Language : analysis & simulationSimulation specificationSimulation SemanticsActive representationsTaskEvent Structure for semantic QA Srini NarayananAspectI/O as Feature StructuresBasic ComponentsPowerPoint PresentationThe Target DomainBayes Nets and Human Probabilistic InferenceMetaphor MapsAn Active Model of EventsSlide 19Simulation hypothesisSlide 21Slide 22Slide 23Slide 24Inference from Domain knowledgeGeneral and Domain KnowledgeSlide 27Modeling Spreading ActivationA computational model: Bayes NetsBayes NetworksExample: AlarmA Simple Bayes NetAssigning Probabilities to RootsConditional Probability TablesSlide 35What the BN MeansCalculation of Joint ProbabilityWhat the BN EncodesSlide 39What can Bayes nets be used for?A Simple Bayes Net for the target domainA Simple Bayes Net for the target domain of Economic PolicyApproaches to inferenceProbabilistic graphical modelsStatesA Simple DBN for the target domainSlide 50Slide 52Slide 53Task: Interpret simple discourse fragmentsSlide 56Slide 57ResultsPsycholinguistic evidenceDiscussionLanguage understanding: analysis & simulationLanguage : analysis & simulation“Harry walked into the cafe.Analysis ProcessSemanticSpecificationUtteranceConstructionsLexiconGeneral KnowledgeBelief StateCAFESimulationconstruction WALKEDform selff.phon  [wakt]meaning : Walk-Action constraints selfm.time before Context.speech-time selfm..aspect  encapsulatedSimulation specificationThe analysis process produces a simulation specification that •includes image-schematic, motor control and conceptual structures •provides parameters for a mental simulationSimulation SemanticsBASIC ASSUMPTION: SAME REPRESENTATION FOR PLANNING AND SIMULATIVE INFERENCE Evidence for common mechanisms for recognition and action (mirror neurons) in the F5 area (Rizzolatti et al (1996), Gallese 96, Boccino 2002) and from motor imagery (Jeannerod 1996)IMPLEMENTATION: x-schemas affect each other by enabling, disabling or modifying execution trajectories. Whenever the CONTROLLER schema makes a transition it may set, get, or modify state leading to triggering or modification of other x-schemas. State is completely distributed (a graph marking) over the network.RESULT: INTERPRETATION IS IMAGINATIVE SIMULATION!Active representationsMany inferences about actions derive from what we know about executing themRepresentation based on stochastic Petri nets captures dynamic, parameterized nature of actionsWalking:bound to a specific walker with a direction or goalconsumes resources (e.g., energy)may have termination condition(e.g., walker at goal) ongoing, iterative actionwalker=Harrygoal=homeenergywalker at goalTaskInterpret simple discourse fragments/blurbsFrance fell into recession. Pulled out by GermanyEconomy moving at the pace of a Clinton jog.US Economy on the verge of falling back into recession after moving forward on an anemic recovery.Indian Government stumbling in implementing Liberalization plan.Moving forward on all fronts, we are going to be ongoing and relentless as we tighten the net of justice.The Government is taking bold new steps. We are loosening the stranglehold on business, slashing tariffs and removing obstacles to international trade.Event Structure for semantic QASrini NarayananReasoning about dynamicsComplex event structureMultiple stages, interruptions, resources, framingEvolving eventsConditional events, presuppositions.Nested temporal and aspectual referencesPast, future event referencesMetaphoric referencesUse of motion domain to describe complex events.Reasoning with UncertaintyCombining Evidence from Multiple, unreliable sourcesNon-monotonic inferenceRetracting previous assertionsConditioning on partial evidenceAspectAspect is the name given to the ways languages describe the structure of events using a variety of lexical and grammatical devices.Viewpointsis walking, walkPhases of eventsStarting to walk, walking, finish walkingInherent Aspectrun vs cough vs. rubComposition withTemporal modifiers, tense..Noun Phrases (count vs. mass) etc..I/O as Feature StructuresIndian Government stumbling in implementing liberalization planBasic ComponentsAn fine-grained executing model of action and events (X-schemas).A simulation of connected embodied x-schemas using a controller x-schemaA representation of the domain/frames (DBN’s) that supports spreading activationA model of metaphor maps that project bindings from source to target domains.The Target DomainSimple knowledge about EconomicsFactual (US is a market economy)Correlational (High Growth => High Inflation)Key Requirement:Must combine background knowledge of economics with inherent structure and constraints of the target domain with inferential products of metaphoric (and other) projections from multiple source domains.Must be able to compute the global impact of new observations (from direct input as well as metaphoric inferences)Bayes Nets and Human Probabilistic InferenceOur use of Bayes Networks will be to model how people reason about uncertain events, such as those in economics and politics. We know that people do reason probabilistically, but also that they do not always act in accord with the formal laws of probability. Daniel Kahneman won the 2002 Nobel Prize largely for his work with Amos Tversky explaining many of the limitations of human probabilistic reasoning. Some of the limitations are obvious, e.g. the calculations might be just too complex. But some are much deeper involving the way a question is stated, a preference for avoiding loss, and some basic misperceptions about large and small probabilities. Bayes nets only approximate the underlying evidential neural computation, but they are by far the best available model.Metaphor MapsStatic Structures that project bindings from source domain f- struct to target domain Belief net nodes by setting evidence on the target network.Different types of mapsPMAPS project X- schema Parameters to abstract domainsOMAPS connect roles between source and target domainSMAPS connect schemas from source to target domains.ASPECT is an invariant in projection.An Active Model of EventsComputationally, actions and events are coded in active representations called x-schemas which are extensions to Stochastic Petri nets.x-schemas are fine-grained action and event representations