CEVO
Contents
CEVO: Comprehensive Event Ontology
Abstract Description While the general analysis of named entities has received substantial research attention, the analysis of relations over named entities has not. In fact, literature review of works on unstructured data as well as structured data revealed a deficiency in research on abstract conceptualization to organize relations. We believe that such an abstract conceptualization can benefit various communities and applications such as natural language processing, information extraction and ontology engineering. Here, we present CEVO (i.e., a comprehensive event ontology) that is built upon Levin's conceptual hierarchy of English verbs. This conceptual hierarchy categorizes verbs with the shared meaning or behavior. We present the fundamental concepts and requirements for this ontology. Furthermore, we present three use cases for demonstrating the benefits of this ontology on annotation tasks. The first use case concerns annotating relations in plain text. The second one annotates ontological properties of a background data model. The third one links textual relations to properties of the background data model.
You can access the ontology here [1]
Beth Levin Conceptual Hierarchy
The entries of Levin's lexical knowledge base are verb classes whose members are English verbs having two characteristics: (i) semantically coherent and (ii) shared syntactic behavior.
(i) Shared meaning: Each class of verbs shows a unique set of properties that shape the meaning of the member verbs. In fact, the conjunction of properties provides a distinctive meaning for each class. A single meaning property might be attributed to several classes depending upon context. Moreover, an individual verb might belong to multiple classes, creating a graph instead of a tree.
(ii) Shared syntactic behavior: Meaning influences the syntactic behavior of a verb in terms of its expression and interpretation of arguments. Verbs with shared meaning exhibit similar syntactic behavior. Figure below illustrates two distinct English verb classes (1) transformation and creation and (2) change of the state that both subsume several verbs (e.g., cook, boil). Thus, depending on the context of the shared verbs, the appropriate class is distinguished.
CEVO Ontology
The core class of CEVO is the class of Generic Event that is the superclass of all specific events. The Generic Event class is formally defined as follows: The `Generic Event' is an owl:Class and refers to `occurrence of anything'. It generally is the superclass of any specific type of event.
cevo:GEvent a owl:Class . cevo:GEvent rdfs:label `generic event' . cevo:GEvent rdfs:comment `something that happens' .
In CEVO, the Levin conceptual hierarchy is incorporated under the Generic Event class. In other words, any class provided for a set of English verbs revealing a specific event is considered as an owl:Class. Formally as: Class of `X' Event: `X' Event is a subclass of the class GEvent. Conceptually it refers to a specific type of event that is associated with an English verb category sharing a common behavior or meaning. For instance, the class communication given below is defined as a subclass of Generic Event. This class refers to occurrence of any activity for communicating or transferring message/idea.
cevo:Communication a owl:Class ;
rdfs:subClassOf cevo:GEvent ;
rdfs:label `communication' ;
rdfs:comment `communication and transfer of idea' .
The next main class is cevo:MainVerb that refers to words with part of speech as verb. This class is equivalent to the class of main verb of OLiA [2] ontology. OLiA [3] is an annotation model based on morphology.
cevo:MainVerb a rdf:Class . cevo:MainVerb owl:equivalentClass OLiA:MainVerb .
CEVO Instance Level (Verb Individuals)
So far, we described the schema level classes; the next important step is to map each individual English verb to the corresponding event class. Thus, we instantiate each English verb at the instance level and type this verb primally as cevo:MainVerb and map it to the associated event. In the following, two English verbs say and cook are defined (i.e., with the type cevo:MainVerb) and furthermore, they are typed to their corresponding event classes Communication and Creation and Transformation respectively. However, each individual verb might be associated with several event classes. For instance, the verb cook, in addition to the class Creation and Transformation, is also associated with the event Change of the State.
(a) cevo#say rdf:type cevo:MainVerb . (a) cevo#say rdf:type cevo:Communication .
(b) cevo#cook rdf:type cevo:MainVerb . (b) cevo#cook rdf:type cevo:Creation_Transformation. (b) cevo#cook rdf:type cevo:Change_of_the_state.
Use Cases
Use Case 1: Annotating Relations in Text
CEVO can promote annotating relations in plain text. Figure in below shows two headline news on Twitter. The first tweet was published by BBC and the second one was published by New York Times. Tweet #1 is headed by the verb announce and the tweet#2 is headed by the verb say. Both of these tweets are similar in the sense that a message is transferred. Annotating these two tweets via CEVO enables us to obtain the same tag communication for both of these verbs, whereas the two verbs announce and say do no hold lexical relations such as synonymy.
Use Case 2: Annotating Properties of Ontologies
CEVO can be utilized for annotating properties of any ontology. One way of providing such an annotation is using the Web Annotation Data Model (WADM, W3C Working Draft 15 October 2015, [4]) which is a framework for expressing annotations. A WADM annotation has two elements (i) a target which indicates the resource being annotated and (ii) the body which indicates the description. Annotating properties of various ontologies according to CEVO addresses integration and alignment problems. Assume that we have the property <http://dbpedia.org/ontology/spouse> from DBpedia ontology that represents the relation of marrying that is semantically equivalent to the class CEVO:Amalgamate. The annotation of this property is presented in Turtle syntax using WADM framework as follows:
example:annotation1 a oa:Annotation ;
oa:hasTarget <http://dbpedia.org/ontology/spouse> ;
oa:hasBody CEVO:Amalgamate .
Use case 3: Linking Relations
CEVO facilitates linking occurrences of relations in plain text to properties in the background knowledge base. We continue with the following example. On the 4th March 2016, BBC published this headline: Rupert Murdoch and Jerry Hall marry. The embedded relation in this part of text is marry. This relation is annotated as CEVO:Amalgamate employing CEVO ontology. We show this annotation using WADM in the following.
example:annotation2 a oa:Annotation ;
oa:hasTarget example:headline1#marry ;
oa:hasBody CEVO:Amalgamate .
