OpenCycBY BENJAMIN SUSMAN

Introductions Commonsense knowledge:

Outline Background information

OpenCyc details

Cyc Ontology

CycL – Cyc’s language

NLP in Cyc

Setting up OpenCyc

Demo

What is Cyc?

1. Rich knowledge modeling language

2. Large corpus of formally modeled knowledge with a wide breadth

3. Efficient inference engine

Written in LISP and CycL (Cyc language)

Designed as a way to provide common sense or world knowledge and provide methods for reasoning around this knowledge

~ 900 person-years of effort

Cycorp headquartered in Austin, Texas; offices in Ljublijana, Slovenia

Timeline 1984: Cyc Project founded by Dr. Douglas Lenat as a lead project in the Microelectronics and Computer Technology Corporation (MCC).

1994: Cycorp was founded to further develop, commercialize, and apply the Cyc technology.

1996: Cycorp gets its first substantial government contract

2002: OpenCyc Version 1 Released: only included 6,000 concepts and 60,000 facts

2006: ResearchCyc Version 1 Released: adds additional knowledge base, English parsing and generation tools, and interface for knowledge editing and querying

2012: OpenCyc Version 4 Released: Includes ~239,000 terms; ~2,093,000 triples

OpenCyc details Available under Apache version 2 license (free software license)

OpenCyc includes a subset of functionality from Cyc

One way for accessing OpenCyc KB (knowledge base) is through OpenCyc KB Browser.

OpenCyc criticized for restrictiveness:

1. Missing instance-level knowledge

2. Lacking any reasoning capability

3. Being formally inconsistent

The 3rd is most difficult to deal with since it has been hand-crafted

Other problems addressed by ResearchCyc or EnterpriseCyc

CycL – The basics Language that makes up most of Cyc

Some similarities to LISP

#$Collection = Class, something that has instances #$Individual

◦ non-set/class (incl. relations, strings, numbers) #$isa = instance of #$genls = subclass of (“is-a”) Operators are built-in code-supported #$Predicates:

◦ #$and #$or #$not #$implies #$arity #$thereExists ◦ #$assertedSentence #$knownSentence #$arg1Isa ◦ #$interArg[Isa/Genl/Reln/Format]N-M #$resultIsa

CycL – The basics Definitional Assertion of isa: (isa isa BinaryPredicate) (arg1Isa isa Thing) (arg2Isa isa Collection)

Example

(isa Benjamin-Susman Individual)

CycL Visualized

Cyc Ontology Separated into 3 parts: upper, middle, and lower

Upper level◦ Contains the most broad abstract concepts, universal truths◦ Smallest but most widely referenced area of Cyc ontology

Middle level◦ Not universal, but widely used abstraction layer◦ Examples: geospatial relationships, broad knowledge of human interactions, or everyday items and

events

Lower level◦ Domain specific knowledge◦ Example: information about chemistry, biology, etc.

Cyc Ontology – Upper Level (comment Event “An important specialization of Situation, and thus also of IntangibleIndividual and TemporallyExistingThing. Each instance of Event is a dynamic situation in which the state of the world changes; each instance is something one would say ‘happens’. Events are intangible because they are changes per se, not tangible objects that effect and undergo changes. Notable specializations of Event include Event-Localized, PhysicalEvent, Action, and GeneralizedTransfer. Events should not be confused with TimeIntervals.”)

(isa Event TemporalStuffType)

(isa Event Collection)

(quotedIsa Event PublicConstant-CommentOK)

(quotedIsa Event VocabularyConstrainingAbstraction)

(genls Event Situation)

(disjointWith Event PositiveDimensionalThing)

Cyc Ontology – Upper Level (cont.)

(genls InstantaneousEvent Event)

(genls HelicopterLanding Event)

inferred knowledge

(genls (BecomingFn Intoxicated) Event)

(relationExistsAll victim Event Victim-UnfortunatePerson)

For every instance of the collection Victim-UnfortunatePerson, there exists an Event in which that

person was the victim—i.e., an event for which these statements hold:

(victim ?SOMEVICTIM ?SOMEEVENT)

(isa ?SOMEVICTIM Victim-UnfortunatePerson)

(isa ?SOMEVICTIM Event))

Cyc Ontology – Middle Level (comment SocialGathering “A specialization of SocialOccurrence. Each instance of SocialGathering is an intentional social gathering of people who have the same or similar purposes in attending, and in which there is communication between the participants. Specializations include BabyShower, Carnival, and Rally. Note that a group of people waiting to board an elevator is not typically a SocialGathering, even though they share a common purpose, since they are not expected to talk to each other.”)

(disjointWith SocialGathering SingleDoerAction)

(disjointWith SocialGathering ConflictEvent)

(disjointWith SocialGathering IntrinsicStateChangeEvent)

(keStrongSuggestionPreds SocialGathering dateOfEvent)

Although it is not semantically required, it is likely that getting a dateOfEvent assertion for any given instance of SocialGathering would be appropriate or desirable.

(requiredActorSlots SocialGathering attendees)

In every social occasion something must play the role of attendees.

Cyc Ontology – Lower Level (comment ChemicalReaction “A collection of events; a subcollection of PhysicalTransformationEvent. Each instance of ChemicalReaction is an event in which two or more substances undergo a chemical change, i.e., some portions of the substances involved are transformed into different ChemicalSubstanceTypes. The transformations are brought about by purely chemical (including biochemical) means which affect chemical bonds between atoms in the molecules of stuff. Examples of ChemicalReaction: instances of CombustionProcess; instances of Photosynthesis-Generic.”)

(keGenlsStrongSuggestionPreds-RelationAllExists ChemicalReaction catalyst)

(genls ChemicalReaction PhysicalTransformationEvent)

(genls CombustionReaction ChemicalReaction)

(genls ExothermicReaction ChemicalReaction)

(genls ChemicalBonding ChemicalReaction)

Cyc Ontology – Lower Level (cont.)

(arg1Genl availableReactantTypeInReactionType ChemicalReaction)

The first argument to an assertion made using the predicate availableReactantTypeInReactionType must generalize to some chemical reaction.

(comment CombustionReaction “A specialization of both ChemicalReaction and CombustionProcess. Each instance of CombustionReaction is a rapid chemical reaction that produces a flame. Many combustion reactions involve oxygen from the air as a reactant.”)

(genls CombustionReaction CombustionProcess)

(genls CombustionReaction ChemicalReaction)

(outputsCreated-TypeType CombustionReaction Flame)

Events which are CombustionReactions have members of the collection Flame as outputs.

CycL Example (isa BurningOfPapalBull SocialGathering)

Because this is an instance of SocialGathering, it is an known to be an instance of Event and to have attendees.

(eventOccursAt BurningOfPapalBull CityOfWittenburgGermany)

(dateOfEvent BurningOfPapalBull

(DayFn 10 (MonthFn December (YearFn 1520))))

(attendee BurningOfPapalBull MartinLuther-ReligiousFigure)

Martin Luther is already represented in the KB, along with basic biographical information such as birth and death date, country of residence, and native language.

(relationInstanceExistsMin BurningOfPapalBull attendees UniversityStudent 40)

At least forty university students attended the event. RelationInstanceExistsMin is a rule macro predicate.

CycL Example (cont.) (isa BurningOfPapalBull-Document CombustionProcess)

(properSubEvent BurningOfPapalBull-Document BurningOfPapalBull)

(relationInstanceExists inputsDestroyed BurningOfPapalBull-Document

(CopyOfConceptualWorkFn PapalBull-ExcommunicationOfLutherCW)

The thing destroyed is a member of the functionally defined collection “all copies of the conceptual work PapalBullExcommunicationOfLuther”. The distinction between the conceptual artifact and the specific copy being burned prevents Cyc from concluding that the conceptual work has been utterly destroyed (in the same way that burning a copy of Moby Dick does not destroy the work Moby Dick generally).

(thereExists ?EVT (and (performedBy ?EVT MartinLuther-ReligiousFigure) (causes ?EVT BurningOfPapalBull-Document)))

The actual burning of a document was caused by some additional sub-event EVT (such as holding a match to it or throwing it into a fire). Unlike the distinction between the social gathering and the actual burning of the document, both of which are reified events, this event is defined implicitly, because there is nothing else to say about it.

NLP in Cyc – The lexicon◦ Backbone of NL system◦ Links between English words and Cyc constants are stored◦ Every word contained as a Cyc constant◦ Constant designated with “#$”

NLP in Cyc – Syntactic Parser

PARSE TREE 1 {:SENTENCE

{:NP

{:DETP {#$Determiner [the]}}

{:N-BAR {#$SimpleNoun [man]}}}

{:VP

{#$Verb [saw]}

{:NP {:DETP {#$Determiner [the]}}

{:N-BAR {#$SimpleNoun [light]}}

{:PP {#$Preposition [with]}

{:NP {:DETP {#$Determiner [the]}}

{:N-BAR {#$SimpleNoun [telescope]}}}}}}}}

PARSE TREE 2 {:SENTENCE

{:NP

{:DETP {#$Determiner [the]}}

{:N-BAR {#$SimpleNoun [man]}}}

{:VP

{#$Verb [saw]}

{:NP {:DETP {#$Determiner [the]}}

{:N-BAR

{:N-BAR {#$SimpleNoun [light]}}

{:PP {#$Preposition [with]}

{:NP {:DETP {#$Determiner [the]}}

{:N-BAR {#$SimpleNoun [telescope]}}}}}}}}

Makes multiple parse trees for ambiguous (outputs all parse trees allowed by rules)

NLP in Cyc – Semantic Interpreter

Cyc-NL's semantic component transforms syntactic parses into CycL formulas

Semantic structures are built up piece-by-piece and combined into larger structures.

For each syntactic rule, there is a corresponding semantic procedure which applies.

Verb driven:◦ Example: "Mary believes that the blue hat is pretty"◦ (#$believes :SUBJECT :CLAUSE).

NLP in Cyc – NL Generation Cyc's natural language generation capabilities include (according to Cyc):

◦ Multi-lingual support◦ Alternative paraphrasing based on desired verbosity, register, etc.◦ Selective inclusion/exclusion of text based on expectations of the user's information needs◦ Extensible lexifications for individual concepts◦ Extensible generation templates for predicates and sentences

Can create output without requiring user to understand CycL

Setting up OpenCyc System requirements

◦ 3G RAM◦ 64 bit system (No 32 bit support)◦ 1G disk space

Setup◦ Download and unzip folder◦ Available for Windows and Linux based systems

◦ Windows: opencyc-4.0\scripts\run-cyc.bat◦ Linux: opencyc-4.0/scripts/run-cyc.sh

Setting up OpenCyc Access

◦ Use web browser: http://localhost:3602/cgi-bin/cyccgi/cg?cb-start ◦ Start as Guest account (documentation bad)◦ Login as CycAdministrator (Note typo in the prompt when OpenCyc starts)◦ Make new account, allow features should now be available

Demo Let’s run it!

Research Cyc Anyone? http://www.cyc.com/downloads/ResearchCyc_License_Rev2.1.pdf

References Primary source of info: http://www.cyc.com/platform/opencyc

Cynthia Matuszek, John Cabral, Michael Witbrock, John DeOliveira. An Introduction to the Syntax and Content of Cyc: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.68.1357&rep=rep1&type=pdf

CycL Visualization courtesy of: http://www.webstructor.net/worlds/cycl.html

Used for timeline and finding release dates: http://en.wikipedia.org/wiki/Cyc

Gave perspective on Cycorp’s funding, long term goals, and organizational setup: http://www.technologyreview.com/news/403803/cycorp-the-cost-of-common-sense/

Copy of ResearchCyc’s License agreement: http://www.cyc.com/downloads/ResearchCyc_License_Rev2.1.pdf

NLP in Cyc: http://www.cyc.com/cyc/nl