1. On the Proper Treatment of Quantifiers in Probabilistic Logic Semantics Islam Beltagy and Katrin Erk The University of Texas at Austin IWCS 2015

2. Logic-based Semantics First-order logic and theorem proving Deep semantic representation: –Negation, Quantifiers, Conjunction, Disjunction …. 2

3. Probabilistic Logic Semantics Logic + Reasoning with Uncertainty –Confidence rating of Word Sense Disambiguation –Weight of Paraphrase rules –Distributional similarity values [Beltagy et al., 2013] baby  toddler | w 1 eating doll  playing with a toy | w 2 –... 3

4. Probabilistic Logic Semantics Quantifiers and Negations do not work as expected Domain Closure Assumption: finite domain –Problems with quantifiers –“Tweety is a bird and it flies”  “All birds fly” Closed-World Assumption: low prior probabilities –Problems with negations –“All birds fly”  “The sky is not blue” 4

5. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks (MLNs) –Recognizing Textual Entailment (RTE) Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 5

6. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 6

7. 7 Probabilistic Logic Frameworks that combine logical and statistical knowledge [Nilsson, 1986], [Getoor and Taskar, 2007] Use weighted first-order logic rules –Weighted rules are soft rules (compared to hard logical constraints) Provide a mechanism for probabilistic inference: P(Q|E, KB) Bayesian Logic Programs (BLP) [Kersting & De Raedt, 2001] Markov Logic Networks (MLN) [Richardson and Domingos, 2006] Probabilistic Soft Logic (PSL) [Kimmig et al., NIPS 2012]

8. Markov Logic Networks [Richardson and Domingos, 2006]  x. smoke(x)  cancer(x) | 1.5  x,y. friend(x,y)  (smoke(x)  smoke(y)) | 1.1 Two constants: Anna (A) and Bob (B) P(Cancer(Anna) | Friends(Anna,Bob), Smokes(Bob)) Cancer(A) Smokes(A)Friends(A,A) Friends(B,A) Smokes(B) Friends(A,B) Cancer(B) Friends(B,B) 8

9. Markov Logic Networks [Richardson and Domingos, 2006] Probability Mass Function (PMF) Inference: calculate probability of atoms given evidence set –P(Cancer(Anna) | Friends(Anna,Bob), Smokes(Bob)) Weight of formula i No. of true groundings of formula i in x Normalization constant a possible truth assignment 9 the set of all atoms

10. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 10

11. Recognizing Textual Entailment (RTE) RTE requires deep semantic understanding [Dagan et al., 2013] Given two sentences Text ( T) and Hypothesis ( H), finding if T Entails, Contradicts or not related (Neutral) to H 11

12. Recognizing Textual Entailment (RTE) Examples (from the SICK dataset) [Marelli et al., 2014] –Entailment: T: “A man is walking through the woods. H: “A man is walking through a wooded area.” –Contradiction: T: “A man is jumping into an empty pool.” H: “A man is jumping into a full pool.” –Neutral: T: “A young girl is dancing.” H: “A young girl is standing on one leg.” 12

13. Recognizing Textual Entailment (RTE) Translate sentences to logic using Boxer [Bos 2008] T: John is driving a car  x,y,z. john(x)  agent(y, x)  drive(y)  patient(y, z)  car(z) H: John is driving a vehicle  x,y,z. john(x)  agent(y, x)  drive(y)  patient(y, z)  vehicle(z) KB: (collected from difference sources)  x. car(x)  vehicle(x) | w P(H|T, KB) 13

14. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 14

15. Domain Closure Assumption (DCA) There are no objects in the world other than the named constants (Finite Domain) e.g.  x. smoke(x)  cancer(x) | 1.5  x,y. friend(x,y)  (smoke(x)  smoke(y)) | 1.1 Two constants: Anna (A) and Bob (B) 15 Cancer(A) Smokes(A)Friends(A,A) Friends(B,A) Smokes(B) Friends(A,B) Cancer(B) Friends(B,B) Ground Atoms

16. Domain Closure Assumption (DCA) There are no objects in the universe other than the named constants (Finite Domain) –Constants need to be explicitly added –Universal quantifiers do not behave as expected because of finite domain –e.g. “Tweety is a bird and it flies”  “All birds fly” P(H|T,KB)TH  SkolemizationNo problems  ExistenceUniversals in H 16

17. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Skolemization:  in T Existence:  in T Universals in Hypothesis :  in H Closed-World Assumption Evaluation Future work and Conclusion 17

18. Skolemization (  in T ) Explicitly introducing constants T:  x,y. john(x)  agent(y, x)  eat(y) Skolemized T: john(J)  agent(T, J)  eat(T) Embedded existentials –T :  x. bird(x)   y. agent(y, x)  fly(y) –Skolemized T:  x. bird(x)  agent(f(x), x)  fly(f(x)) –Simulate skolem functions –  x. bird(x)   y. skolem f (x,y)  agent(y, x)  fly(y) –skolem f (B1, C1), skolem f (B2, C2) … 18

19. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Skolemization:  in T Existence:  in T Universals in Hypothesis :  in H Closed-World Assumption Evaluation Future work and Conclusion 19

20. Existence (  in T ) T: All birds fly H: Some birds fly Logically, T ⇏ H but pragmatically it does –“All birds fly” presupposes that “there exist birds” Solution: simulate this existential presupposition –From parse tree, Q(restrictor, body) –“All birds fly” becomes: all(bird, fly) –Introduce additional evidence for the restrictor bird(B) 20

21. Existence (  in T ) Negated Existential –T: No bird flies = no (bird, fly)   x,y. bird(x)  agent(y, x)  fly(y)  x. bird(x)    y. agent(y, x)  fly(y) –Additional evidence bird(B) Exception –T: There are no birds   x. bird(x) –No additional evidence because the existence presupposition is explicitly negated 21

22. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Skolemization:  in T Existence:  in T Universals in Hypothesis :  in H Closed-World Assumption Evaluation Future work and Conclusion 22

23. Universals in Hypothesis (  in H ) T: Tweety is a bird, and Tweety flies bird(Tweety)  agent(F, Tweety)  fly (F) H: All birds fly  x. bird(x)   y. agent(y, x)  fly(y) T  H because universal quantifiers work only on the constants of the given finite domain Solution: –As in Existence, add evidence for the restrictor: bird(Woody) –If the new bird can be shown to fly, then there is an explicit universal quantification in T 23

24. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 24

25. Closed-World Assumption (CWA) The assumption that everything (all ground atoms) have very low prior probability CWA fits the RTE task because: –In the world, most things are false –Inference results are less sensitive to the domain size –Enable inference optimization [Beltagy and Mooney, 2014] 25

26. Closed-World Assumption (CWA) Because of CWA, negated H comes true regardless of T H :   x,y. bird(x)  agent(y, x)  fly(y) Solution –Add positive evidence that contradicts the negated parts of H –A set of ground atoms with high prior probability (in contrast with low prior probability on all other ground atoms) –R: bird(B)  agent(F, B)  fly(F) | w=1.5 –P(H| CWA)  1 –P(H|R, CWA)  0 26

27. Closed-World Assumption (CWA) Entailing example: T: No bird flies:   x,y. bird(x)  agent(y, x)  fly(y) H: No penguin flies:   x,y. penguin(x)  agent(y, x)  fly(y) R: penguin(P)  agent(F, P)  fly(F) | w=1.5 KB:  x. penguin(x)  bird(x) P(H|T, R, KB) = 1 T  KB contradicts R, which lets H be true. 27

28. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 28

29. 29 Evaluation Probabilistic Logic Framework: Markov Logic Network –Proposed handling of DCA and CWA applies to other Probabilistic Logic frameworks that make similar assumptions, e.g, PSL (Probabilistic Soft Logic) Evaluation Task: RTE –Proposed handling of DCA and CWA applies to other tasks where the logical formulas have existential and universal quantifiers, e.g, STS (Textual Similarity) and Question Answering

30. 30 Evaluation 1)Synthetic Dataset Template: Q 1 NP 1 V Q 2 NP 2 = Q 1 (NP 1, Q 2 (NP 2,V)) Example –T: No man eats all food –H: Some hungry men eat not all delicious food

31. Evaluation 31 1) Synthetic Dataset Dataset size: 952 Neutral + 72 Entail = 1024

32. 32 Detection of Contradiction Entailment: P(H| T, KB, W t,h ) Contradiction: P(  H| T, KB, W t,  h ) World configuration: Domain size Prior probabilities

33. 33 Evaluation 2) Sentences Involving Compositional Knowledge (SICK) [Marelli et al., SemEval 2014] –10,000 pairs of sentences annotated as Entail, Contradict or Neutral

34. 34 Evaluation 3) FraCas [Cooper et al., 1996]: hand-built entailments pairs –We evaluate of the first section (out of 9 sections) –Unsupported quantifiers (few, most, many, at least) (28/74 pairs)

35. Outline Probabilistic Logic Semantics (overview of previous work) –Markov Logic Networks –Recognizing Textual Entailment Domain Closure Assumption –Definition –Inference problems with Quantifiers Closed-World Assumption Evaluation Future work and Conclusion 35

36. Future Work Generalized Quantifiers: How to extend this work to generalized quantifiers like Few and Most 36

37. 37 Conclusion Domain Closure Assumption, its implication on the probabilistic logic inferences, and how to formulate the RTE problem in a way that we get the expected inferences Closed-World Assumption, why we make that assumption, and what its effect on the negation, and how to formulate the RTE problem to get correct inferences.

38. Thank You 38