Вопросно-ответные системы (ВОС). Определение

Вопросно-ответные системы (ВОС). Определение

Friendly software should listen and speak - .3539 .3539 1 QA- Start 2

3 - - : -

4 5

QA- 1960- . QA- - QA-: (closed-domain) :

(open-domain) Start (1993 .)

6 QA- (NLP)) 7

QA-

() QA- 8 - Start 9 QA- Start. http://start.csail.mit.edu MIT Artificial Intelligence Laboratory 1993 Boris Katz

: ( ) Start: NLP), Start

10 Start. What is a fractal? : Who invented the telegraph? :

: What country is bigger, Russia or USA? : Show me some poems by Alexander P)ushkin 11 Start. 1.

Give me the states that border Colorado. What's the largest city in Florida? Show me a map of Denmark List some large cities in Argentina Which is deeper, the Baltic Sea or the North Sea? Show the capital of the 2nd largest country in Asia 2.

When was Beethoven born? Who composed the opera Semiramide? What movies has Dustin Hoffman been in? 12 Start. 3. What is Jupiter's atmosphere made of? Why is the sky blue? Convert 100 dollars into Euros How is the weather in Boston today?

How far is Neptune from the sun? Show me a metro map of Moscow. 4. What countries speak Spanish? Who was the fifth president of the United States? What languages are spoken in the most populous country in Africa? How many people live on Earth? 13 Start.

doc1 doc2 doc3 P)arser Generator text doc4 14 Start.

3- : (T-) / (S-) () 15 Start.

T- < > / T- T- , , T- (, , , , ..) History, T-. 16 Start. T- Bill surprised Hillary with his answer Parser < with answer>

17 Start. T- Whom did Bill surprise with his answer? Bill surprised whom with his answer? P)arser < with answer> Whom = Hillary < with answer> Bill surprised Hillary with his answer 18

Start. T- Did Bill surprise with his answer? Bill surprised Hillary with his answer? P)arser < with answer> Yes! Yes, Bill surprised Hillary with his answer 19 - vs. The bird ate the young snake

The snake ate the young bird The meaning of life A meaningful life The bank of the river The bank near the river 20 - vs. :

, T-: T- 21 What do frogs eat? ,

T- 6 , 3 : Adult frogs eat mainly insects and other small animals, including earthworms, minnows, and spiders One group of South American frogs feeds mainly on other frogs Frogs eat many other animals, including spiders, flies, and worms

22 What do frogs eat? , , 33 , What eats What eats frog? What eats eat What eats frog: Bowfins eat mainly other fish, frogs, and crayfish Cranes eat a variety of foods, including frogs, fishes, birds, and various small mammals. 23

Start. S- : What eats Bills answer surprised Hillary = What eats Bill surprised Hillary with his answer P)arser P)arser = < with answer>

: S- Surprise < with n3> Where ni Nouns 24 Start. S- S- :

/ 25

Start. S-. S- Sell- Buy < to n3> < from n1> Where ni Nouns S- Kill- Die Where ni Nouns 26 Start. S- S- 2- :

T 27 Start. S-

, , S- ERV < with n3> Where ni Nouns and v emotional- reaction- verbs 28 WordNet

() : () 150 000 , 115 000 , 207 000 29 WordNet.

: Y X, X Y : Y X, Y X : X Y , : Y X, X Y : Y X, Y X 30

WordNet. 31 Start. WordNet WordNet Start

T- T- Canary Bird : What eats Can canary fly? Start What eats Yes 32 Start. Omnibase --

: Federico Fellini is a director of La StradaFederico Fellini is a director of La Strada : La Strada : director : Federico Fellini (data source): Star Wars imdb-movie 33

Start. Omnibase. Who wrote the music for Star Wars? Star Wars Composer John Williams Who invented

dynamite? Dynamite Inventor Alfred Nobel How big is Costa Rica? Costa Rica Area 51,100 sq. km How many people live in Kiribati?

Kiribati P)opulation 94,149 What languages are spoken in Guernsey? Guernsey Languages English, French Show me paintings by Monet

Monet Works [images] 34 Start. Omnibase. Who directed gone with the wind? Start Who directed X?, X = Gone with the wind Omnibase Gone with the wind imdb-movie (get Federico Fellini is a director of La Stradaimdb- movie Federico Fellini is a director of La StradaGone with the Wind Federico Fellini is a director of La StradaDirector)

imdb-movie Victor Fleming Victor Fleming directed Gone with the wind 35 Start. Omnibase : --

: 36 Start.

Wikipedia The World Factbook 2006 Google Yahoo The Internet Movie Database Internet P)ublic Library The P)oetry Archives Biography.com Merriam-Webster Dictionary WorldBook Infoplease.com Metropla.net Weather.com 37

, : RDF (Resource Description Framework), :

XML 38 Start. Natural Language Annotations

- : 39 Start. Natural Language Annotations

: RDF ( ) 40 Start. How many people live in Kiribati? What is the population of the Bahamas? Tell me Guams population.

41 Start. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. A Country in the CIA Factbook

42 Start. What is the country in Africa with the largest area? Tell me what Asian country has the highest population density. What country in Europe has the lowest infant mortality rate? What is the most populated South American

country? 43 Start. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

what country in $region has the largest $attribute ?x a :Country ?x map($attribute) ?val ?x :location $region display(boundto(?x, max(?val))) :population :area ... 44 Start.

Is Canadas coastline longer than Russias coastline? Which country has the larger population, Germany or Japan? Is Nigerias population bigger than that of South Africa? 45 Start. 1. 2. 3. 4. 5. 6. 7.

8. 9. 10. 11. 12. 13. 14. 15. $country-1s $att is larger than $country-2s $att ?x a :Country ?x map($att) ?val-1 ?y a :Country ?y map($att) ?val-2 display(gt(?val-1, ?val-2))) :population

:area ... 46 Start. What is the distance from Japan to South Korea? How far is the United States from Russia? Whats the distance between Germany and England? :

47 Start. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

distance between $country1 and $country2 what is the capital of $country1 := ?capital1 what is the capital of $country1 := ?capital2 what is the distance between ?capital1 and ? capital2 := ?distance display(?distance) 48 Start. Natural Language Annotations :

, Omnibase : 49

Start. WordNet 1 Parser T-

2 Omnibase Omnibase doc2 doc1 passage1 doc3 docN WWW

passage2 passageM 50 Start.

51 52 The dog ate. s np vp det

noun verb the dog ate 53 s s np noun vp

verb sold Salespeople np det noun noun np vp noun verb np

sold the dog biscuits Salespeople np det noun the dog noun biscuits

Salespeople sold the dog biscuits. 54 Sentence The det can will rust modal verb modal verb noun

noun noun verb verb verb 55 n arg max p (ti wi ) t1..tn t 1

n arg max p ( wi ti ) p (ti ti 1 ) t1..tn t 1 t (det, noun, ) w (can, will) p(t | w) , t w p(w | t) , w t p(t1 | t2) , t1 t2 56

90% 97% 98% 57 adj 0.218 0.0016 large 0.004 0.45 small 0.005 det a 0.245 noun

the 0.586 0.475 house 0.001 stock 0.001 58 (transformational tagging) . : X Y, Z.

- . HMM vs. TT ( ) 59 Treebank , . : (s (np (det The) (noun stranger)) (vp (verb ate) (np (det the) (noun doughnut) (pp (prep with) (np (det a) (noun fork))))) 60 Statistical

Parser P)en treebank 61 PCFG (Probabilistic contextfree grammars) sp np vp

vp verb np vp verb np np np det noun np noun np det noun noun np np np (1.0) (0.8) (0.2) (0.5) (0.3) (0.15) (0.05) 62 p ( s, ) p (r (c))

c s r(c) r 63 PCFG. . P)en treebank

P)() = , 64 Two state-of-the-art statistical parsers. Markov grammars , , , np = prep + p (r f ) p (ti f , ti 1 )

ti r p(t1 | f, t2) t2 t1 f. 65 Lexicalized parsing p( s, ) p (h(c) m(c), t (c)) p (r (c) h(c)) c (head), .

p(r | h) , r h. p(h | m, t) , h head = m t. 66 Lexicalized parsing (S (NP) The (ADJP) most troublesome) report) (VP) may (VP) be (NP) (NP) the August merchandise trade deficit) (ADJP) due (ADVP) out) (NP) tomorrow))))) p(h | m, t) = p(be | may, vp) p(r | h) = p(posvp aux np | be)

67 Lexicalized parsing What eats the August merchandise trade deficit rule = np det propernoun noun noun noun Conditioning events p(What eats August) p(rule) Nothing 2.7*10^(-4) 3.8*10^(-5)

P)art of speech 2.8*10^(-3) 9.4*10^(-5) h(c) = What eats deficit 1.9*10^(-1) 6.3*10^(-3) 68 (Causal Reconstruction) 69

, . 70 . , . ?

? 2? 71 CR (Causal Reconstruction) , . 72 (Transition space) ,

The perception of causality 73 (Causal modeling)

74 , CR 75 76

? What eats ? 77 . 3

78 79 2 80

3 . Windows? 81 3 .

? , , ? 82

breakpoint 83 (States vs. Changes) Changes 84 The

contact between the steam and the metal plate appears. The concentration of the solution increases. The appearance of the film changes. The pin becomes a part of the structure. The water remains inside the tank. 85 The contact between the steam and the metal plate appears. The concentration of the solution increases. The appearance of the film changes. The pin becomes a part of the structure. The water remains inside the tank. 86

AP)P)EAR NOT-AP)P)EAR DISAP)P)EAR NOT-DISAP)P)EAR 87 NOT-DISAPPEAR CHANGE

NOT-CHANGE INCREASE DECREASE NOT-INCREASE NOT-DECREASE 88 -

89 AP)P)EAR(contact, , t1, t2) INCREASE(concentration, the-solution, t3, t4) CHANGE(appearance, the-film, t5, t6) AP)P)EAR(a-part-of, , t7, t8) NOT-DISAP)P)EAR(inside, , t9, t10) 90

. ::= ::= the { { | and } }* ::= CHANGE | AP)P)EAR .. The concentration of the solution increases. 91 . ::= ::= [ [] { { | and} }* ]

::= becomes | becomes not | remains | remains not The water becomes a vapor. 92 1. 2. 3. 4. 5. 6. 7. CLEF. http://clef-qa.itc.it/ WordNet. http://wordnet.princeton.edu/ P)en treebank. http://www.cis.upenn.edu/~treebank/

Start. http://start.csail.mit.edu/ TREC. http://trec.nist.gov/ Eugene Charniak [1997], What eats Statistical Techniques for Natural Language P)arsing Gary C. Borchardt [1993], What eats Causal Reconstruction 93 8. 9. 10. 11. 12. Boris Katz, Beth Levin [1988] What eats Exploiting Lexical

Regularities in Designing Natural Language Systems Boris Katz and Jimmy Lin. Annotating the Semantic Web Using Natural Language. September, 2002. Boris Katz, Sue Felshin, Deniz Yuret, Ali Ibrahim, Jimmy Lin, Gregory Marton, Alton Jerome McFarland and Baris Temelkuran. Omnibase: Uniform Access to Heterogeneous Data for Question Answering. June, 2002. SEMLP). http://semlp.com/ RCO. http://www.rco.ru/ 94

Recently Viewed Presentations

  • Delaware&#x27;s Pretrial Modernization Effort

    Delaware's Pretrial Modernization Effort

    Delaware Pretrial Modernization Effort. Discussion Points. In 2014, the Delaware Center for Justice was awarded a three year grant to coordinate stakeholders' efforts to improve the pretrial justice system. The Policy team is chaired by Chief Magistrate Davis and represents...
  • 1 MATLAB  MATLAB  MATrix   LABoratory  20  Cleve Moler
  • SEIS RELACIONES DEL DISCÍPULO - WordPress.com

    SEIS RELACIONES DEL DISCÍPULO - WordPress.com

    Podemos llegar a afirmar que la experiencia de ser un hijo amado de Dios, es el rasgo primordial que configura al discípulo de Jesús. No puede haber discípulo que no viva la relación filial con Dios, su padre. Y, naturalmente,...
  • OligoArchive: Using DNA in the DBMS storage hierarchy

    OligoArchive: Using DNA in the DBMS storage hierarchy

    Introduction. In this paper, we investigate the implications of integrating DNA in the DBMS storage hierarchy. We present OligoArchive, an architecture for using a DNA-based storage system as the archival tier of a relational DBMS.
  • Identifying and Supporting Doctors in Difficulty  Bringing Research

    Identifying and Supporting Doctors in Difficulty Bringing Research

    Nationally through the Renal SAC and JRCPTB through External Advisor Training Working to roll out to Foundation through Horus Added in trainee feedback of quality of ESR in 2016 Started to assess quality of SLEs in 2016 ESR feedback -Trainee...
  • Biliary Endoscopy in the Management of Choledocholithiasis CSGNA

    Biliary Endoscopy in the Management of Choledocholithiasis CSGNA

    Acute biliary pancreatitis should have: Mild = No procedure if no evidence of stone. Cholangitis or obstruction go for ERCP. What about cholangitis? Impacted stones biliary stasis. When ductal pressure >30 cm H2O, flow stops IgA secretion decreases
  • What is IPA TouchPoints? - JICMAIL

    What is IPA TouchPoints? - JICMAIL

    What is IPA TouchPoints? Specified by the requirements of the industry, on a Joint industry basis for IPA Member Agencies and Media Owners. Trusted accountable data with a complete transparency of methodology. Only incorporating the highest quality of industry data...
  • Florida&#x27;s Bathymetry - University of Florida

    Florida's Bathymetry - University of Florida

    Ejection trajectories of grains blasted into the air by rainsplash The Raindrop CTSV Raindrop Size and Effects Rain Splash Experiments Measuring Net Downslope Rainsplash Transport Creep General term for slow, downslope mass movement of material in response to gravity on...