SOCoP and Topic Overview John Moeller (24G4)
The workshop continues the SOCoP effort to foster collaboration and open standards for increased interoperability of spatial data across government among researchers, technologists & users of spatial knowledge representations and reasoning towards the development of spatial ontologies for use by all in the Semantic Web. A particular issue is whether a meaningful reference model can be crafted to guide work to accelerate the pace of advancement for the development and use of geospatial ontologies and related semantics. (24G5)
One issue raised by Jacqueline Knudson concerned outreach on this issue with the GWG metadata group as well as the National Core Vocabulary group. These and others are interested in Best Practices for building and using ontologies. (24G8)
Workshop Keynote Ontologies for Reasoning, Action and Interaction in Space John Bateman Professor of Applied Linguistics at the University of Bremen (24G6)
This talk draws from Batemans experience with Computational Linguistics, Formal Ontology and Spatial representation and language and the work of the SFB/TR8 team (http://www.sfbtr8.uni-bremen.de). (24G9)
John noted that many application domains, but surprisingly few, related useful architectural abstractions are agreed upon. The work discussed here is an attempt to close this gap motivated in part to better support applications like wheelchair mobility, partially-embedded task (assisted living), and free space maps generated from sensor data. Often available information (e.g. GPS: 34° 15´ N / 3° 27´ E) does not help with spatial situations and natural route descriptions using language require considerable semantic interpretation. A high degree of interoperability is difficult because different source/communities of geospatial information have different interests for developing their models in different context. Besides this they use different representations (or way-finding abstractions such as choremes) and so require different expertise to talk to each other. The lack of a foundation leads to a proliferation of standardisation efforts and developmental efforts going on over and over again by different groups. (24GA)
The Onto-Space (SFB/TR8 ) plan is to ontological engineer a highly structured and motivated semantics to achieve better interoperability by using commonsense knowledge from the natural human and synthetic robot world combined with geo-knowledge (i.e. GML, and other standards), spatial knowledge (spatial calculi, spatial ontologies) and linguistic knowledge. Work on a Spatial Ontology Baseline balances primitives (relations, concept of spatial region, shortcut) from general ontologies, linguistics and Qualitative Spatial Reasoning (QSR) for routing and paths. The taxonomies and classes of SUMO, CYC, and DOLCE were highlighted and contrasted. There are distinct theories and varying primitives in each of these ontologies (as well as the several QSR/Qualitative Calculi approaches (RCC8, Region Connecting Spaces, Double Cross Calculus-reasoning composition, star calculus, dipoles, OPRA and Qualitative Trajectory Calculus) that were reviewed and there is no easy merging of the various ontologies with the QSRs (or with spatial language). There remain many perspectives and therefore many ontologies and ways of doing spatial reasoning. For example DOLCEs general ontology represents the classes of agents, states of belief, plans, and goals and a class for quality. Others do not. (24GB)
The situation is no less complicated when we look at evidence from linguistic usage reflecting statements about perceived situations/relations in the world (e.g. the book in ON the table) Several of the meaning of in were shown from Herskovits (1986). The conclusion is that spatial language is sensitive to function and purpose as shown in the idea of an umbrella being over ones head or in the statement that an overalls curvy road is straight ahead. We need hyper-ontologies in the face of diverse perspectives/perspectivalism -a philosophical view that human ideations arise from particular perspectives so that there are naturally many possible conceptual schemes affecting the possible validity and relevance of our judgments. Such hyper-ontologies will help map ontologies of such things as time, event types, landmarks, routes etc. As a base for these DOLCE is used for cross-category binding and axiomatization (entities, qualities, physical regions), while is BFO is used for sites, niches and places and for SNAP/SPAN) and a generalized upper model (GUM) is used for linguistic semantics. In a DOLCE conceptualization space is a quality so we have space regions like we have color regions. This is valuable for `swappable treatments of space where we select formalization for the reasoning task at hand but have them linked/plugged to play with a DOLCE Physical Endurant (PED) with a Physical Quality (PQ) say Quality Space quale of Cardinal directions or Double Cross appropriate for a task at hand. We dont commit to just one, but preserve modular flexibility without sacrificing formality. (24GC)
This work draws on 2 Formal and computational tools: CASL and HETS. CASL (Common Algebraic Specification Language) is a standardised first-order language with useful constructs for specification, structuring and relating while HETS (Heterogeneous Tool Set) is used for connecting to the range of reasoners including First-Order Reasoning, description logic reasoning with DL reasoners and spatial reasoning with specialized spatial reasoners (SparQ, GQR). Together these provide somethings missing in Owl and the Semantic Web. CASL supports a family of logics (OWL-DL etc.) and is formalizing the entire family of qualitative spatial calculi. (For details see K. Lüttich, T. Mossakowski (2004). Specification of Ontologies in CASL. In Achille C. Varzi, Laure Vieu (Eds.), Formal Ontology in Information Systems -- Proceedings of the Third International Conference (FOIS-2004), Vol. 114, pp. 140150, Frontiers in Artificial Intelligence and Applications. ) (24GD)
Summary of Lessons and Conclusions from the work to date. 1. Geospatial Information may be set to become the next major area of ontological development, however, just converting existing schema to OWL is probably not going to be adequate because loose and sparse definitions in such schema do not give much to get hold of for relating distinct accounts/levels of abstraction (24GE)
2. A proper approach to ontologies and semantic precision rather than syntactic orientation of current metamodels enables access to detailed contextual world-knowledge that does not then have to be continually re-invented. This avoids: (24GF)
a. proliferation of unrelated designs, b. impoverished or application-specific semantics, c. roll your own ignoring previous attempts d. lack of interoperability (24GG)
3. Unlike much of the light ontologies of the semantic web proper ontologies are heavy in that they have: (24GH)
a. Rich axiomatization b. Formal principles c. Well-defined design criteria 4. Lesson for Ontological best design principles include the role of: a. axiomatization b. modularity c. heterogeneity d. perspectivalism (24GI)
5. Next steps include selected applications for assisted ambient living (AAL), geographic information science (GIS) and assisted architectural design (AAD) (24GJ)
PS See http://irtg-sigi.uni-muenster.de for applications for International Research Training Group: semantic integration of geospatial information. (24G7)