KR Meeting Jan 06 Summary

This is version 37. It is not the current version, and thus it cannot be edited.
[Back to current version] [Restore this version]

Meeting Summary

The Knowledge Representation working group met for three days in San Antonio, TX, with the aim to produce a core ontology for ecological observation. The proposed scope of the observation ontology for ecology (OBOE) was to capture the information/concepts needed to combine or determine the differences/similarities between two data-sets. The proposed uses for such as ontology are consistency checking, data annotation and data integration. This scope and uses of OBOE will be elaborated in a formal document that will be appended here in the near future.

During the first day, discussion centered on what information is necessary to distinguish between two data, and resulted in the development of a model that captured the "entity" observed, the "trait" measured, and the "measurement" taken in terms of value and dimensionality. The first day model can be found here, but will not be discussed in any detail due to fundamental changes in subsequent days of the meeting.

The second day focusing on capturing "measurement" of an observed trait (e.g., biomass or height). This notorious issue raised much debate about dimensionality, units, standards and culminated in the idea of dimension components (the ability to build an observed dimensionality via a simple component construct in OWL). For example, an ecologist may want to capture a measurement with units of rabbits per fox as one observation or grams phosphorus per grams carbon. To deal with this in a generic framework, a measurement would have a value and an observed dimensionality, which is like a "standard" that is mapped to the observation via a value (e.g., an observation of 2 meters would be mapped to the standard "meter" and the value "2"). An observed dimensionality is built from one or more dimension components and each component has a base dimension (e.g., gram), a power, and a categorical dimension, which is just an observed entity (e.g., carbon). Therefore, grams phosphorus per grams carbon can be built using two dimension components: base dimension = gram, power = 1, categorical dimension = phosphorus; base dimension = gram, power = -1, categorical dimension = carbon. The Day 2 model including the dimension component construct can be found here. Also included in the figure is a preliminary model for observation context that was originally frame in terms of an observation window (the contextual window through which an observation was made), which is termed observation domain.

The third day primarily focused on observation context and prompted several changes to the "entity/trait" model developed on the first day. The days debate centered about how context should be incorporated into the observation ontology and resulted in several contextual constructs. The first introduced the idea of using associative and mereological properties to relate observations (e.g., biomass isAssociatedWith Species, Species isContainedIn Plot, and so forth), and debate focused on the level at which such a contextual "builder" should be placed, i.e., entity, trait or measurement. A variant of this model can be found here. A second variant was then discussed in which some class names were changed (e.g., observed entity became observable, and observed trait became observable trait), but the main difference was that the contextual construct was moved to act upon the trait. This model can be found here.

During the final half-day, the model changed fundamentally again. Continuing debate about the difference between entities and traits culminated in modeling them as the same thing: observables. This move greatly simplified the ontology (below). Therefore, an observation has a subject (an observable) and subsumes measurement (also an observation). Because traits are also observable, then an observation can have a trait observation that uses the same construct. Further, an observation has context also being an observation with an observable. This model greatly simplifies the concept of observation and renders it flexible to a range of different observation structures (data-sets). This final model was the final product of the working group and can be viewed here. Definitions for the main classes and properties are also given here. The next step is to build use case scenarios which are being outlined here. The final model also allows for modularization of domain specific concepts that can be termed observables. Therefore, domain models of observables can be plugged in to the observable class when annotating data. Such domain models include spatial and temporal concept ontologies and the ecological concepts ontology. The "units" ontology will be plugged into the measurement construct. We divided these tasks, which are outlined here.

Observation: An assertion that an entity (observable) was observed. The observable is the subject of the observation. An observation may be composed of other observations serving as traits that are required to define (observe) the subject entity. An observation may also contribute to defining the context for another observation.

Observable: A particular entity that can be observed to have one or more characteristics, or traits.

Measurement: An observation that assigns a value and dimension to the observable.

Properties

hasObservation: To observe a observable your statement of what the subject is may require observing other subjects. This relationship is the way you connect the observation of those traits to the overall observation. The traits further define the overall observable.

hasSubject: Relates the observable to the observation. States the overall subject of the observation.

hasContext: States that one observation serves as the context for another observation. Defines a semantic relationship between two observables that is a fundamental aspect of the observations, but not of the observables themselves. For example, most measurements are accomplished in a spatio-temporal framework that might be valuable context.