Getting and Utilizing World Information utilizing a Confined Subset of English

0
0
1607 days ago, 634 views
PowerPoint PPT Presentation
Information obtaining is still a noteworthy bottleneck. computerized techniques ... World. Information. Etymological. Learning. Entering Quantified Expressions (Rules) Seven

Presentation Transcript

Slide 1

Procuring and Using World Knowledge utilizing a Restricted Subset of English Peter Clark, Phil Harrison, Tom Jenkins, John Thompson, Rick Wojcik Boeing Phantom Works, Seattle

Slide 2

Introduction Knowledge securing is still a noteworthy bottleneck mechanized strategies are great yet extremely limited Our approach: Knowledge section utilizing Controlled Language Hits "sweet spot" amongst rationale and full NLP dialect translator creates rationale yield Outline: Our Controlled Language Processing innovation Discussion on Natural Language as a reason for KR

Slide 3

"A ball tumbles from a precipice" " x  y B(x)  R(x,y)  C(y)" "Consider the accompanying conceivable circumstance in which a ball first… " too hard for the client too hard for the PC to comprehend The Language Spectrum Unrestricted normal dialect Formal dialect Controlled English

Slide 4

Short sentences No pronouns Rewritten in CPL (PC can comprehend): A protest is tossed from a bluff. The level speed of the question is 20 m/s. The highest point of the precipice is 125 m above level ground. The question falls 125 m to the ground. What is the term of the fall? Basic sentence structures CPL (Computer-Processable Language) Original content (endless to PC): A question is tossed with an even speed of 20 m/s from a precipice that is 125 m above level ground. In the event that air resistance is unimportant, to what extent does it take the protest tumble to the ground?

Slide 5

isa(_Object1, object_n1) isa(_Cliff2, cliff_n1) isa(_Throw3, throw_v1) object(_Throw3, _Object1) origin(_Throw3, _Cliff2) Throw question birthplace Object Cliff Target Interpretation Sentences in first-arrange rationale Capable of supporting machine surmising "A protest is tossed from a precipice"

Slide 6

isa(_Person1, person_n1) isa(_Room2, room_n1) isa(_Entity3, entity_n1) isa(_Carry4, carry_v1) object(_Carry4, _Entity3) agent(_Carry4, _Person1) is-inside(_Entity4, _Room2) =====> is-inside(_Person1, _Room2) Carry operator question Person Object is-inside will be inside Room Target Interpretation Sentences in first-arrange rationale Capable of supporting machine deduction IF "a man is conveying a substance that is inside a room" THEN "the individual is in the room."

Slide 7

Throw question starting point Object Cliff Overview of Processing "A question is tossed from a bluff" Parser & LF Generator Word sense disambiguator Linguistic Knowledge Relational disambiguator Coreference identifier World Knowledge Structural reorganizer (_Object13320 instance_of object_n1) (_Cliff13321 instance_of cliff_n1) (_Throw13319 instance_of throw_v1) (_Throw13319 question _Object13320) (_Throw13319 root _Cliff13321)

Slide 8

Entering Quantified Expressions (Rules) Seven "manage layouts" utilized: IF sentence THEN sentence ABOUT protest : sentence protest IS thing/verb state BEFORE sentence , sentence BEFORE sentence , it is not genuine that sentence AFTER sentence , sentence AFTER sentence , it is not genuine that sentence Processing: Each sentence prepared as a ground declaration Quantifiers are included (Prolog-style) "Activity" formats get to be circumstance analytics rules

Slide 9

CPL (Controlled english) A protest is tossed from a precipice. The level speed of the question is 20 m/s. The highest point of the bluff is 125 m above level ground. Revamping exhortation Logic A protest is tossed from a bluff. The even speed of the protest is 20 m/s. The highest point of the bluff is 125 m above level ground. Summarize of framework's understanding KB Overall Flow of Processing Original content

Slide 17

Part II: Discussion Controlled Languages: Strengths and difficulties

Slide 18

Strengths…  x  y B(x)  R(x,y)  C(y)??? "A man is driving a truck towards the industrial facility" CPL is anything but difficult to utilize , seems feasible fabricated KB with more than 1000 tenets KB is derivation skilled simple to assess and sort out Makes information passage open to numerous clients real accomplishment

Slide 19

Original content: "assault: serious unfriendly feedback" CPL: "IF a man assaults a 2 nd individual THEN the main individual censures the 2 nd individual strongly." Challenges: 1. Reformulating in a Controlled Language is not trifling Task is not simply syntactic reformulation Rather: "common" English leaves much learning certain CPL creator must make that express

Slide 20

Original content: "pivot: the middle around which something turns" CPL: "IF a question is turning THEN the protest is pivoting the protest's hub." Challenges: 1. Reformulating in a Controlled Language is not trifling Task is not simply syntactic reformulation Rather: "normal" English leaves much learning certain CPL creator must make that express

Slide 21

"The man ate the sandwich on the plate" "The man ate on the plate. He ate the sandwich." ?????? 2. Clients may not know about framework's slip-ups User must have the capacity to spot misinterpretations effectively System's summary must be unambiguous User must know how to right them

Slide 22

"The man ate the sandwich on the plate" "The man ate on the plate. He ate the sandwich." "the man ate the sandwich that was on the plate" 2. Clients may not know about their missteps User must have the capacity to spot mistakes effectively System's summarization must be unambiguous User must know how to right them

Slide 23

3. Regular Language-based information representations have restricted expressivity "Common dialect is extremely expressive" … not to the PC! (Stay away from "pie in the sky semantics") Expressiveness = the sum the PC comprehends the sum it can use to reach determinations from Everything else is useless to the PC e.g., CPL can't express: requirements, defaults, some measurement designs

Slide 24

NL-based KR "Customary" KR distance(_Walk1, _Mile1) count(_Mile1, 10) distance(_Walk1, _Distance1) value(_Distance1, 10, mile)   4. In some cases, etymologically spurred representations are poor Language-based KR: Most ideas relate to words Structure of KB will reflect structure of dialect Is this awful? Here and there… "… strolled for 10 miles"

Slide 25

5. (Absence of) Canonicalization "leading a trial of an element" "testing a substance" Many approaches to state a similar thing System needs to understand the equality BUT: regularly NL-based KRs won't  Solutions: Add identicalness rules . (Yet, there are parts!!) e.g., "Directing a X of Y ↔ Xing a Y" Have the translator standardize the info . Confine the info dialect.

Slide 26

Summary CPL = a confined English dialect for information Hits "sweet spot" amongst rationale and full NLP Produces surmising fit representations Is reasonable , used to construct an expansive KB But: No "free lunch" obliges expertise to utilize it viably NL-based KRs are turning out to be more essential! Web: require semantically significant explanations AI: require better information securing instruments Some energizing potential outcomes ahead (esp. at Boeing!)

SPONSORS