Semantic Mediation across Communities and Sensors

Leaders

Werner Kuhn (Münster), John Bateman (Bremen)

Problem

Geospatial information is obtained from a variety of sources, typically sensors and other signal sources, and is used by an equally varied range of communities, including humans working in different disciplines, social communities, and artificial systems. The complete information flow involved is complex and involves varying levels of abstraction, granularity, and organization. Moreover, the process is not unidirectional but embedded in an active action-perception cycle where data from a source will be used by a destination that might change the next data gathering step from the source. Each source has its own particular structures and performs measurements in its own way. Similarly, each type of destination may differ in its structure and in the ways it needs to reason with the information. At higher levels of abstraction, geospatial information is used among and across communities of users that may differ considerably concerning the categories and relationships they use for constructing their respective worlds. These differing organizations arise with respect to communities defined in several ways. Most broadly, we can see communities arising with respect to distinct cultures with distinct conventions and needs—as for example in the contrasts observable between cultures living in areas where water is a transient resource (deserts with mostly dry river courses) and those where water features are permanent (rivers, lakes); such communities divide up the geographic world quite differently. Further significant distinctions stem from different scientific communities and disciplines—land terrain information requires distinct organizations and constructs than navigation information, water authorities require distinct information to wildlife organizations, and so on. In addition to these complexities, we also find uncertainty across the entire measurement and reasoning process; sensor readings, for example, are often uncertain and have to be combined with other uncertain sensor readings. This uncertainty propagates to more abstract levels and so must be treated as an integral part of the entire active perception-action cycle: sensory information influences rules and knowledge on abstract levels and vice versa. To summarize: geospatial information originates increasingly in sensors and the individual types of sensors employed commit to their own particular abstractions over the environmental features measured. Starting from this most basic level environmental modeling and moving on to the decision-making demands of distinct user communities requires the semantic integration of sensor data both across distinct sensor data models and across information communities.

Challenge

Providing models for the integration of sources and destinations of geospatial information and the communication of this information over different levels of abstraction is the major challenge addressed in this initiative. Appropriate models need to rely on an active-perception loop and to integrate geospatial information from basic levels such as sensors and across communities of users. Enabling communication among such heterogeneous groupings demands an unprecedented degree of semantically rich content. To achieve this, semantic reference systems must be constructed in such a way that individual modeling demands are met, but interoperability and mediation across information communities remain possible. This kind of mediation among users and between users and systems is increasingly modeled in terms of inter-ontology mappings. Applying this approach to geospatial information and mediation across communities and sensor models requires the design, construction, and evaluation of formal ontologies of language, space and everyday ‘common sense’ domains and the detailed investigation of how best these constructs are to be placed in relation to one another. A consideration of distinct cultural and disciplinary starting points also provides insight into the flexibility of mappings required. The general goal of this research initiative is then to address the central problem of how communities with different perspectives on the world can come to exchange and share information.

Possible PhD Topics

  • what ontological models specify the sensing process?
  • to what extent are ontological modules for distinct user communities consistent with foundational ontologies (BFO, DOLCE, SUMO, etc.)?
  • what classes of formal relationships are necessary for relating generic foundational ontologies, user community or task-specific ontological specifications of geographic space and objects, and the language requirements of those user communities? Can information be ‘translated’ via formal mappings so as to be made appropriate for consumption by different communities?
  • how can the formal inter- and intra-ontological relationships across perspectives, layers and partitions be constrained? Are there particular classes of tasks, domains, users that allow relationships to be pre-configured so as to support more effective reasoning? To what extent do static alignments between ontologies suffice and to what extent are more dynamic ‘negotiations’ of correspondences required? How can such negotiations be informed ontologically?
  • what does measurement theory offer for a formal treatment of observations in ontologies?
  • what are suitable ontology-based representations for measurement units and value types?
  • what is needed to ontologically capture the nature of observables like direction, speed or flow rates?
  • how can the interpretation of observables be constrained?
  • how practical is the idea of defining semantics through a loop linking observations to actions to observations of resulting changes?
  • which types of uncertainty in the context of sensor readings exist and how can they be formally described?
  • how does visual and auditory information processing occur and determine meaning in biological und technical systems?
  • how can sensory information be integrated into knowledge-based processes?