Cyc Representing Acquiring and Using Knowledge People have

Cyc Representing, Acquiring and Using Knowledge “People have silly reasons why computers don’t really think. The answer is we haven’t programmed them right; they just don’t have much common sense. There’s been only one large project to do something about that, that’s the famous Cyc project…” - Prof. Marvin Minsky MIT, May, 2001 Dr. Michael Witbrock Cycorp, RER d. o. o. http: //www. cycorp. eu/ witbrock@cycorp. eu May 24 th, 2007

1984: Increase human capabilities by building the first true Artificial Intelligence. Revised: Increase human capabilities by teaching the first true Artificial Intelligence to build itself. Cycorp Corporate Mission witbrock@cyc. com

1984: Increase human capabilities by building the first true Artificial Intelligence. Further Revised: Increase human capabilities by teaching the first true Artificial Intelligence to parts of itself in Slovenia. Cycorp d. o. o Corporate Mission witbrock@cyc. com

The Cyc Ontology and Content Tens of thousands of Predicates between, owns, isa, beliefs Millions of Assertions Thing “Causes precede effects” Sets Hundreds of thousands of Concepts Relations Spatial Temporal (Forall x (Forall y Thing Partially (implies City. Of. Ljubljana. Slovenia, Space Paths Tangible Time Thing (causes x y) Witbrock, Slovenia, Events Spatial Logic Paths Math Scripts (starts. After. Start. Of y x)))) Slovenian. Language Physical Borders Agents Artifacts Intangible Individual Thing Objects Living Things Ecology Natural Geography Life Forms Human Beings Geometry Materials Parts Statics State Change Dynamics Plants Actors Actions Movement Plans Goals Physical Agents Organizational Actions Organizational Plans Types of Organizations Human Activities Human Organizations Human Nations Business & Politics Agent Anatomy & Governments Animals Commerce Warfare Organizations Physiology Geo-Politics Human Products Conceptual Emotion Sports Perception Behavior & Social Purchasing Professions Devices Works Weather Law Recreation Belief Actions Shopping Occupations Behavior Entertainment Vehicles Mechanical Software Social Buildings & Electrical Literature Language Relations, Business, Earth & Social Transportation Travel Everyday Weapons Devices Works of Art Culture Military Solar System Activities & Logistics Communication Living Organizations Political Geography Human Artifacts General Knowledge various domains Specific data, facts, and observations about those domains witbrock@cyc. com

Millions of facts, rules of thumb, etc. that capture human common sense about our everyday world ØThe typical bird has 1 beak, 1 heart, lots of feathers, … ØHearts are internal organs; feathers are external protrusions ØMost vehicles are steered by an awake, sane, adult, … human ØTangible objects can’t be in 2 (disjoint) places at once ØBadly injuring a child is much worse than killing a dog ØCauses temporally precede (i. e. , start before) their effects ØA stabbing requires 2 co-temporal and proximate actors ◦ etc. , … Cyc is… Cycorp © 2007

Upper Ontology Core Theories ( a, b ) a EVENT b EVENT causes( a, b ) precedes( a, b ) Domain-Specific Theories Very specific information (some indirect, via SKSI) ( m, a ) m MAMMAL a ANTHRAX causes( exposed-to( m, a ), infected-by( m, a ) ) (ist Ft. Laud. Holy. Cross. ERCase#403921 (caused Cutaneous. Anthrax (Skin. Lesions Ahmed_al-Haznawit))) First Order Predicate Calculus: unambiguous; enable mechanical reasoning Every Slovene has a president. Every Slovene has a mother. y. x. Slov(x) president(x, y) x. y. Slov(x) mother(x, y) Higher Order Logic (nth-order predicate calculus): contexts, predicates as variables, nested modals, reflection, … Cyc Reasons via Formal Logic Cycorp © 2007

Cyc contains: >15, 000 Predicates >300, 000 Concepts >3, 500, 000 Assertions Intangible Individual Thing Sets Relations Space Physical Objects Living Things Ecology Natural Geography Political Geography Weather Earth & Solar System Paths Human Anatomy & Physiology Actors Actions Movement State Change Dynamics Plants Temporal Thing Partially Tangible Thing Logic Math Borders Geometry Plans Goals Physical Agents Animals Emotion Human Products Conceptual Perception Behavior & Devices Works Belief Actions Vehicles Buildings Weapons Spatial Thing Spatial Paths Materials Parts Statics Life Forms Human Beings Human Artifacts Represented in: • First Order Logic • Higher Order Logic Time • Context Logic Events Scripts • Micro-theories Agents Artifacts Thing Mechanical Software Social Language Relations, & Electrical Literature Devices Works of Art Culture Organizational Actions Organizational Plans Agent Organizations Social Behavior Organization Social Activities Human Activities Business & Commerce Purchasing Shopping Types of Organizations Politics Warfare Sports Recreation Entertainment Transportation & Logistics Human Organizations Nations Governments Geo-Politics Professions Occupations Travel Communication Law Everyday Living Business, Military Organizations General Knowledge about Various Domains The Cyc Knowledge Base Specific data, facts, and observations Cycorp © 2007

Cycorp © 2007

Cyc KB Extended w/Domain Knowledge Thing Intangible Individual Thing Sets Relations Spatial Thing Temporal Thing General Knowledge about Terrorism: Space Paths Partially Tangible Thing Time Terrorist groups are capable of directing assassinations: Events Spatial Logic Paths Math Scripts (implies Agents Physical Borders Artifacts Geometry (isa ? GROUP Terrorist. Group) Objects (behavior. Capable ? GROUP Assassinating. Someone directing. Agent)) Materials Living Organ. Actors Parts Movement Actions Statics Things ization … If a terrorist group considers an agent State enemy, Plans agent is vulnerable to an attack by that group: an Change that Life Organizational Types of Ecology Dynamics Actions Organizations Goals Forms (implies Human Physical Natural Organizational Human (and Plants Geography Plans Agents Beings Activities Organizations (isa ? GROUP Terrorist. Group) Human Nations Human Political Agent Business & Politics Anatomy & Governments Animals (considers. As. Enemy ? GROUP ? TARGET)) Artifacts Geography Organizations Commerce Warfare Physiology Geo-Politics Emotion Human Sports (vulnerable. To ? GROUPPerception Behavior &Terrorist. Attack)) ? TARGET Social Products Conceptual Purchasing Professions Weather Law Recreation Devices Earth & Solar System Vehicles Buildings Weapons Works Belief Actions Behavior Mechanical Software Social Language Relations, & Electrical Literature Devices Works of Art Culture Shopping Social Activities Entertainment Transportation & Logistics Occupations Travel Communication Everyday Living General Knowledge about Terrorism Specific data, facts, and observations Cycorp about terrorist groups and activities © 2007 Business, Military Organizations

Cyc KB Extended w/Domain Knowledge Thing Intangible Individual Thing Sets Relations Space Specific Facts about Al Qaida: Physical Spatial Thing Paths Spatial Paths Borders Temporal Thing Partially Tangible Thing Logic Math Time Events Scripts Agents Artifacts (based. In. Region Al. Qaida Afghanistan) Objects Geometry Al-Qaida is based in Afghanistan. (has. Belief. Systems Al. Qaida Islamic. Fundamentalist. Beliefs) Al-Qaida has Islamic fundamentalist beliefs. Materials Living Organ. Actors Parts Actions Statics Things ization (has. Leaders Al. Qaida Osama. Bin. Laden) Al-Qaida is. Movement Osama bin Laden. led by … State Change Plans Life Organizational Types of Ecology Dynamics Actions Organizations Goals Forms (affiliated. With Al. Qaida Al. Quds. Mosque. Organization) Al-Qaida is affiliated with the Al Quds Mosque. (affiliated. With Al. Qaida Sudanese. Intelligence. Service)Physical Organizational Human the Sudanese Intell Service Al-Qaida is affiliated with Human Natural Plants Geography Plans Agents Beings Activities Organizations … Human Nations Human Political Agent Business & (sponsors Al. Qaida Harakat. Ul. Ansar) Al-Qaida sponsors Harakat ul-Ansar. Politics Anatomy & Governments Animals Artifacts Geography Organizations Commerce Warfare Physiology Geo-Politics (sponsors Al. Qaida Laskar. Jihad) Al-Qaida sponsors Social Jihad. Laskar Emotion Human Sports Products Conceptual Purchasing Professions Perception Behavior & Weather Law Recreation Devices Works Shopping Occupations Behavior … Belief Actions Entertainment Vehicles Mechanical Software Business, (performed. By Embassy. Bombing. In. Nairobi Language Social Al-Qaida bombed the Embassy in Nairobi. Al. Qaida) Social Transportation Travel Everyday Earth & Buildings & Electrical Literature Relations, Military Activities & Logistics Communication Living Solar System Weapons Devices Works of Art Culture Organizations (performed. By Embassy. Bombing. In. Tanzania Al. Qaida) Al-Qaida bombed the Embassy in Tanzania. General Knowledge about Terrorism Specific data, facts, and observations Cycorp about terrorist groups and activities © 2007

Knowledge Users Other Applications User Interface (with Natural Language Dialog) Cyc Reasoning Modules Cyc API Knowledge Entry Tools Knowledge Authors Cyc Ontology & Knowledge Base Interface to External Data Sources Data Bases Web Pages Text Sources Other KBs Cyc High-level Architecture Cycorp © 2007

Cyc is not monolithic The Knowledge Base is divided into thousands of contexts by: granularity, topic, culture, geospatial place, time, . . . Cyc is not committed to any one reasoning mechanism The inference engine has more than 1, 000 “agents” that attack every problem and, recursively, every sub-problem. They have specialised representations and methods to quickly solve common problems. Soon: Distributed, Parallel Reasoning Multiple Contexts and Reasoners Cycorp © 2007

Cyc is not monotonic 98% of its content is marked as merely being usually true. Reasoning in Cyc is default (gather up pro/con arguments, and compare them). Soon: Probabilistic Reasoning Cyc is not committed to its own reasoning mechanisms Additional reasoning modules can include any manner of external databases, simulators, translators… Multiple Contexts and Reasoners Cycorp © 2007

Crossover Point: Cyc learns iscovery g via d learnin by e g faster by itself ng langua rni natural lea than by manual knowledge entry Google books project finished rate of learning In this area, empirical discovery and learning via NL come together. codify & enter each piece of knowledge, by hand d wle CYC ge no nk a um er ti ron h of F amount known Today: Acquiring Knowledge witbrock@cyc. com

Cyc Today Reasoning via deduction, abduction, and induction; access to other reasoning capabilities (through heuristic modules) Growing ability to support Knowledge Acquisition and use via Natural Language Research Platform: Open. Cyc, Research. Cyc General-purpose platform: Supports applications involving search, question answering, knowledge management, scenario generation, situation reporting, (re-)planning, etc. witbrock@cyc. com

Cyc Analytical Environment Michael Witbrock © Cycorp 2007

Content and Representation Drill-down Michael Witbrock © Cycorp 2007

Temporal Relations Senses of “x is a physical part of y” Senses of “x is physically in y” Events and their performers (role types) Organizations Propositional Attitudes Biology Materials Devices Weather Cyc KB Sampler Cycorp © 2007

37 Relations Between Temporal Things #$temporal. Bounds. Intersect #$temporal. Bounds. Contain #$temporally. Intersects #$temporal. Bounds. Identical #$starts. After. Starting. Of #$starts. During #$ends. After. Ending. Of #$overlaps. Start #$starting. Date #$starting. Point #$temporally. Contains #$simultaneous. With #$temporally. Cooriginating #$after Temporal Relations Cycorp © 2007

Some of these Relations are very General, such as: #$temporally. Intersects Such relations are particularly useful when they are known not to hold between a pair of individuals: (#$not (#$temporally. Intersects ? X ? Y)) That implies all of these: (#$not (#$spouse PERSON-X PERSON-Y)) (#$not (#$consultant AGENT-X AGENT-Y)) (#$not (#$account. Holder ACCOUNT-X AGENT-Y)) (#$not (#$resides. In. Region AGENT-X REGION-Y)) (#$not (#$officiator EVENT-X PERSON-Y)) Temporal Relations Cycorp © 2007

#$parts #$intangible. Parts #$sub. Information #$sub. Events #$physical. Decompositions #$physical. Portions • • • #$physical. Parts #$external. Parts #$internal. Parts #$anatomical. Parts #$constituents #$functional. Part Senses of ‘Part’ Cycorp © 2007

Can the inner object leave by passing between members of the outer group? • Yes -- Try #$in-Among Senses of ‘In’ Cycorp © 2007

Does part of the inner object stick out of the container? ◦ None of it. -- Try #$in-Cont. Completely ◦ If the container were turned around could the contained object fall out? Yes -- Try #$in. Cont. Open – ◦ Yes -- Try #$in-Cont. Partially ◦ No -- Try • #$in-Cont. Closed Senses of ‘In’ Cycorp © 2007

Is it attached to the inside of the outer object? – Yes -- Try #$connected. To. Inside Can it be removed by pulling, if enough force is used, without damaging either object? – No -- Try #$in-Snugly or #$screwed. In Does the inner object stick into the outer object? –Yes – Try #$sticks. Into Senses of ‘In’ Cycorp © 2007

#$Physical. State. Change. Ev ent #$Temperature. Changing. Pr ocess #$Biological. Development. E vent #$Shape. Change. Event #$Movement. Event #$Changing. Device. State 11, 000 more #$Giving. Something #$Discovery. Event #$Cracking #$Carving #$Buying #$Thinking #$Mixing #$Singing #$Cutting. Nails #$Pumping. Fluid Event Types Cycorp © 2007

#$performed. By #$causes-Event #$object. Placed #$object. Of. State. Change #$outputs. Created #$inputs. Destroyed #$assisting. Agent #$beneficiary #$from. Location #$to. Location #$device. Used #$driver. Actor #$damages #$vehicle #$provider. Of. Motive Force Over 400 more. #$transportees Relations Between Events and Participants Cycorp © 2007

Here are some roles for Attack 874 isa: Terrorist. Attack. performed. By: Jihad. Group. device. Used: Bomb 8388. event. Occurs. At: City. Of. London. England. victim: Person 9399. victim: Person 52666. assisting. Agent: Al. Qaeda. objects. Destroyed: Structure 2990. objects. Destroyed: Vehicle 523452. Representation of A Sample Event Cycorp © 2007

日本 語 Nederla nds Polski ქართულ ი Ελληνικά 中文 Italiano Portuguê s עברית Българс ки Русский Español Français ﺍﻟﻌﺮﺑﻴﺔ 한국 어 ( x) (isa x Living. Language) ( y) (isa y Person) (language. Spoken y x) Deutsch English ﻓﺎﺭﺳی ব ল Natural Language Drill-down Michael Witbrock © Cycorp 2007

English Words Syntactic Frame Links Single-word Denotation Mappings Multi-word Phrase Denotation Mappings 18, 796 23, 336 27, 681 44, 298 Verbal Semantic Frame Links Noun Semantic Frame Links Word. Net 2. 0 Links Names 3, 701 2, 578 (Includes chemical symbols, person/place/organization names, acronyms, etc. ) 11, 322 100, 811 Predicate-based Phrasal Links 9, 637 (gen. Templates for paraphrase) Cyc NL Lexicon Cycorp © 2007

Constant : Coke-The. Word isa : English. Word Mt : English. Mt singular : “coke” mass. Number : “coke” pn. Singular : “Coke” pn. Mass. Number : “Coke” (denotation Coke-The. Word Proper. Count. Noun 0 (Serving. Fn Coca. Cola)) G LAN (denotation Coke-The. Word Proper. Mass. Noun 0 S G LAN G S Coca. Cola) LAN S (denotation Coke-The. Word Mass. Noun 0 Cocaine. Powder) (denotation Entry Example: Cola. Soft. Drink) Lexical. Coke-The. Word Mass. Noun 2 0 Coke (denotation Coke-The. Word Simple. Noun (Serving. Fn Cycorp © 2007 Cola. Soft. Drink))

Constant: Eat. The. Word isa: English. Word Mt: English. Mt infinitive: “eat” perfect: “eaten” past. Tense: “ate” agentive-Sg: “eater” (subcat. Frame Eat-The. Word Verb 0 Transitive. NPComp. Frame) (verb. Sem. Trans Eat-The. Word 0 Transitive. NPComp. Frame (and (isa : ACTION Eating. Event) (performed. By : ACTION : SUBJECT) (inputs. Destroyed : ACTION : OBJECT))) Lexical Entry Example: Eat Cycorp © 2007

Blah blah al queda blah embassy blah grenade blah suicide attack • Concept spotting [Zaquawi] was involved in kidnapping [seven israeli children] • Extraction Templates A polygraph, also known as a lie detector, is an [electrical device] • EBMT: syntax templates The heart consists of four chambers, the two upper atria (singular: atrium) and the two lower ventricles. • Last resort, full parsing CYC Knowledge Reduces Work Cycorp © 2007

Document Tagging … Document Tagging

Entire Cyc KB

◦… warplanes ◦ bombers (warplanes) B 24 bomber B-1 bombers B-2 stealth bombers B-29 Superfortress B-52 bombers … A-5 C fighter planes A 10 fighter plane F-117 Nighthawks F-14 fighter plane F-15 eagles F-16 falcons … ◦ fighter planes Military Taxonomy

“The plants that produced the cranes that NASA deployed in space in the 1990 s are in Canada. ” The plants (#$Factory. Building. Complex 0. 8817 #$Plant 0. 0967) that produced (#$Production-Generic 0. 6017) the cranes (#$Crane-Motorized. Device 0. 9387 #$Crane. Bird 0. 0408) that NASA (#$NASA) deployed (#$Deploying. Material) in space (#$Outer. Space 0. 51 #$Space. In. AHOC 0. 1473 #$Reserved. Space. Region 0. 0459 #$Area 0) in the 1990 s ((#$Decade. Fn 199) 1) are in Canada (#$Canada 1). Probabilistic Concept Tagging

Simple Example: #$isa “… natural resources, of which oil and diamonds are the most relevant. ” “oil” • #$Oil • #$Petroleum-Crude. Oil • #$Artist. Oil. Paint • #$Petroleum. Product “diamonds” • #$Diamond-Gem • #$Diamonds-Suit #$isa licence Looks for collections in the text of which a given object is an instance “natural resources” • #$Natural. Resource. Type #$siblings. WRTType licence Looks for collections in the text that share a type Knowledge for Disambiguation

Example: “common sense” “The heart pumps blood to the lungs. ” “heart” #$type. Behavior. Capable Cyc Knows: The heart is a kind of pump. • #$Heart-Suit Are there any pairs Cyc Knows: Every pump can pump fluid. • #$Heart-Locus. Of. Feeling such that every C 1 is capable of playing a • #$Center. Of. Region role in an event of type C 2? • #$Heart “pumps” • #$Pumping. Fluid • #$Pumping-Making. Something Available Knowledge Driven Disambiguation

Renaissance Artists Bronze Age Farmers (Subcollection. Of. With. Relation To. Fn Artist active. During. Period The. Renaissance) (Subcollection. Of. With. Relation To. Fn Farmer active. During. Period The. Bronze. Age) Kind of Time. Interval Kind of Agent-Generic Noun Form: not plural Noun form Noun Compounds More Complex Parsing Cycorp © 2007

Train +----------------------Xp----------------------+ +------Wd------+ +----------MVp-----------+ | | +----A----+ | +------Jp-----+----Mp----+ | | +--G--+--G-+--Ss--+---Os---+--Mp-+ +--Dmcn--+ +N Sa+ +-Js-+ | | | | LEFT Royal. a Dutch Shell Plc halted. v output. n of 455, 000 barrels. n a day. p in Nigeria. (#$and (#$isa (#$The. Fn #$Decrease. Event) (#$Decrease. In. Value. Returned. By. Fn (#$Export. Rate. Of. By. Fn #$Petroleum-Crude. Oil) #$Nigeria)) (#$done. By (#$The. Fn #$Decrease. Event) #$Royal. Dutch. Shell) (#$quantity. Change. Amount (#$The. Fn #$Decrease. Event) (#$Barrels. Per. Day 455000))) Template +----------------------Xp----------------------+ +------Wd------+ +----------MVp-----------+ | | +------Jp-----+ | | | +------+--Ss--+---Os---+--Mp-+ + +-Js-+ | | | | | LEFT [Agent] halted. v output. n of [Quantity] in [Locn]. (#$and (#$isa (#$The. Fn #$Decrease. Event) (#$Decrease. In. Value. Returned. By. Fn (#$Export. Rate. Of. By. Fn #$Petroleum-Crude. Oil) [Locn])) (#$done. By (#$The. Fn #$Decrease. Event) [Agent]) (#$quantity. Change. Amount (#$The. Fn #$Decrease. Event) [Quantity])) Use Petróleos de Venezuela S. A. halted output of 760 000 barrels a week in Maracaibo. (#$and (#$isa (#$The. Fn #$Decrease. Event) (#$Decrease. In. Value. Returned. By. Fn (#$Export. Rate. Of. By. Fn #$Petroleum-Crude. Oil) #$City. Of. Maracaibo. Venezuela )) (#$done. By (#$The. Fn #$Decrease. Event) #$Petroleosde. Venezuela. SA (#$quantity. Change. Amount (#$The. Fn #$Decrease. Event) (#$Barrels. Per. Day 760000)))

Mohamed Atta Knowledge Acquisition Drill-down Michael Witbrock © Cycorp 2007

… Klingberg contacted the USSR for the first time in 1957, and soon after that he started his espionage activity. Israel's foreign and domestic intelligence agencies, Mossad and Shin Bet, started suspecting Klingberg of espionage, but shadowing brought no results. At one point, the scientist also successfully passed the Unknown word “polygraph” Device-Physical genls Polygraph polygraph test… Page Download EBMT Parser Sentence Extractor Wikipedia No page found Uninformative sentence Success (#$genls #$Polygraph #$Device. Physical) CYC Unable to parse Semantic Checker Hypothesis not logically consistent Automatically Adding to the Model Cycorp © 2007

Query “What are symptoms of Whooping Cough? ” (symptom. Of. Ailment Whooping. Cough ? SYMP ) NL Generation Partial English sentences “A symptom of whooping cough is ___” “Whooping cough can cause ___” “A symptom of Pertussis Bordetella is ___” “Symptoms (such as ____) of whooping cough” Learning Facts by Search Michael Witbrock © Cycorp 2007

Looking for something that matches the argument constraints on the predicate… “… symptoms of pertussis such as fever and a dry cough …” Parse back into existing Cyc. L concepts (symptom. Of. Ailment Whooping. Cough Fever) (symptom. Of. Ailment Whooping. Coughing-Ailment. Condition) Parsing Results C. Matuszek, R. C. Kahlert , M Witbrock FACTory © Cycorp 2007

• Explicitly: perform one step of inference to throw out facts inconsistent with KB • Implicitly: don’t even look at things that don’t match argument constraints Throw out provably wrong answers Skip already known (provably right) knowledge KB Consistency Check C. Matuszek, R. C. Kahlert , M Witbrock FACTory © Cycorp 2007

Given a set of events that Cyc already knows about… January 15, 2006 Group of Pirates 1 Piracy Event 1 Somalia date. Of. Event January 20, 2006 MV Delta Ranger Group of Pirates 3 perpetrator date. Of. Event perpetrator event. Occurs. Near intended. Attack. Targets Group of Pirates 2 Piracy Event 2 event. Occurs. Near February 18, 2006 date. Of. Event event. Occurs. Near intended. Attack. Targets Philippines Nigeria Piracy Event 3 device. Used Speed Boat 1 MV Man Chu Yi …recognize new instances of that event type in text Malacca Straits: On 17 April 2006, a Malaysian fishing vessel was attacked by armed pirates at approximately nine nautical miles off Parit Haji Baki coast in the Malacca Straits at about 0200 Hrs LT. Six pirates armed with guns in a speedboat closed in rapidly and opened fire at the fishing vessel underway. Several shots hit the side of the vessel but the crew escaped injuries. The fishing vessel crew lodged a police report. New Piracy Event Probabilistic Event Extraction

determines appropriate relations for event type perpetrator Piracy Event 1 perpetrator date. Of. Event event. Occurs. Near intended. Attack. Targets date. Of. Event perpetrator event. Occurs. Near Piracy Event 2 event. Occurs. Near Piracy Event 3 device. Used intended. Attack. Targets that occur in >1/2 of the known events. P(device. Used) = 1/3 = 0. 33 p(perpetrator) = 3/3 = 1. 0 p(date. Of. Event) = 3/3 = 1. 0 p(event. Occurs. Near) = 3/3 = 1. 0 p(intended. Attack. Targets) = 2/3 = 0. 66 intended. Attack. Targets New Piracy Event perpetrator date. Of. Event event. Occurs. Near

…look at role fillers for known events… January 15, 2006 Group of Pirates 1 Piracy Event 1 Somalia date. Of. Event January 20, 2006 MV Delta Ranger Group of Pirates 3 perpetrator February 18, 2006 date. Of. Event perpetrator event. Occurs. Near intended. Attack. Targets Group of Pirates 2 Piracy Event 2 event. Occurs. Near Piracy Event 3 event. Occurs. Near intended. Attack. Targets Philippines Nigeria device. Used Speed Boat 1 MV Man Chu Yi and find similar types of concepts mentioned in the text. Malacca Straits: On 17 April 2006, a Malaysian fishing vessel was attacked by armed pirates at approximately nine nautical miles off Parit Haji Baki coast in the Malacca Straits at about 0200 Hrs LT. Six pirates armed with guns in a speedboat closed in rapidly and opened fire at the fishing vessel underway. Several shots hit the side of the vessel but the crew escaped injuries. The fishing vessel crew lodged a police report. intended. Attack. Targets New Piracy Event ? ? ? perpetrator ? ? ? date. Of. Event ? ? ? event. Occurs. Near ? ? ?

Concepts in Cyc’s ontology are found in the text Things People & Org. s Places Malacca Straits Malaysia Dates Parit Haji Baki April 17, 2006 People Vehicles Pirates Speed Boats Some Speed Boat Malacca Straits: On 17 April 2006, a Malaysian fishing vessel was attacked by armed pirates at approximately nine nautical miles off Parit Haji Baki coast in the Malacca Straits at about 0200 Hrs LT. Six pirates armed with guns in a speedboat closed in rapidly and opened fire at the fishing vessel underway. Several shots hit the side of the vessel but the crew escaped injuries. The fishing vessel crew lodged a police report. Police Some Police Officer Some Pirate

The concepts associated with the known events are also mapped to the ontology. Things People & Org. s Places Malacca Straits Somalia Pirates January 20, 2006 Malaysia Parit Haji Baki Philippines People Vehicles Dates April 17, 2006 Org. s February 18, 2006 January 15, 2006 Speed Boats MV Man Chu Yi MV Delta Ranger Police Some Speed Boat 1 Groups of Pirates Group of Pirates 1 Group of Pirates 3 Some Police Officer Group of Pirates 2 Some Pirate

Collections are associated with each relation… Things People & Org. s Places Dates People Vehicles Org. s Somalia Philippines The most relevant Collections are those where the probability of seeing them is high for concepts that have this relation to an event of this type, and have low probability of being seen otherwise. P(isa(X, Thing) | (event. Occurs. Near EVENT X)) = 1 P(isa(X, Place) | (event. Occurs. Near EVENT X)) = 1 P(isa(X, Date) | (event. Occurs. Near EVENT X)) = 0 P(isa(X, Vehicle) | (event. Occurs. Near EVENT X)) = 0 P(isa(X, Person or Org. ) | (event. Occurs. Near EVENT X)) = 0 P(isa(X, Person) | (event. Occurs. Near EVENT X)) = 0 P(isa(X, Org. ) | (event. Occurs. Near EVENT X)) = 0 P(isa(X, Group of Pirates) | (event. Occurs. Near EVENT X)) = 0 New Piracy Event event. Occurs. Near ? ? ?

Probabilities can be estimated for extracted concepts… Things People & Org. s Places Malacca Straits Malaysia Dates People Vehicles Pirates Org. s Parit Haji Baki April 17, 2006 Speed Boats Some Speed Boat Police Groups of Pirates Some Police Officer to measure how well they fit the relation. p(Malacca Straits) = 1 p(Malaysia) = 1 p(Parit Haji Baki) = 1 p(April 17, 2006) = 0 p(Speed Boats) = 0 P(Some Speed Boat) = 0 … event. Occurs. Near New Piracy Event Some Pirate Parit Haji Baki event. Occurs. Near Malacca Straits Malaysia

After repeating this process for every relation, choosing relation/concept pairs with >0. 5 probability, a potential event has been extracted from the text. Some Speed Boat intended. Attack. Targets date. Of. Event Parit Haji Baki New Piracy Event April 17, 2006 Malacca Straits: On 17 April 2006, a Malaysian fishing vessel was attacked by armed pirates at approximately nine nautical miles off Parit Haji Baki coast in the Malacca Straits at about 0200 Hrs LT. Six pirates armed with guns in a speedboat closed in rapidly and opened fire at the fishing vessel underway. Several shots hit the side of the vessel but the crew escaped injuries. The fishing vessel crew lodged a police report. event. Occurs. Near Malacca Straits Malaysia

Probabilities for each relation/concept pair can then be used to estimate the overall probability of the extracted event being correct. Some Speed Boat (1. 0) intended. Attack. Targets (0. 67) Parit Haji Baki (1. 0) Piracy Event (0. 67) event. Occurs. Near (1. 0) date. Of. Event (1. 0) April 17, 2006 (1. 0) Malacca Straits: On 17 April 2006, a Malaysian fishing vessel was attacked by armed pirates at approximately nine nautical miles off Parit Haji Baki coast in the Malacca Straits at about 0200 Hrs LT. Six pirates armed with guns in a speedboat closed in rapidly and opened fire at the fishing vessel underway. Several shots hit the side of the vessel but the crew escaped injuries. The fishing vessel crew lodged a police report. Malaysia (1. 0) Malacca Straits (1. 0)

Test corpus of 27 piracy documents. Extractor found 578 relation/concept pairs. 34. 78% (201) were correct. Good enough to feed human validation tools for partially automated data entry. Initial Results C. Matuszek, R. C. Kahlert FACTory © Cycorp 2006

By giving the extractor feedback about what concepts it got correct and incorrect, the precision of the extractor can be greatly increased. In initial experiments, giving feedback on the 27 piracy paragraphs raised precision from. 39 to. 61 using 2 -fold crossvalidation. Using Human Feedback

ØLarge vocabularies with detailed semantics are useful ØMore and more of the conceptual structure and knowledge will be automatically learned ØThe more you know, the less you need to understand ØDetailed knowledge really is needed for meaningful reasoning Wrap-up 2007 © Cycorp