1. February 2012 Princeton Plasma Physics Laboratory With thanks to: Collaborators: Ming-Wei Chang, James Clarke, Michael Connor, Dan Goldwasser, Lev Ratinov, Vivek Srikumar, Many others Funding: NSF; DHS; NIH; DARPA. DASH Optimization (Xpress-MP) Learning and Inference for Natural Language Understanding Dan Roth Department of Computer Science University of Illinois at Urbana-Champaign

2. Nice to Meet You Page 2

3. Page 3 Learning and Inference in Natural Language  Global decisions in which several local decisions play a role but there are mutual dependencies on their outcome.  Structured Output Problems – multiple dependent output variables  (Learned) models/classifiers for different sub-problems  In some cases, not all local models can be learned simultaneously  In these cases, constraints may appear only at evaluation time  Incorporate models ’ information, along with prior knowledge (constraints), in making coherent decisions  decisions that respect the local models as well as domain & context specific knowledge/constraints.

4. Page 4 Comprehension 1. Christopher Robin was born in England. 2. Winnie the Pooh is a title of a book. 3. Christopher Robin’s dad was a magician. 4. Christopher Robin must be at least 65 now. A process that maintains and updates a collection of propositions about the state of affairs. This is an Inference Problem (ENGLAND, June, 1989) - Christopher Robin is alive and well. He lives in England. He is the same person that you read about in the book, Winnie the Pooh. As a boy, Chris lived in a pretty home called Cotchfield Farm. When Chris was three years old, his father wrote a poem about him. The poem was printed in a magazine for others to read. Mr. Robin then wrote a book. He made up a fairy tale land where Chris lived. His friends were animals. There was a bear called Winnie the Pooh. There was also an owl and a young pig, called a piglet. All the animals were stuffed toys that Chris owned. Mr. Robin made them come to life with his words. The places in the story were all near Cotchfield Farm. Winnie the Pooh was written in 1925. Children still love to read about Christopher Robin and his animal friends. Most people don't know he is a real person who is grown now. He has written two books of his own. They tell what it is like to be famous.

5. Why is it difficult? Meaning Language Ambiguity Variability Page 5

6. Context Sensitive Paraphrasing He used a Phillips head to tighten the screw. The bank owner tightened security after a spat of local crimes. The Federal Reserve will aggressively tighten monetary policy. Loosen Strengthen Step up Toughen Improve Fasten Impose Intensify Ease Beef up Simplify Curb Reduce Loosen Strengthen Step up Toughen Improve Fasten Impose Intensify Ease Beef up Simplify Curb Reduce Page 6

7. Variability in Natural Language Expressions Example: Relation Extraction: “Works_for” Jim Carpenter works for the U.S. Government. The American government employed Jim Carpenter. Jim Carpenter was fired by the US Government. Jim Carpenter worked in a number of important positions. …. As a press liaison for the IRS, he made contacts in the white house. Top Russian interior minister Yevgeny Topolov met yesterday with his US counterpart, Jim Carpenter. Former US Secretary of Defence Jim Carpenter spoke today… Page 7

8. Textual Entailment Eyeing the huge market potential, currently led by Google, Yahoo took over search company Overture Services Inc. last year Yahoo acquired Overture Is it true that…? (Textual Entailment)  Overture is a search company Google is a search company ………. Google owns Overture Page 8 A key problem in natural language understanding is to abstract over the inherent syntactic and semantic variability in natural language.

9. Page 9 Why? A law office wants to get the list of all people that were mentioned in email correspondence with the office.  For each name, determine whether is was mentioned adversarially or not. A political scientist studies Climate Change and its effect on Societal instability. He wants to identify all events related to demonstrations, protests, parades, elections, analyze them (who, when, where, why) and generate a timeline An electronic health record (EHR) is a personal health record in digital format. Includes information relating to:  Current and historical health, medical conditions and medical tests; medical referrals, treatments, medications, demographic information etc.  Today: a write only document  Can we use it in medical advice systems; medication selection and tracking (Vioxx…); disease outbreak and control; science – correlating response to drugs with other conditions

10. Page 10 Background It’s difficult to program predicates of interest due to  Ambiguity (everything has multiple meanings)  Variability (everything you want to say you can say in many ways) Consequently: all of Natural Language Processing is driven by Statistical Machine Learning Even simple predicates like:  What is the part of speech of the word “can”, or  Correct: I’d like a peace of cake Not to mention harder problems like co-reference resolution, parsing, semantic parsing, named entity recognition,… Machine Learning problems in NLP are large:  Often, 10 6 features (due to lexical items, conjunctions of them, etc.)  We are pretty good for some classes of problems.

11. 11 Illinois’ bored of education [board] Nissan Car and truck plant; plant and animal kingdom (This Art) (can N) (will MD) (rust V) V,N,N The dog bit the kid. He was taken to a veterinarian; a hospital Tiger was in Washington for the PGA Tour  Finance; Banking; World News; Sports Important or not important; love or hate Classification: Ambiguity Resolution

12. 12  Theoretically: generalization bounds  How many example does one need to see in order to guarantee good behavior on previously unobserved examples.  Algorithmically: good learning algorithms for linear representations.  Can deal with very high dimensionality (10 6 features)  Very efficient in terms of computation and # of examples. On-line.  Key issues remaining:  Learning protocols: how to minimize interaction (supervision); how to map domain/task information to supervision; semi-supervised learning; active learning; ranking; adaptation to new domains.  What are the features? No good theoretical understanding here.  Is it sufficient for making progress in NLP? Classification is Well Understood Classification: learn a function f: X  Y that maps observations in a domain to one of several categories.

13. Page 13 Significant Progress in NLP and Information Extraction

14. Page 14 Comprehension 1. Christopher Robin was born in England. 2. Winnie the Pooh is a title of a book. 3. Christopher Robin’s dad was a magician. 4. Christopher Robin must be at least 65 now. A process that maintains and updates a collection of propositions about the state of affairs. This is an Inference Problem (ENGLAND, June, 1989) - Christopher Robin is alive and well. He lives in England. He is the same person that you read about in the book, Winnie the Pooh. As a boy, Chris lived in a pretty home called Cotchfield Farm. When Chris was three years old, his father wrote a poem about him. The poem was printed in a magazine for others to read. Mr. Robin then wrote a book. He made up a fairy tale land where Chris lived. His friends were animals. There was a bear called Winnie the Pooh. There was also an owl and a young pig, called a piglet. All the animals were stuffed toys that Chris owned. Mr. Robin made them come to life with his words. The places in the story were all near Cotchfield Farm. Winnie the Pooh was written in 1925. Children still love to read about Christopher Robin and his animal friends. Most people don't know he is a real person who is grown now. He has written two books of his own. They tell what it is like to be famous.

15. Page 15 Coherency in Semantic Role Labeling Predicate-arguments generated should be consistent across phenomena The touchdown scored by Cutler cemented the victory of the Bears. VerbNominalizationPreposition Predicate: score A0: Cutler (scorer) A1: The touchdown (points scored) Predicate: win A0: the Bears (winner) Sense: 11(6) “the object of the preposition is the object of the underlying verb of the nominalization” Linguistic Constraints: A0: the Bears  Sense(of): 11(6) A0: Cutler  Sense(by): 1(1)

16. Page 16 Semantic Parsing Successful interpretation involves multiple decisions  What entities appear in the interpretation?  “New York” refers to a state or a city?  How to compose fragments together? state(next_to()) >< next_to(state()) X :“What is the largest state that borders New York and Maryland ?" Y: largest( state( next_to( state(NY) AND next_to (state(MD))))

17. Page 17 Learning and Inference Natural Language Decisions are Structured  Global decisions in which several local decisions play a role but there are mutual dependencies on their outcome. It is essential to make coherent decisions in a way that takes the interdependencies into account. Joint, Global Inference. Today:  How to support Structured Predictions in NLP Using declarative constraints in Natural Language Processing  Learning and Inference issues  Mostly Examples

18. Outline Background:  Natural Language Processing: problems and difficulties Global Inference with expressive structural constraints in NLP  Constrained Conditional Models Some Learning Issues in the presence of minimal supervision  Constraints Driven Learning  Learning with Indirect Supervision  Response based Learning More Examples Page 18

19. Page 19 Statistics or Linguistics? Statistical approaches were very successful in NLP But, it has become clear that there is a need to move from strictly Data Driven approaches to Knowledge Driven approaches Knowledge: Linguistics, Background world knowledge How to incorporate Knowledge into Statistical Learning & Decision Making? In many respects Structured Prediction addresses this question.  This also distinguishes it from the “standard” study of probabilistic models.

20. Page 20 Inference with General Constraint Structure Recognizing Entities and Relations Dole ’s wife, Elizabeth, is a native of N.C. E 1 E 2 E 3 R 12 R 23 other 0.05 per 0.85 loc 0.10 other 0.05 per 0.50 loc 0.45 other 0.10 per 0.60 loc 0.30 irrelevant 0.10 spouse_of 0.05 born_in 0.85 irrelevant 0.05 spouse_of 0.45 born_in 0.50 irrelevant 0.05 spouse_of 0.45 born_in 0.50 other 0.05 per 0.85 loc 0.10 other 0.10 per 0.60 loc 0.30 other 0.05 per 0.50 loc 0.45 irrelevant 0.05 spouse_of 0.45 born_in 0.50 irrelevant 0.10 spouse_of 0.05 born_in 0.85 other 0.05 per 0.50 loc 0.45 How to guide the global inference? Why not learn jointly?

21. Page 21 Pipeline Conceptually, Pipelining is a crude approximation  Interactions occur across levels and down stream decisions often interact with previous decisions.  Leads to propagation of errors  Occasionally, later stage problems are easier but cannot correct earlier errors. But, there are good reasons to use pipelines  Putting everything in one basket may not be right  How about choosing some stages and think about them jointly? POS TaggingPhrasesSemantic EntitiesRelations Most problems are not single classification problems ParsingWSDSemantic Role Labeling Raw Data

22. Page 22 Semantic Role Labeling I left my pearls to my daughter in my will. [ I ] A0 left [ my pearls ] A1 [ to my daughter ] A2 [ in my will ] AM-LOC. A0Leaver A1Things left A2Benefactor AM-LOCLocation I left my pearls to my daughter in my will. Overlapping arguments If A2 is present, A1 must also be present. Who did what to whom, when, where, why,… How to express the constraints on the decisions? How to “enforce” them?

23. Page 23 PropBank [Palmer et. al. 05] provides a large human-annotated corpus of semantic verb-argument relations.  It adds a layer of generic semantic labels to Penn Tree Bank II.  (Almost) all the labels are on the constituents of the parse trees. Core arguments: A0-A5 and AA  different semantics for each verb  specified in the PropBank Frame files 13 types of adjuncts labeled as AM- arg  where arg specifies the adjunct type Semantic Role Labeling (2/2)

24. Page 24 Algorithmic Approach Identify argument candidates  Pruning [Xue&Palmer, EMNLP’04]  Argument Identifier Binary classification (A-Perc) Classify argument candidates  Argument Classifier Multi-class classification (A-Perc) Inference  Use the estimated probability distribution given by the argument classifier  Use structural and linguistic constraints  Infer the optimal global output I left my nice pearls to her [ [ [ [ [ ] ] ] ] ] I left my nice pearls to her [ [ [ [ [ ] ] ] ] ] I left my nice pearls to her candidate arguments

25. Page 25 Semantic Role Labeling (SRL) I left my pearls to my daughter in my will. 0.5 0.15 0.1 0.15 0.6 0.05 0.1 0.2 0.6 0.05 0.7 0.05 0.15 0.3 0.2 0.1 0.2

26. Page 26 Semantic Role Labeling (SRL) I left my pearls to my daughter in my will. 0.5 0.15 0.1 0.15 0.6 0.05 0.1 0.2 0.6 0.05 0.7 0.05 0.15 0.3 0.2 0.1 0.2

27. Page 27 Semantic Role Labeling (SRL) I left my pearls to my daughter in my will. 0.5 0.15 0.1 0.15 0.6 0.05 0.1 0.2 0.6 0.05 0.7 0.05 0.15 0.3 0.2 0.1 0.2 One inference problem for each verb predicate.

28. Page 28 Integer Linear Programming Inference For each argument a i  Set up a Boolean variable: a i,t indicating whether a i is classified as t Goal is to maximize   i score(a i = t ) a i,t  Subject to the (linear) constraints If score(a i = t ) = P(a i = t ), the objective is to find the assignment that maximizes the expected number of arguments that are correct and satisfies the constraints. The Constrained Conditional Model is completely decomposed during training

29. Page 29 No duplicate argument classes  a  P OT A RG x { a = A0 }  1 R-ARG  a2  P OT A RG,  a  P OT A RG x { a = A0 }  x { a2 = R-A0 } C-ARG  a2  P OT A RG,  (a  P OT A RG )  (a is before a2 ) x { a = A0 }  x { a2 = C-A0 } Many other possible constraints: Unique labels No overlapping or embedding Relations between number of arguments; order constraints If verb is of type A, no argument of type B Any Boolean rule can be encoded as a (collection of) linear constraints. If there is an R-ARG phrase, there is an ARG Phrase If there is an C-ARG phrase, there is an ARG before it Constraints Joint inference can be used also to combine different (SRL) Systems. Universally quantified rules LBJ: allows a developer to encode constraints in FOL to be compiled into linear inequalities automatically.

30. Page 30 SRL: Formulation & Outcomes 2:30 Demo:http://cogcomp.cs.illinois.edu/page/demoshttp://cogcomp.cs.illinois.edu/page/demos Top ranked system in CoNLL’05 shared task Key difference is the Inference 2) Produces a very good semantic parser. F1~90% 3) Easy and fast: ~7 Sent/Sec (using Xpress-MP)

31. Constrained Conditional Models (aka ILP Inference) How to solve? This is an Integer Linear Program Solving using ILP packages gives an exact solution. Cutting Planes, Dual Decomposition & other search techniques are possible (Soft) constraints component Weight Vector for “local” models Penalty for violating the constraint. How far y is from a “legal” assignment Features, classifiers; log- linear models (HMM, CRF) or a combination How to train? Training is learning the objective function Decouple? Decompose? How to exploit the structure to minimize supervision?

32. Three Ideas Idea 1: Separate modeling and problem formulation from algorithms  Similar to the philosophy of probabilistic modeling Idea 2: Keep model simple, make expressive decisions (via constraints)  Unlike probabilistic modeling, where models become more expressive Idea 3: Expressive structured decisions can be supervised indirectly via related simple binary decisions  Global Inference can be used to amplify the minimal supervision. Modeling Inference Learning

33. Linguistics Constraints Cannot have both A states and B states in an output sequence. Linguistics Constraints If a modifier chosen, include its head If verb is chosen, include its arguments Examples: CCM Formulations (aka ILP for NLP) CCMs can be viewed as a general interface to easily combine declarative domain knowledge with data driven statistical models Sequential Prediction HMM/CRF based: Argmax  ¸ ij x ij Sentence Compression/Summarization: Language Model based: Argmax  ¸ ijk x ijk Formulate NLP Problems as ILP problems (inference may be done otherwise) 1. Sequence tagging (HMM/CRF + Global constraints) 2. Sentence Compression (Language Model + Global Constraints) 3. SRL (Independent classifiers + Global Constraints)

34. Outline Background:  Natural Language Processing: problems and difficulties Global Inference with expressive structural constraints in NLP  Constrained Conditional Models Some Learning Issues in the presence of minimal supervision  Constraints Driven Learning  Learning with Indirect Supervision  Response based Learning More Examples Page 34 Learning structured models requires annotating structures Interdependencies among decision variables should be exploited in Inference & Learning. Goal: learn from minimal, indirect supervision. Amplify it using variables’ interdependencies

35. Information extraction without Prior Knowledge Prediction result of a trained HMM Lars Ole Andersen. Program analysis and specialization for the C Programming language. PhD thesis. DIKU, University of Copenhagen, May 1994. [AUTHOR] [TITLE] [EDITOR] [BOOKTITLE] [TECH-REPORT] [INSTITUTION] [DATE] Violates lots of natural constraints! Lars Ole Andersen. Program analysis and specialization for the C Programming language. PhD thesis. DIKU, University of Copenhagen, May 1994.

36. Page 36 Strategies for Improving the Results (Pure) Machine Learning Approaches  Higher Order HMM/CRF?  Increasing the window size?  Adding a lot of new features Requires a lot of labeled examples  What if we only have a few labeled examples? Other options?  Constrain the output to make sense  Push the (simple) model in a direction that makes sense Increasing the model complexity Can we keep the learned model simple and still make expressive decisions? Increase difficulty of Learning

37. Page 37 Examples of Constraints Each field must be a consecutive list of words and can appear at most once in a citation. State transitions must occur on punctuation marks. The citation can only start with AUTHOR or EDITOR. The words pp., pages correspond to PAGE. Four digits starting with 20xx and 19xx are DATE. Quotations can appear only in TITLE ……. Easy to express pieces of “knowledge” Non Propositional; May use Quantifiers

38. Page 38 Information Extraction with Constraints Adding constraints, we get correct results!  Without changing the model [AUTHOR] Lars Ole Andersen. [TITLE] Program analysis and specialization for the C Programming language. [TECH-REPORT] PhD thesis. [INSTITUTION] DIKU, University of Copenhagen, [DATE] May, 1994. Constrained Conditional Models Allow: Learning a simple model Make decisions with a more complex model Accomplished by directly incorporating constraints to bias/re- rank decisions made by the simpler model

39. Page 39 Guiding (Semi-Supervised) Learning with Constraints Model Decision Time Constraints Un-labeled Data Constraints In traditional Semi-Supervised learning the model can drift away from the correct one. Constraints can be used to generate better training data  At training to improve labeling of un-labeled data (and thus improve the model)  At decision time, to bias the objective function towards favoring constraint satisfaction.

40. Can be viewed as Constrained Expectation Maximization (EM) algorithm. This is the hard EM version; can be generalized in several directions Page 40 Constraints Driven Learning (CoDL) (w 0, ½ 0 )= learn(L)‏ For N iterations do T=  For each x in unlabeled dataset h à argmax y w T Á (x,y) -  ½ k d C (x,y) T=T  {(x, h)} (w, ½ ) =  (w 0, ½ 0 ) + (1-  ) learn(T) [Chang, Ratinov, Roth, ACL’07;ICML’08,ML, to appear] Related to Ganchev et. al [PR work 09,10] Supervised learning algorithm parameterized by (w, ½ ). Learning can be justified as an optimization procedure for an objective function Inference with constraints: augment the training set Learn from new training data Weigh supervised & unsupervised models. Excellent Experimental Results showing the advantages of using constraints, especially with small amounts on labeled data [Chang et. al, Others] Several Training Paradigms

41. Page 41 Objective function: Constraints Driven Learning (CODL) # of available labeled examples Learning w 10 Constraints Poor model + constraints Constraints are used to:  Bootstrap a semi-supervised learner  Correct weak models predictions on unlabeled data, which in turn are used to keep training the model. Learning w/o Constraints: 300 examples. Semi-Supervised Learning Paradigm that makes use of constraints to bootstrap from a small number of examples [Chang, Ratinov, Roth, ACL’07;ICML’08,ML, to appear] Related to Ganchev et. al [PR work 09,10]

42. Constrained Conditional Models – ILP formulations – have been shown useful in the context of many NLP problems, [Roth&Yih, 04,07; Chang et. al. 07,08,…]  SRL, Summarization; Co-reference; Information & Relation Extraction; Event Identifications; Transliteration; Textual Entailment; Knowledge Acquisition Some theoretical work on training paradigms [Punyakanok et. al., 05 more] See a NAACL’10 tutorial on my web page & an NAACL’09 ILPNLP workshop Summary of work & a bibliography: http://L2R.cs.uiuc.edu/tutorials.htmlhttp://L2R.cs.uiuc.edu/tutorials.html Constrained Conditional Models [AAAI’08, MLJ’12]

43. Outline Background:  Natural Language Processing: problems and difficulties Global Inference with expressive structural constraints in NLP  Constrained Conditional Models Some Learning Issues in the presence of minimal supervision  Constraints Driven Learning  Learning with Indirect Supervision  Response based Learning More Examples Page 43 Learning structured models requires annotating structures Interdependencies among decision variables should be exploited in Inference & Learning. Goal: learn from minimal, indirect supervision. Amplify it using variables’ interdependencies

44. Page 44 Connecting Language to the World [CoNLL’10,ACL’11,IJCAI’11] Can I get a coffee with no sugar and just a bit of milk Can we rely on this interaction to provide supervision? MAKE(COFFEE,SUGAR=NO,MILK=LITTLE) Arggg Great! Semantic Parser This requires that we use the minimal binary supervision (“good structure” /”bad structure”) as a way to learn how to generate good structures. SKIP

45. Page 45 Key Ideas in Learning Structures Idea1: Simple, easy to supervise, binary decisions often depend on the structure you care about. Learning to do well on the binary task can drive the structure learning. Idea2: Global Inference can be used to amplify the minimal supervision.  Idea 2 ½: There are several settings where a binary label can be used to replace a structured label. Perhaps the most intriguing is where you use the world response to the model’s actions.

46. Page 46 I. Paraphrase Identification Consider the following sentences: S1: Druce will face murder charges, Conte said. S2: Conte said Druce will be charged with murder. Are S1 and S2 a paraphrase of each other? There is a need for an intermediate representation to justify this decision Given an input x 2 X Learn a model f : X ! {-1, 1} We need latent variables that explain why this is a positive example. Given an input x 2 X Learn a model f : X ! H ! {-1, 1} XY H

47. Page 47 Algorithms: Two Conceptual Approaches Two stage approach (a pipeline; typically used for TE, paraphrase id, others)  Learn hidden variables; fix it Need supervision for the hidden layer (or heuristics)  For each example, extract features over x and (the fixed) h.  Learn a binary classier for the target task Proposed Approach: Joint Learning  Drive the learning of h from the binary labels  Find the best h(x)  An intermediate structure representation is good to the extent is supports better final prediction.  Algorithm? How to drive learning a good H? XY H

48. Page 48 Algorithmic Intuition If x is positive  There must exist a good explanation (intermediate representation)  9 h, w T Á (x,h) ¸ 0  or, max h w T Á (x,h) ¸ 0 If x is negative  No explanation is good enough to support the answer  8 h, w T Á (x,h) · 0  or, max h w T Á (x,h) · 0 Altogether, this can be combined into an objective function: Min w ¸ /2 ||w|| 2 + C  i L(1-z i max h 2 C w T  {s} h s Á s (x i )) Why does inference help?  Constrains intermediate representations supporting good predictions New feature vector for the final decision. Chosen h selects a representation. Inference: best h subject to constraints C

49. Page 49 Optimization Non Convex, due to the maximization term inside the global minimization problem In each iteration:  Find the best feature representation h* for all positive examples (off- the shelf ILP solver)  Having fixed the representation for the positive examples, update w solving the convex optimization problem:  Not the standard SVM/LR: need inference Asymmetry: Only positive examples require a good intermediate representation that justifies the positive label.  Consequently, the objective function decreases monotonically

50. Page 50  Formalized as Structured SVM + Constrained Hidden Structure  LCRL: Learning Constrained Latent Representation Iterative Objective Function Learning Inference best h subj. to C Prediction with inferred h Training w/r to binary decision label Initial Objective Function Generate features Update weight vector Feedback relative to binary problem ILP inference discussed earlier; restrict possible hidden structures considered.

51. Page 51 Experimental Results Transliteration: Recognizing Textual Entailment: Paraphrase Identification:*

52. Page 52 II. Structured Prediction Before, the structure was in the intermediate level  We cared about the structured representation only to the extent it helped the final binary decision  The binary decision variable was given as supervision What if we care about the structure?  Information Extraction; Relation Extraction; POS tagging, many others. Invent a companion binary decision problem!  Parse Citations: Lars Ole Andersen. Program analysis and specialization for the C Programming language. PhD thesis. DIKU, University of Copenhagen, May 1994.  Companion: Given a citation; does it have a legitimate citation parse?  POS Tagging  Companion: Given a word sequence, does it have a legitimate POS tagging sequence? Binary Supervision is almost free XY H

53. Page 53 Predicting phonetic alignment (For Transliteration) Target Task  Input: an English Named Entity and its Hebrew Transliteration  Output: Phonetic Alignment (character sequence mapping)  A structured output prediction task (many constraints), hard to label Companion Task  Input: an English Named Entity and an Hebrew Named Entity  Companion Output: Do they form a transliteration pair?  A binary output problem, easy to label  Negative Examples are FREE, given positive examples ylatI י ט לי הא Target Task Yes/No Why it is a companion task? Companion Task נ ל יי וא י I l l i n o i s

54. Page 54 Companion Task Binary Label as Indirect Supervision The two tasks are related just like the binary and structured tasks discussed earlier All positive examples must have a good structure Negative examples cannot have a good structure We are in the same setting as before  Binary labeled examples are easier to obtain  We can take advantage of this to help learning a structured model Algorithm: combine binary learning and structured learning XY H Positive transliteration pairs must have “good” phonetic alignments Negative transliteration pairs cannot have “good” phonetic alignments

55. Page 55 Joint Learning Framework Joint learning : If available, make use of both supervision types ylatI י ט לי הא Target Task Yes/No Loss on Target TaskLoss on Companion Task Loss function – same as described earlier. Key: the same parameter w for both components Companion Task נ ל יי וא י I l l i n o i s

56. Page 56 Experimental Result Very little direct (structured) supervision.

57. Page 57 Experimental Result Very little direct (structured) supervision. (Almost free) Large amount binary indirect supervision

58. Page 58 Driving supervision signal from World’s Response Can I get a coffee with no sugar and just a bit of milk Can we rely on this interaction to provide supervision? MAKE(COFFEE,SUGAR=NO,MILK=LITTLE) Arggg Great! Semantic Parser

59. Page 59 Traditional approach: learn from logical forms and gold alignments EXPENSIVE! Semantic parsing is a structured prediction problem: identify mappings from text to a meaning representation Query Response: Supervision = Expected Response Check if Predicted response == Expected response Logical Query Real World Feedback Interactive Computer System Pennsylvania Query Response : r largest( state( next_to( const(NY)))) y “What is the largest state that borders NY?" NL Query x Train a structured predictor with this binary supervision ! Expected : Pennsylvania Predicted : NYC Negative Response Pennsylvania r Binary Supervision Expected : Pennsylvania Predicted : Pennsylvania Positive Response Our approach: use only the responses

60. Page 60 Response Based Learning X: What is the largest state that borders NY? Y: largest( state( next_to( const(NY)))) Use the expected response as supervision Additional difficulty:  Algorithm generates potential structures, tested against the DB oracle.  Difficult to generate positive examples (good structures) Learning approach – iteratively identify more correct structures  DIRECT protocol: Convert the learning problem into binary prediction  AGGRESSIVE protocol: Convert the feedback into structured supervision  COMBINED approach, with a weighted loss function Domain’s semantics is used to constrain interpretations  declarative constraints: Lexical resources (wordnet); type consistency: distance in sentence, in dependency tree,… Repeat for all input sentences do Find best structured output Query feedback function end for Learn new W using feedback Until Convergence

61. Page 61 Empirical Evaluation [CoNLL’10,ACL’11,Current] Key Question: Can we learn from this type of supervision? Algorithm# training structures Test set accuracy No Learning: Initial Objective Fn Binary signal: DIRECT Protocol 0000 22.2% 69.2 % Binary signal: AGGRESSIVE Protocol073.2 % Binary Signal: COMBINED Protocol081.6% WM*2007 (fully supervised – uses gold structures) 31075 % *[WM] Y.-W. Wong and R. Mooney. 2007. Learning synchronous grammars for semantic parsing with lambda calculus. ACL. Current emphasis: Learning to understand natural language instructions for games via response based learning

62. Page 62 Data vs. Annotated Data One of the key challenges facing Machine Learning today is that of lack of annotated data. We do have huge amounts of data…but not annotated as required by many tasks:  Semantic Parsing  Event Extraction and Analysis  Co-reference resolution  …. Our goal was to show that sometimes, minimal, easy to come by supervision can be amplified and used The next case provides an extreme example.

63. Reference Resolution 63

64. Document 1: The Justice Department has officially ended its inquiry into the assassinations of John F. Kennedy and Martin Luther King Jr., finding ``no persuasive evidence'' to support conspiracy theories, according to department documents. The House Assassinations Committee concluded in 1978 that Kennedy was ``probably'' assassinated as the result of a conspiracy involving a second gunman, a finding that broke from the Warren Commission 's belief that Lee Harvey Oswald acted alone in Dallas on Nov. 22, 1963. Document 2: In 1953, Massachusetts Sen. John F. Kennedy married Jacqueline Lee Bouvier in Newport, R.I. In 1960, Democratic presidential candidate John F. Kennedy confronted the issue of his Roman Catholic faith by telling a Protestant group in Houston, ``I do not speak for my church on public matters, and the church does not speak for me.'‘ Document 3: David Kennedy was born in Leicester, England in 1959. …Kennedy co- edited The New Poetry (Bloodaxe Books 1993), and is the author of New Relations: The Refashioning Of British Poetry 1980-1994 (Seren 1996). Kennedy The Reference Problem The same problem exists with other types of entities & Concepts Decisions with respect to entities and concepts are interdependent (and depend on a lot more things beyond the variables of interests) 64

65. Organizing knowledge It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II. 65

66. Organizing knowledge It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II. 66

67. Cross-document co-reference resolution It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II. 67

68. Reference resolution: (disambiguation to Wikipedia) 68 It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II.

69. The “reference” collection has structure 69 It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II. Used_In Is_a Succeeded Released

70. Analysis of Information Networks 70 It’s a version of Chicago – the standard classic Macintosh menu font, with that distinctive thick diagonal in the ”N”. Chicago was used by default for Mac menus through MacOS 7.6, and OS 8 was released mid-1997.. Chicago VIII was one of the early 70s-era Chicago albums to catch my ear, along with Chicago II.

71. Wikipedia as a knowledge (and supervision) resource 71 Used_In Is_a Succeeded Released

72. Problem formulation - Matching/Ranking problem Text Document(s)—News, Blogs,… Wikipedia Articles 72 We view this, too, as a constrained optimization problem

73. Local approach  Γ is a solution to the problem  A set of pairs (m,t)  m: a mention in the document  t: the matched Wikipedia title Text Document(s)—News, Blogs,… Wikipedia Articles 73

74. Local approach  Γ is a solution to the problem  A set of pairs (m,t)  m: a mention in the document  t: the matched Wikipedia title Local score of matching the mention to the title Text Document(s)—News, Blogs,… Wikipedia Articles 74

75. A Global Augmentation 1.Invent a surrogate solution Γ’; disambiguate each mention independently. 2.Evaluate the structure based on pair- wise coherence scores Ψ(t i,t j ) Text Document(s)—News, Blogs,… Wikipedia Articles Training the models is very involved and relies on the correctness of the (partial) link structure in Wikipedia But – requires no annotation, relying on Wikipedia 75

76. Demo State of the art results [ACL’11] 76

77. Wikifikation: Demo Screen Shot (Demo)Demo http://en.wikipedia.org/wiki/Mahmoud_Abbas 77

78. Robustness Across Domains (Demo)Demo http://en.wikipedia.org/wiki/Bone_tumor http://en.wikipedia.org/wiki/Protein_precursor The training method used resulted in robustness across domains 78

79. Page 79 Conclusion Understanding Natural Language requires making decisions that rely both on statistical learning and on the incorporation of (declarative) background knowledge. We presented Constrained Conditional Models: A computational framework for learning and inference that augments probabilistic models with declarative constraints (within an ILP formulation). We discussed learning protocols that aim at reduced annotation cost:  Constrained Driven Learning (CoDL)  Learning structure with indirect supervision  Response Based Learning Many open issues… Check out our tools & demos

80. Page 80 Questions? Thank you