250 likes | 355 Views
From Question-Answering to Information-Seeking Dialogs. Jerry R. Hobbs Artificial Intelligence Center SRI International Menlo Park, California. (with Douglas Appelt, Chris Culy, David Israel, David Martin, Srini Narayanan, Mark Stickel, and Richard Waldinger). Key Ideas.
E N D
From Question-Answeringto Information-Seeking Dialogs Jerry R. Hobbs Artificial Intelligence Center SRI International Menlo Park, California (with Douglas Appelt, Chris Culy, David Israel, David Martin, Srini Narayanan, Mark Stickel, and Richard Waldinger)
Key Ideas 1. Logical analysis/decomposition of questions into component questions, using a reasoning engine 2. Use of component questions to drive subsequent dialogue, for elaboration, revision, and clarification 3. Bottoming out in variety of web resources and information extraction engine 4. Use of analysis of questions to determine, formulate, and present answers. Principal Investigator: Jerry R. Hobbs, SRI International
Composition of Informationfrom Multiple Sources GEMINI SNARK How far is it from Muscat to Kandahar? Question Decomposition via Logical Rules What is the distance between the two lat/longs? What is the lat/long of Muscat? What is the lat/long of Kandahar? Resources Attached to Reasoning Process Alexandrian Digital Library Gazetteer Alexandrian Digital Library Gazetteer Geographical Formula or www.nau.edu/~cvm/latlongdist.html Principal Investigator: Jerry R. Hobbs, SRI International
An Information-Seeking Scenario How safe is the Muscat harbor for refueling US Navy ships? Question Decomposition via Logical Rules Are relations between Oman and US friendly? What recent terrorist incidents in Oman? Asking User is one such Resource How secure is the Muscat harbor? Resources Attached to Reasoning Process Ask Analyst IR + IE Engine for searching recent news feeds Find map of harbor from DAML-encoded Semantic Web/Intelink Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? Open Domain Information Extraction System Purchase: Agent: Patient: Date: Location: Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? parent(y,x) subsidiary(x,y) Subsidiaries: XYZ: ABC, ... DEF: ..., XYZ, ... DB Purchase: Agent: XYZ, ABC, DEF, ... Patient: Date: Location: Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? illegal User Model biowarfare Purchase: Agent: XYZ, ABC, DEF, ... Patient: anthrax, ... Date: Location: DB of bio-equip Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? Ask User Purchase: Agent: XYZ, ABC, DEF, ... Patient: anthrax, ... Date: since Jun01 Location: Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? not USA Filter Information Extraction System invoked Purchase: Agent: XYZ, ABC, DEF, ... Patient: anthrax, ... Date: since Jun01 Location: -- Principal Investigator: Jerry R. Hobbs, SRI International
A Complex Query What recent purchases of suspicious equipment has XYZ Corp or its subsidiaries or parent firm made in foreign countries? parent(y,x) illegal not USA Ask User subsidiary(x,y) Subsidiaries: XYZ: ABC, ... DEF: ..., XYZ, ... biowarfare Purchase: Agent: XYZ, ABC, DEF, ... Patient: anthrax, ... Date: since Jun01 Location: -- DB of bio-equip Principal Investigator: Jerry R. Hobbs, SRI International
A Dialog What recent purchases of suspicious equipment has XYZ Corp made? illegal Ask User biowarfare Purchase: Agent: XYZ Patient: anthrax, ... Date: since Jun01 Location: DB of bio-equip Principal Investigator: Jerry R. Hobbs, SRI International
A Dialog What recent purchases of suspicious equipment has XYZ Corp made? How about its subsidiaries or parent firm? illegal Ask User parent(y,x) subsidiary(x,y) Subsidiaries: XYZ: ABC, ... DEF: ..., XYZ, ... biowarfare Purchase: Agent: ABC, DEF, ... Patient: anthrax, ... Date: since Jun01 Location: DB of bio-equip Principal Investigator: Jerry R. Hobbs, SRI International
Parsing Queries Use GEMINI Parser, Grammar, and Semantic Interpretation Complete for fronted wh-questions, yes-no questions Minor augmentations may be necessary Map current logical forms into SNARK expressions Principal Investigator: Jerry R. Hobbs, SRI International
Resolving Indeterminacies Determine precise interpretation of vague query: Specific interpretations for general predicates Resolving coreference and syntactic ambiguities Expanding metonymies Interpreting simple metaphors Use abduction capabilities of SNARK: Proof of query is answer, but also more precise interpretation of query Principal Investigator: Jerry R. Hobbs, SRI International
Decomposing Questions Decompose questions into subquestions via inference: P1 & P2 --> Q Axioms direct system toward specific available resources Principal Investigator: Jerry R. Hobbs, SRI International
Articulating with Resource Ontologies Axioms link language of general reasoning with language used by specific resource: Pspec --> Pgen Possible resources: Specific highly useful web sites DAML-encoded web sites in general FASTUS-based open-domain information extraction engine Quick&dirty Q&A system Principal Investigator: Jerry R. Hobbs, SRI International
Invoking Resources Resources will be invoked through Procedural attachment features of SNARK Open Agent Architecture (OAA) Principal Investigator: Jerry R. Hobbs, SRI International
Encoding Axioms Multiple knowledge bases will be used and linked to through the same kind of articulation axioms. Possible knowledge bases: CIA World Fact Book (partially axiomatized) DAML ontologies Structural Evidential Argumentation System (SEAS) (Genoa program) HPKB and RKF knowledge bases Subsets of CYC Core theories in some critical domains will be developed in this project Principal Investigator: Jerry R. Hobbs, SRI International
Filtering Responses from Resources Often an analyst requires specific information, but the resource can only be queried on general criteria Inference engine will filter answers from resource by more specific constraints Principal Investigator: Jerry R. Hobbs, SRI International
Structuring Answers Recompose answers to subquestions into answer to whole, using the structure of the proof graph that decomposed the question Construct hierarchically organized answers Enables drill-down and explanation Principal Investigator: Jerry R. Hobbs, SRI International
Structuring Dialog Proof structure of query is likely to be mirrored by structure of subsequent dialog -- details, fine modifications, clarifications, and elaborations Query may be a subquestion in a larger query that emerges in subsequent dialog -- proof tree of subsequent queries locates place for earlier query Principal Investigator: Jerry R. Hobbs, SRI International
Contributions:AQUAINT Areas of Interest Question-Answering: by inferential question-decomposition, bottoming out in calls to web and other resources. Information-seeking dialogs with subtopics driven by question-decomposition. Determining the Answer: by using the proof graph to recompose the information discovered in question analysis. Formulating and Presenting the Answer: by using the proof graph to structure the responses. Principal Investigator: Jerry R. Hobbs, SRI International
Contributions:Enabling Technologies Advanced Reasoning for Question-Answering Interactive Question-Answering Role of Context Sharable Knowledge Sources Content Representation Role of Knowledge Language Processing Our Primary Foci in AQUAINT Related SRI Projects + Strong Background Strong Background + Strong Software Base + Secondary Focus in AQUAINT Principal Investigator: Jerry R. Hobbs, SRI International
Summary Inference provides a uniform framework for Analyzing questions into their components Linking to external resources Recomposing answers Structuring information-seeking dialogs Principal Investigator: Jerry R. Hobbs, SRI International