Taming Semantic Interoperability Demons Lurking in Military Training & Testing Events By Exploiting Semantic Web Technology David Hanz Senior Principal Engineer 20 March 2013 4955-1 Background – Challenge (1) • With the advent of Net-Enabled Capabilities, Precision Guidance, and similar technological advances, military operations have become exceedingly complex • But before conducting such operations: • The capabilities of the systems must be verified • The personnel using the systems must learn how to employ them effectively • Creating the environment -- that is both appropriate and affordable -- in which to conduct such complex testing and training events has become a significant challenge 4955-2 Background – Challenge (2) • General consensus has emerged on the approach: lash together various Live, Virtual, and Constructive (LVC) simulation systems in a confederation • Much progress has been made in standards & tools to simplify the formation of such LVC confederations (e.g., DIS, HLA, TENA) • But these only establish interoperability at the technical & syntactic levels (bits flow and message structures are mutually comprehended) • Systems can still reach wildly different conclusions from the same data 4955-3 Background – Challenge (2) • General consensus has emerged on the approach: lash together various Live, Virtual, and Constructive (LVC) simulation systems in a confederation • Much progress has been made in standards & tools to simplify the formation of such LVC confederations (e.g., DIS, HLA, TENA) • But these only establish interoperability at the technical & syntactic levels (bits flow and message structures are mutually comprehended) • Systems can still reach wildly different conclusions from the same data That’s a problem! 4955-4 Typical Source of the Problem • The root cause of these problems typically turns out to derive from different native semantics – different understandings of shared information objects A “Semantic Gap” • Semantic gaps are not necessarily troublesome… only when they are large enough to cause disagreement on a result derived from the same data • But determining whether they will cause a disagreement is “complicated” • It depends not only on properties of the specific systems involved, but also on what they are doing • And sometimes where, when, and other factors 4955-5 Semantic Interoperability Problem Example (LOS Fair Fight Issue) Live OPFOR hiding behind real terrain feature can’t be seen by Live Blue combatant But Virtual Blue combatant has no problem seeing avatar for Live OPFOR • This example (from a training exercise held in California) was traced to ~ 2 meters of tectonic plate movement since 1984 that was included in the native semantics of the Live system -- but not in native semantics of the Virtual system • If the exercise had been held in Kansas, this semantic gap would probably still be undiscovered 4955-6 6 ONISTT • The Open Net-centric Interoperability Standards for Training and Testing (ONISTT) was developed as a framework and toolsuite that facilitates planning LiveVirtual-Constructive (LVC) training exercises and test events that: • Satisfy a specific event’s training / testing objectives • Optimize the utilization of available resources • Avoids (or mitigates) pernicious problems resulting from Semantic Gaps among the participating systems • But the technology under the hood deals with a more general problem set • It can be (and is being) used to solve other kinds of problems that require logical reasoning 4955-7 How can ontologies & AI help with the thorny problem of semantic interoperability? • By capturing the knowledge about root causes of past Semantic Interoperability problems in a declarative form, and then… • Performing automated reasoning to predict if those kinds of issues might pop up in a new candidate confederation, and then… • For those cases where there appears to be a problem, determine if a relatively simple inline mediator (active gateway) could be synthesized to “pre-warp” the data-in-transit and bridge the semantic gap, or… • For those cases where the gap is too large to bridge, warn the human-in-charge to find another resource (or relax the expectations) 4955-8 Inline Mediator: Concept & Motivation • Mitigate semantic interoperability problems by reducing the size of the semantic gap for specific information objects exchanged between designated systems • But avoid the cost of changing (“fixing”) the native semantics of the systems • Intuitive example: “Corrective” lenses distort images to match the persons visual acuity deficit 4955-9 “Purpose-aware” Interoperability Analysis Event Knowledge Base Resource Knowledge Base •Task •Role •Capability Needed •Constraints •Confederation •Resource •Capability Capabilities Needed Capabilities Available Representation and analysis of capabilities needed is structured around tasks (i.e., description of purpose) that the systems are intended to perform 4955-10 Basic ONISTT Concept • Create semantic-rich Knowledge Bases (KBs) that describe: • The precise capabilities_needed to support interactions between roles associated with a specific task • The precise capabilities_available from candidate resources that may be available to play those roles • Employ a domain-agnostic inference engine to use the information from the capabilities_needed KBs as a template for performing reasoning against the capabilities_available KBs • Objective: Synthesize a confederation of resources that meets the specific needs of the exercise/event 4955-11 ONISTT CONOPS for Exercise Planning Automate the composition of LVC confederations Deployment Knowledge Bases • • • • • Resource pools Confederations Taskplans Role assignments Task constraints 3. Testing /Training Planner uses Knowledge Bases to a) Define Taskplans (full or partial) b) Propose candidate Confederation(s) (full or partial) • • • • Task Knowledge Bases Resource and Domain Knowledge Bases Tasks Roles Capabilities needed Task constraints • Resources • Capabilities • Domain knowledge Verified Taskplan(s) & Confederation(s) Develop formal ontologies for ONISTT core, DoD, domain and and general domain knowledge, suitable for machine reasoning 2. SMEs populate and maintain distributed Knowledge Bases with ontology-based descriptions of tasks & resources, 4. Analyzer uses information in Knowledge Bases to complete Taskplan and a) Assess given Confederation or b) Generate & rank possible Confederations from Resource Pool Analyzer Decision 5a. Return Taskplan with problem diagnosis and solution options. Back to Step 1 1. 5b. Return verified Taskplan(s), Confederation(s), and Configuration Artifacts for Mediator Configuration Artifacts 4955-12 Analyzer/Synthesizer Architecture • Leverages standards-based semantic and logical reasoning technologies • Knowledge captured declaratively in Web Ontology Language (OWL) + Semantic Web Rule Language (SWRL) • Prolog well suited to the kind of reasoning we need to do with tasks – COTS Description Logic (DL) reasoner engines inadequate • Task Engine is implemented as meta-interpreter. Task plan is proof tree • Can be hosted as a web service 4955-13 ONISTT Top-level Ontology (simplified) 4955-14 Example Domain Ontology (Spatial Reference Frame -- partial) 4955-15 Accomplishments vs. Remaining Issues • The technical feasibility of the ONISTT concept has been demonstrated by conducting a half-dozen, increasingly complex experiments using real tasks and real systems • The Analyzer/Synthesizer renders solutions that are at least as good as the traditional BOGSAT approach • Practicality-related issues remain an open question, and are the focus of our current efforts • These include: • Making the semantic artifacts (ontologies and rules) accessible to domain subject matter experts (SMEs) who are not also Semantic Technology Experts (STEs) • Tools to help deal with the well known “Knowledge Acquisition Bottleneck” for legacy systems • Standards, patterns, & best practice guidelines to allow useful semantic artifacts to be obtained as part of new system acquisitions 4955-16 OMG activities that have helped • The Ontology Definition Metamodel (ODM) has allowed development of tools (like the Visual Ontology Modeler) that scratches a portion of the “Make Accessible to SMEs” itch 4955-17 Possible new activity for OMG • There is a dearth of standards, patterns, & best practice guidelines that would allow useful semantic artifacts to be obtained as part of new system acquisitions • Given prior history (e.g., ODM) OMG seems like the most logical choice as the standards-making body to pursue that goal 4955-18 Conclusions • Ontologies/Rules provide a means to express key capabilities needed and key capabilities available from component resources • Formal declarative expression understandable by a machine • Can be extended to an arbitrary level of granularity • AI inference engine technology provides a tool that can • Discover/Synthesize resource compositions tailored for a given purpose and determine if known interoperability defeaters are potentially present • Determine if potential problems can be mitigated by in-line data mediation • Although Description Logic is inherently “black/white” (not fuzzy) at the atomic level, the ONISTT framework provides means to reason about “gray areas” at the molecular level • A necessity since many of the important issues cannot be reduced to strictly-true or strictly-false facts 4955-19 4955-20