This well-written paper proposes a model of a system of events, as opposed to the existing models of (more or less) isolated events. Scherp et al.’s model supports structural aspects--that is, relationships--and separates the generic structural knowledge about events and objects from the knowledge of a specific application domain. The authors apply events “to capture and represent human experience,” as opposed to low-level signals and actions, and observe that the existing models “are conceptually narrow and their semantics is typically ambiguous.”
Thus, the most interesting characteristics of this event model are in the definitions of the relationships between events. Indeed, the authors introduce and describe “mereological, causal, and correlation relationships,” and substantially use a subtyping relationship. They also introduce viewpoints, or interpretation patterns, which are based on “the context and point of view of the observer.”
Unfortunately, the authors neither use nor reference the standard definitions [1]. The authors’ descriptions of generic relationships are arguably less clear and less explicit than the definitions of subtyping, composition, and reference relationships in the standards [1]. In fact, Scherp et al. fail to explicitly describe or define subtyping, and the causal and correlation relationships may be defined using the composition and reference relationships. Their descriptions of generic relationships are obviously event-specific, without referring to the actual concepts.
If the authors’ semantics definitions were clear and explicit, it would have been impossible for the authors to state, for example, in Section 4.1 that “correlation refers to two events that have a common cause,” and then in Section 5.4 state that a “set of events is called correlated if they have a common cause.” Furthermore, Scherp et al. do not explicitly refer to property determination as the most important characteristic of generic relationships (such as composition), although they tacitly acknowledge the existence and importance of this characteristic.
From the authors’ descriptions and Figure 2, it is apparent that their patterns--participation, mereology, causality, correlation, documentation, and interpretation--have important similarities, including similar generic and pattern-specific relationships among the components. It would certainly be worthwhile to discover a common pattern or patterns that generalize these, especially since the authors include modularity and reusability among nonfunctional requirements for their model.
In Section 5, Scherp et al. refer in passing to “the n-ary relation that exists between multiple individuals of events and objects.” Various generic relationships may be considered as different types of this relation; it is quite possible, for example, to generalize the authors’ composition of events only, and consider--in their participation and mereology patterns--an event composed of other events and of objects participating in that event. In accordance with the standard definition of composition [1], the properties of the composite are defined by those components and by the way they are combined. Although various situations and descriptions may also be decomposed in this manner, the paper does not describe this.
Finally, it is difficult to agree with the authors’ statement that “event-based systems need to be able to automatically check the validity of the exchanged knowledge ... with respect to the semantics,” for two reasons: validation is and can only be done by humans, and we are still too far from “formal precision” in the definition of semantics.
In summary, while this appears to be a nice and interesting model, it needs further development, preferably with existing standards.