Spatial Knowledge Representation for Human-Robot Interaction
Shared by: tck21090
Connecting Language to the World: Ontology Baseline Scott Farrar Project I1[OntoSpace] Brief overview of I1[Ontospace] I1 is a subproject within the SFB/TR Spatial Cognition: Reasoning, Action, Interaction other members: John Bateman, Kerstin Fischer, Reinhard Moratz, Thora Tenbrink I1 primary deliverable Ontologies for spatiallyaware systems and NL components. A registry of interontology mappings that allows for communication between spatiallyaware systems and NL components. Modeling space in the SBF/TR To illuminate the relationship between the spatiallyaware systems and the natural language components (world/language). To facilitate the information exchange between individual spatiallyaware systems (software integration). To investigate certain issues concerning spatial reasoning in robots. e tio f -[Otopc] thr ro c sin n loie Rla noI1 n Saetoo ep jetsu goto g s Goals of presentation To present a review of the major ontologies available for reuse within the SFB and beyond Discuss some basic parameters important to ontology building in general. Discuss the parameters important for the SFB— spatial aspects of ontology building. Major issues within the ontology community Philosophical approach Metalevel distinctions Ontological primitives Set theory vs. mereology Representation Philosophical approach What are the categories and relations supposed to be modelling? the worldrealism (Husserl, Ingarden, Smith) conceptsconceptualism (commonsense realism) (Hayes, Guarino et al.) Metalevel distinctions formal vs. material level in ontology member, familyMember perspective endurantism (3D), perdurantism (4D) granularity microscopic, mesoscopic, macroscopic Ontological primitives Intraontological universals, particulars subsumption, instantiation Transontological [more later… ] Set theory vs. mereology domainindependent tools/theories set theory (ST) Primitives are those from ST: union, intersection, subset, urelements. mereology Primitives are those from classical mereology: part, interiorPart Representation logical language expressiveness computational tractability firstorder logic (SUOKIF, CycL) description logic (OWL) The leading candidates OpenCyc SUMO DOLCE BFO OpenCyc Cycorp http://www.opencyc.org 6,000 categories and relations 60,000 assertions about those c’s and r’s. sort of free bundled with the OpenCyc KB system OpenCyc Based on set theory— everything is either an instance or subset of Thing. Thing partitioned into: SetOrCollection: the collection of sets; instances have members, not parts. and Individual: the collection of individuals; instances have parts, not members. OpenCyc— Top level OpenCyc Entities in OpenCyc are considered to be “concepts”, not necessarily dependent on the human mind— commonsense view of the world. OpenCyc is actually mixture of realism and conceptualism. Compare FoodGroup to SpatialThing Localized OpenCyc Intricate arrangement of all categories, relations, and assertions in the system of microtheories. Everything in OpenCyc is a member of some microtheory. KB partition based on shared assumptions, fields of study, and/or sources of knowledge. See taxonomy of microtheories… OpenCyc and more than 200 others… OpenCyc summary Philosophical approach— mixed (realism, conceptualism) Metalevel distinctions (formal, material, highly granular) Ontological primitives (microtheories) Set theory vs. mereology— strongly set theoretic Representation— CycL (firstorder logic) SUMO Suggest Upper Merged Ontology http://ontology.teknowledge.com 1019 categories and relations, 4181 axioms, and 822 implication rules. Developed by IEEE working group, administered by Teknowledge sort of free Partial SUMO taxonomy Entity isa isa Physical Abstract Object Process Relation Attribute Proposition Region Agent Quantity Collection SetOrClass SelfConnected Object SUMO Informally modular structural ontology base ontology set theory mereotopology numeric, temporal objects, process SUMO Encoded in SUOKIF (<=> (instance ?PHYS Physical) (exists (?LOC ?TIME) (and (located ?PHYS ?LOC) (time ?PHYS ?TIME)))) “All physical entities are located in time and space” SUMO summary Philosophical approach— mixed, less focus on realism Metalevel distinctions— formal, material Ontological primitives Set theory vs. mereology— set theory Representation— SUOKIF (firstorder logic) DOLCE Discriptive Ontology for Linguistic and Cognitive Engineering http://www.loacnr.it/index.html Based on a particular research line within ontology (Guarino, Masolo, Gangemi, Colomb, et al.) Aimed at supporting the Semantic Web (Wonderweb project) DOLCE DOLCE Axioms in DOLCE are represented in KIF and OWL. Initial axiomatization of DOLCE appears to be aimed the very upper formal level, in contrast to SUMO. An example… DOLCE (D1) RG: Rigid Universal (defrelation RG (?w0 ?f) := (and (UNIVERSAL ?f) (WORLD ?w0) (forall (?w ?x) (=> (and (WLDR ?w0 ?w) (WORLD ?w) (PARTICULAR ?x)) (=> (?f ?w ?x) (forall (?u) (=> (and (WLDR ?w ?u) (WORLD ?u)) (?f ?u ?x)))))))) DOLCE Introduces qualities and quality regions. DOLCE summary Philosophical approach— conceptualism, though more realism than OpenCyc/SUMO. Metalevel distinctions— mostly formal Ontological primitives Set theory vs. mereology— strongly mereological Representation— KIF (firstorder logic); OWL (but only the taxonomy+) BFO Basic Formal Ontology http://www.ifomis.unileipzig.de/ Developed by IFOMIS in Leipzig Smith, Grenon, et al. Actually a framework of ontologies SNAP and SPAN BFO “Perspectivalism maintains that there may be alternative, equally legitimate perspectives on reality. But perspectivalism is constrained by realism: …” SNAP vs. SPAN BFO— SNAP Time is not present. SNAP entities have other SNAP entities as parts. Parthood behaves simply and reliably. the totality of space, my_hand, color_red, point_233 BFO BFO— SPAN SPAN entities are necessarily temporal. SPAN entities have other SNAN entities as parts. Your childhood is part of your life; you are not a part of your life. Time itself, spacetime, the year 1971, the shaving of my beard BFO BFO— relations (SNAP, SPAN), e.g., participant, as in ‘the jumper is a participant in the jumping event’ (SNAP, SNAP), e.g., part, as in ‘the leg is part of the table’ (SPAN, SNAP), e.g., creation, as in ‘burning creates ashes’ (SPAN, SPAN), e.g., as in ‘the poor driving caused an accident’ BFO summary Philosophical approach— realist Metalevel distinctions— formal Ontological primitives— transontological relations Set theory vs. mereology— strongly mereological Representation— no implementation yet, but will likely be some variant of first order logic SFBspecific issues Spatial categories and relations Reference systems Reasoning support others? Quick overview of space Space has been approached either from the standpoint of: traditional geometry— a set of points or region calculus— a collection of primitive regions Areas of spatial concern qualitative reasoning surfaces, edges, boundaries topology— theory of boundaries, contact, and separation (also mereotopolgy) commonsense reasoning with objects spatial relations— beyond above, beside… path and reference systems The leading candidates OpenCyc SUMO DOLCE (will not be presented) BFO (will not be presented) RCC (will not be presented) OpenCyc— Space The structure of OpenCyc’s spatially related categories are difficult to visualize. lattice structure cut across microtheories Nevertheless, spatial entities and relations can be discussed separately. SpatialThing spatiallyRelated OpenCyc Subsumed by SpatialThing BiologicalLivingObject GeometricThing Border Localized CavityInteriorGeneric InformationBearingThing CavityOrContainer PartiallyTangible CloudlikeObject PathSpatial Configuration Place CustomarySystemOfLinks SomethingToWear EmptyRegionGeneric SpaceRegionEmpirical EventLocalized SpatialPathSystem Food Trajectory GeographicalThing OpenCyc— SpaceRegion SpaceRegion is subsumed by SpatialThing… OpenCyc— Space OpenCyc— Relations spatiallyRelated aligned connectedTo hasBeenIn parallelObjects notFarFrom connectedTo spatiallyDisjoint physicalPartsdisjoint onSamePlanetSurfaceAs OpenCyc— Granularity touches touches (cont.) aboveTouching hangsFrom groundOf hangsAround inFloating inHeld onPhysical inImmersedGeneric adjacentTo inImmersedFully bordersOn inImmersedPartly spaceRegionPortals inSnugly alignedAlongSurface touchesAtEnd connectedAlongSurface touchesDirectly sheetSurfaceTouches wearer connectedAtContact wearsClothing connectedAlongSurface wornOn connectedAtContact connectedAtEnd inEmbedded pipesDirectlyConnected OpenCyc— Path systems OpenCyc has a rich inventory of categories related to paths and path systems, based on graph theory. OpenCyc— PathSystem PathSpatial can be anything… EdgeOnObject PathCustomary CordlikeObject FlowPath PathArtifact Portal Doorway PipeEndToCavityJunction WindowPortal SolidSurfacePathThroughAir PathForWheeledVehicles Railroad Roadway OpenCyc— Observations OpenCyc is neatly, sometimes not so neatly, partitioned into microtheories. covers many subdomains related to space. is axiomatically impoverished, though future releases are supposed to include the axioms. OpenCyc— Observations Whereas OpenCyc has a rich inventory of spatial categories and relations, it seems more like an extension of a lexical resource such as WordNet than an ontology. OpenCyc is useful as a measure for breadth, though the quality of breadth is questionable in many cases. Particularly useful may be OpenCyc’s rich vocabulary for paths and path systems. SUMO— Space SUMO— Parts and wholes SUMO— Parts and wholes (<=> (instance ?OBJ SelfConnectedObject) (forall (?PART1 ?PART2) (=> (equal ?OBJ (MereologicalSumFn ?PART1 ?PART2)) (connected ?PART1 ?PART2)))) “A SCO is the sum of its parts which are all connected.” SUMO— Space SUMO— Regions (=> (instance ?REGION Region) (exists (?PHYS) (located ?PHYS ?REGION))) “If something is a region, then there is something located in that region.” SUMO— Relations SUMO spatial relations are partially based on the Allen axioms: SUMO— Attributes SUMO— Parts SUMO— Observations SUMO contains considerable breadth, but also depth concerning certain aspects of space. particularly rich concerning relations. axioms are included (Allen axioms for space). leans less towards the lexical semantics of English than OpenCyc. Summary The aims of the SFB/TR have been introduced, with particular emphasis on the creation of ontologies for spatially aware systems. The basics of four ontologies have been overviewed: OpenCyc, SUMO, DOLCE, and BFO. Spatial aspects of OpenCyc and SUMO have been presented. Conclusions No single ontology satisfies the needs of the SFB/TR in terms of an ontology for space. OpenCyc and SUMO lack the theoretical consistency needed for scaling up to detailed software integration. Both Open and SUMO rely on categories and relations more appropriate for the lexical semantics of English, not a general spatial ontology. Conclusions Neither DOLCE nor BFO are developed enough to evaluate in terms of space. Even so, certain aspects of the reviewed ontologies should be considered as starting points for constructing a general ontology for space. Conclusions OpenCyc— microtheories, path systems, breadth SUMO— commonsense mereology, example axioms, spatial relations DOLCE— qualities/q. regions, basic mereology BFO— perspectivalism, transontological relations, basic mereology, theories of space/time Next steps Consider the needs of the particular projects within the SBF/TR; nail down a methodology for incorporating experimental results. Determine weather or not DLs are adequate. Survey available ontologies for application towards a linguistically motivated ontology (add GUM, GOLD). Survey methods to relate ontologies; define the relationship between linguistically motivated ontology and ontology for space.