buttered the toast in the bathroom with a knife to approach 9 from Figure 1 and only works because ob- at midnight. tains is inverse functional (a characteristics applicable to The binary base relation butter (we assume a direct map- properties in OWL). Ideally, obtains should be modeled as ping of the transitive verb to the relation name here) now a quinternary relation, having one domain argument, two needs to be split and/or extended by further optional argu- range arguments, and two extra temporal arguments: ments, as the following sentences are perfectly legal: obtains ⊆ Person × // domain Jones buttered the toast. Degree × EducationalOrganization × // range Jones buttered the toast in the bathroom. xsd:dateTime × xsd:dateTime // extra Jones buttered the toast with a knife. In order to easily define such non-binary relations, ontology Jones buttered the toast at midnight. editors need to be extended by Cartesian types. In Krieger Jones buttered the toast in the bathroom with a knife. and Willms (2015), we described ×-Protégé, an extension Jones buttered the toast in the bathroom at midnight. of the Protégé ontology editor that provides means to define ..... etc. such Cartesian types and to use them to type the domain, In principle, the number of adjuncts is not bounded, thus range, and extra arguments of non-binary relations. A first adding a large number of potentially underspecified direct public version of ×-Protégé will be available in mid 2015. relation arguments is probably a bad solution. Today’s tech- nologies often address such hidden arguments through a 4 Relation vs. Event Representation kind of relation composition as we have seen above for the The approaches considered in Section 2 were investigated obtains example from the last section and listed as approach on how well they perform w.r.t. binary relations whose two 9 in Figure 1. We think that this modeling “trick” is unsat- arguments are considered to be obligatory. Such a kind of isfactory as it operates on the object of the binary relation relation is the default case in today’s popular knowledge instance, but not on the relation instance itself (besides be- resources, such as YAGO, DBpedia, BabelNet, or Google’s ing only correct if the original relation is inverse functional, Knowledge Graph. as explained before). 107 Our personal solution would model the obligatory argu- to stick to an event-like representation in which all infor- ments, including (under- or unspecified) time and perhaps mation is hidden in an object and time is accessible through space, as direct arguments of the corresponding relation in- properties, similar to approach 3 in Figure 1. None of them stance or tuple (approach 1). A further argument, an event are able to encode time as direct arguments of a relation in- identifier, also takes part in the relation. Optional argu- stance (approach 1). A comparison of some of these event ments, however, would be addressed through binary rela- ontologies is presented in (Shaw et al., 2009, section 2) and tions, now working on the event argument. Applying this (van Hage et al., 2011, section 5). kind of Davidsonian or event representation to the above As we have indicated in the beginning of Section 3 example gives us (informal relational notation): (Journalist example), OWL axiom constructors and do- ∃e . butter(e, Jones, toast, at midnight) ∧ main/range restrictions allow us to manually interface location(e, bathroom) ∧ instrument(e, knife) our biography ontology with other ontologies, may they It is worth noting that two of the approaches from Figure 1 be complimentary domain ontologies (opinion, politics, are related to such an event representation, viz., 3 and 4. sociology), overlapping biography event ontologies (see above), or even OWL versions of upper ontologies (if Approach 3 (internal reification) can be seen as a kind desired), such as DOLCE+DnS (Gangemi et al., 2002), of “owlfication” of Neo-Davidsonian semantics (Parsons, SUMO (Niles and Pease, 2001), or Cyc (Reed and Lenat, 1990), as the original relation is always turned into an event 2002). For instance, if we would like to interface the BBC (an OWL class). Here the event identifier e from above di- storyline ontology, the following single axiom suffices: rectly corresponds to a URI, referring to an instance of the OWL class. For instance, the marriedTo relation is turned bio:Happening ≡ nsl:Event into an event class, say Marry; thus: Connecting with LODE essentially reduces to: tony blair marriedTo cherie booth bio:Happening ≡ lode:Event ”1980-03-29”ˆˆxsd:date ”2015-05-08”ˆˆxsd:date bio:happensAt ≡ lode:atPlace needs to be expressed by (we use VerbNet terminology): bio:involves ≡ lode:involvedAgent e rdf:type Marry bio:basedOn v lode:involved e agent tony blair bio:leadsTo v lode:involved e co-agent cherie booth Other properties from LODE either do not have a di- e starts ”1980-03-29”ˆˆxsd:date rect counterpart (lode:illustrate) or need to be decom- e ends ”2015-05-08”ˆˆxsd:date posed (lode:atTime onto bio:startDate and bio:endDate). Approach 4 (fact identifier) is a kind of external reification. The sub-properties bio:startsWith, bio:continuesWith, and YAGO uses its own extension of the N3 plain triple for- bio:endsWith from the class bio:Event would even allow us mat, called N4, which associate unique identifiers i with to decompose LODE events into smaller units, a feature each time-dependent fact. However, the association i := partially available in the SEM ontology: marriedTo(p, p0 ) has the disadvantage of not being part of bio:startsWith v sem:hasSubEvent the triple repository, as it is a quadruple technically. So we bio:continuesWith v sem:hasSubEvent guess that there exists a separate extendable mapping table bio:endsWith v sem:hasSubEvent outside of the semantic repository, storing the triples. Luckily, the biography ontology presented in Section 3 both As our ontology comes with the class bio:Happening, it is allows for extended relation instances (as shown before), possible to take advantage of the great effort invested in but also Davidsonian-like events through the class Happen- the definition of event types in the ESO ontology. We fi- ing and its subclasses Event and Situation (see Figure 2). As nally note that some of the mappings are not expressible there does not exist a Marry event class so far (but only the through simple OWL axiom constructors, because they in- marriedTo property), such a class needs to be introduced as volve a translation from n-ary relation instances to sets of a subclass of class Event, if needed. triples (and vice versa). This would require to apply HFC migration rules, similar to the rewrite rule of approach 3 in 5 Related Ontologies Figure 1 which mediates between the quaternary marriedTo relation and its event representation MarriedToEvent. Several ontologies addressing the representation of bio- graphical information, cultural heritage information, and news-related information exist today, all building on De- 6 Summary and Conclusion scription Logics and Semantic Web technology stan- In this paper, we have presented an overview of nine ap- dards. These include ESO (Segers et al., 2015), Wikidata proaches to the representation of time-dependent knowl- (Erxleben et al., 2014), the BiographyNet ontology (Ock- edge and have favored the direct encoding of the tempo- eloen et al., 2013), the BBC Storyline Ontology (Wilton ral information as extra arguments of the original relation et al., 2013), SEM (van Hage et al., 2011), FRBR OO (Le instance. Nevertheless, allowing at the same time for an Bœuf, 2010), LODE (Shaw et al., 2009), or Event-Model-F event-based representation of situations, happening in the (Scherp et al., 2009). Some of these ontologies make use of real world, is profitable as a knowledge engineer might other resources, such as WordNet, FrameNet, Wikipedia, choose the representation which fits her/his needs. For in- SUMO , DOLCE , or CIDOC CRM . In order to represent stance, a marriage ceremony between two people is prob- time-dependent knowledge, these approaches always need ably modeled best as an event, whereas the fact that these 108 two people are married for a specific time period is bet- Wikidata to the Linked Data Web. In Proceedings of the ter represented as a quaternary relation. The lightweight 13th International Semantic Web Conference (ISWC), biography ontology, presented in this paper, allows both pages 50–65. views through the very general class Happening and re- Aldo Gangemi, Nicola Guarino, Claudio Masolo, Alessan- lations defined between classes which are extended by a dro Oltramari, and Luc Schneider. 2002. Sweetening starting and ending time, expressing the temporal extent in ontologies with DOLCE. In Proceedings of the 13th In- which the atemporal fact is true (called valid time in tem- ternational Conference on Knowledge Engineering and poral databases). Knowledge Management (EKAW), pages 166–181. Our debate on the right representation format can even be Aldo Gangemi. 2011. SuperDuper schema: an viewed as the more general quest on how to integrate/add OWL2+RIF DnS pattern. In KCAP 2011 Deep Knowl- important (meta) information that has been neglected in the edge Representation Challenge Workshop. past for practical matters, but has gained a lot of atten- Patrick Hayes and Chris Welty. 2006. Defining N-ary rela- tion recently; see the W3C recommendation for the prove- tions on the Semantic Web. Technical report, W3C. nance data model PROV-DM (Moreau and Missier, 2013). Patrick Hayes. 2004. RDF semantics. Technical report, This additional information might include the holder of a W3C. time-dependent statement or event (person, website, pro- Johannes Hoffart, Fabian M. Suchanek, Klaus Berberich, gram/service), the spacial location of the holder, the time Edwin Lewis Kelham, Gerard de Melo, and Gerhard when the statement/event was communicated by the holder Weikum. 2011. YAGO2: Exploring and querying world or made public on the Web (related to transaction time in knowledge in time, space, context, and many languages. temporal databases), the trustworthiness of the holder, and In Proceedings of the 20th International World Wide Web the attitude of the holder w.r.t. the statement/event (senti- Conference (WWW 2011), pages 229–232. ment/opinion). Ontologies for all these different aspects al- Hans-Ulrich Krieger and Thierry Declerck. 2014. TMO— ready exist today (for instance, the BiographNet ontology the federated ontology of the TrendMiner project. In (Ockeloen et al., 2013) which incorporates a multi-level, Proceedings of the 9th edition of the Language Re- multi-perspective model for provenance), but a unified stan- sources and Evaluation Conference (LREC). dard is still missing. As a short-/mid-term workaround, we Hans-Ulrich Krieger and Christian Willms. 2015. Extend- suggest to manually interface these different sources of in- ing OWL ontologies by Cartesian types to represent N- formation, as indicated in Section 5, thus making it possible ary relations in natural language. In Proceedings of the to incorporate work carried out by other researchers. IWCS Workshop on Language and Ontologies. Hans-Ulrich Krieger. 2008. Where temporal description Acknowledgements logics fail: Representing temporally-changing relation- The work described in this paper has been partially car- ships. In KI 2008: Advances in Artificial Intelligence, ried out in the ALL SIDES project, funded by the German volume 5243 of Lecture Notes in Artificial Intelligence, Federal Ministry of Education and Research under contract pages 249–257. Springer. number FKZ 01IW14002 and the PHEME project, funded Hans-Ulrich Krieger. 2012. A temporal extension of the by the European Union’s 7th Framework Programme un- Hayes/ter Horst entailment rules and an alternative to der grant agreement no 611233. The authors would like to W3C’s n-ary relations. In Proceedings of the 7th Inter- thank our three reviewers and Antske Fokkens for their use- national Conference on Formal Ontology in Information ful comments, Hans Uszkoreit for giving the initial impe- Systems (FOIS), pages 323–336. tus for the development of the biography ontology several Hans-Ulrich Krieger. 2013. An efficient implementation of years ago, and Bernd Kiefer for always having an open ear equivalence relations in OWL via rule and query rewrit- in many discussions. ing. In Proceedings of the 7th IEEE International Con- ference on Semantic Computing (ICSC), pages 260–263. 7 References Hans-Ulrich Krieger. 2014. A detailed comparison of Peter Adolphs, Xiwen Cheng, Tina Klüwer, Hans Uszkor- seven approaches for the annotation of time-dependent eit, and Feiyu Xu. 2010. Question answering biographic factual knowledge in RDF and OWL. In Proceedings of information and social network powered by the semantic the 10th Joint ACL-ISO Workshop on Interoperable Se- web. In Proceedings of the 7th International Conference mantic Annotation (ISA). on Language Resources and Evaluation (LREC), pages Patrick Le Bœuf. 2010. From FRBR to FRBROO through 2764–2768. CIDOC CRM. Slide presentation, October. Interna- Gavin Carothers and Andy Seaborne. 2014. RDF 1.1 N- tional Symposium on the Future of Information Organi- Triples. a line-based syntax for an RDF graph. Technical zation Research, Taipei, Taiwan. report, W3C. Frank Manola and Eric Miller. 2004. RDF primer. Techni- Donald Davidson. 1967. The logical form of action sen- cal report, W3C. tences. In Nicholas Rescher, editor, The Logic of Deci- John McCarthy and Patrick J. Hayes. 1969. Some philo- sion and Action, pages 81–95. University of Pittsburgh sophical problems from the standpoint of artificial intelli- Press. gence. In B. Meltzer and D. Michie, editors, Machine In- Fredo Erxleben, Michael Günther, Markus Krötzsch, Ju- telligence 4, pages 463–502. Edinburgh University Press. lian Mendez, and Denny Vrandec̆ić. 2014. Introducing Luc Moreau and Paolo Missier. 2013. PROV-DM: The 109 PROV data model. Technical report, W3C. W3C Rec- lines. http://www.bbc.co.uk/ontologies/storyline, May. ommendation 30 April 2013. Contributors: Paul Rissen, Helen Lippell, Matt Chad- Boris Motik, Bernardo Cuenca Grau, Ian Horrocks, Zhe burn, Tom Leitch, Dan Brickley, Michael Smethurst, Se- Wu, Achille Fokoue, and Carsten Lutz. 2012. OWL 2 bastien Cevey. web ontology language profiles. Technical report, W3C. W3C Recommendation 11 December 2012. Ian Niles and Adam Pease. 2001. Towards a standard up- per ontology. In Proceedings of the 2nd International Conference on Formal Ontology in Information Systems (FOIS), pages 2–9. Niels Ockeloen, Antske Fokkens, Serge ter Braake, Piek Vossen, Victor de Boer, Guus Schreiber, and Susan Legêne. 2013. BiographyNet: Managing provenance at multiple levels and from different perspectives. In Pro- ceedings of the 3rd International Workshop on Linked Science (LISC) at ISWC, pages 59–71. Terence Parsons. 1990. Events in the Semantics of English. A Study in Subatomic Semantics. MIT Press, Cambridge, MA. Stephen L. Reed and DouglaS B. Lenat. 2002. Mapping ontologies into Cyc. In Proceedings of the AAAI 2002 Conference Workshop on Ontologies For The Semantic Web, pages 1–6. Dave Reynolds. 2006. Jena rules tutorial. In Jena User Conference. PowerPoint presentation. Ansgar Scherp, Thomas Franz, Carsten Saathoff, and Stef- fen Staab. 2009. F—a model of events based on the foundational ontology DOLCE+DnS Ultralite. In Pro- ceedings of the 5th International Conference on Knowl- edge Capture (K-CAP), pages 137–144. James G. Schmolze. 1989. Terminological knowledge rep- resentation systems supporting n-ary terms. In Proceed- ings of the 1st International Conference on Principles of Knowledge Representation and Reasoning (KR), pages 432–443. Roxane Segers, Piek Vossen, Marco Rospocher, Luciano Serafini, Egoitz Laparra, and German Rigau. 2015. ESO: a frame based ontology for events and implied sit- uations. In Proceedings of MAPLEX. Ryan Shaw, Raphaël Troncy, and Lynda Hardman. 2009. LODE: Linking open descriptions of events. In Proceed- ings of the 4th Asian Semantic Web Conference (ASWC), pages 153–167. Richard T. Snodgrass. 2000. Developing Time-Oriented Database Applications in SQL. Morgan Kaufmann, San Francisco, CA. Herman J. ter Horst. 2005. Completeness, decidability and complexity of entailment for RDF Schema and a seman- tic extension involving the OWL vocabulary. Journal of Web Semantics, 3:79–115. Willem Robert van Hage, Véronique Malaisé, Roxane Segers, Laura Hollink, and Guus Schreiber. 2011. De- sign and use of the simple event model (SEM). Journal of Web Semantics, 9(2):128–136. Christopher Welty and Richard Fikes. 2006. A reusable ontology for fluents in OWL. In Proceedings of 4th FOIS, pages 226–236. Paul Wilton, Jeremy Tarling, and Jarred McGinnis. 2013. Storyline ontology: An ontology to represent news story- 110