1 / 41

Minimally Supervised Event Causality Identification

Minimally Supervised Event Causality Identification. Quang Do, Yee Seng, and Dan Roth University of Illinois at Urbana-Champaign. EMNLP-2011. Event Causality. The police arrested him because he killed someone. Event Causality. The police arrested him because he killed someone.

lewis-dixon
Télécharger la présentation

Minimally Supervised Event Causality Identification

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Minimally Supervised Event Causality Identification Quang Do, Yee Seng, and Dan Roth University of Illinois at Urbana-Champaign EMNLP-2011

  2. Event Causality The police arrested him because he killed someone.

  3. Event Causality The police arrested him because he killed someone. event trigger event trigger

  4. Event Causality causality The police arrested him because he killed someone. event trigger event trigger • We identify causality between event pairs, but not the direction

  5. Event Causality calculate causality association: co-occurrence counts, pointwise mutual information (PMI)… The police arrested him because he killed someone.

  6. Event Causality connective The police arrested him because he killed someone. contingency discourse relation

  7. Event Causality Distributional distributional association score The police arrested him because he killed someone. Discourse discourse relation prediction • Identify multiple cues to jointly identify event causality: • Distributional association scores • discourse relation predictions

  8. Cause-Effect Association (CEA) andDiscourse Relations [ … … … ] connective [ … … … ] Distributional association between the predicate of an event and the arguments of the other event association between event predicates association between event arguments Discourse e e e e • A connective is associated with two text spans • Training on the Penn Discourse Treebank (PDTB), we developed a system that predicts the discourse relations of expressed by the connectives We define an event e as: p(a1, a2, …, an):

  9. Event Definition • We define an event e as: p(a1, a2, …, an): • predicate p: the event trigger word • a1, a2, …, an: arguments associated with e • Examples: • Verbs: “… hekilledsomeone …” • Nominals: “… the attack by the troops …”

  10. Contributions (Event Causality) • We identify causality between event pairs in context: • verb-verb, verb-noun, noun-noun triggered event pairs • (prior work usually focus on just verb triggers) • A minimally supervised approach to detect event causality using distributional similarity methods • Leverage the interactions between event causality prediction and discourse relations prediction

  11. Overview (Event Causality) • Event causality: • Interaction between event causality and discourse relations • Event predicates: verbs, nominals • Cause-Effect Association (CEA) • Discourse and Causality: • Discourse relations • Constraints for joint inference with CEA • Experiments: • Settings • Evaluation • Analysis • Conclusion

  12. Overview (Event Causality) • Event causality: • Interaction between event causality and discourse relations • Event predicates: verbs, nominals • Cause-Effect Association (CEA) • Discourse and Causality: • Discourse relations • Constraints for joint inference • Experiments: • Settings • Evaluation • Analysis • Conclusion

  13. Cause-Effect Association (CEA) CEA: prediction of whether two events are causally related The police arrested him because he killed someone.

  14. Cause-Effect Association (CEA) association between event predicates association between event arguments association between the predicate of an event and the arguments of the other event • We define an event e as: p(a1, a2, …, an): • predicate p: the event trigger word (e.g.: arrested, killed) • a1, a2, …, an: arguments associated with e

  15. Predicate-Predicate Association

  16. Predicate-Predicate Association D: total number of documents in the collection N: number of documents that p occurs in

  17. Predicate-Predicate Association awards event pairs that are closer together in the texts (in terms of num# of sentences apart), while penalizing event pairs that are further apart

  18. Predicate-Predicate Association takes into account whether predicates (events) pi and pj appear most frequently with each other

  19. Predicate-Predicate Association takes into account whether predicates (events) pi and pj appear most frequently with each other ui will be maximized if there is no other predicate pk (as compared to pj) having a higher co-occurrence probability with pi

  20. Predicate-Argument Association Pair up the predicates and arguments across events, calculate the PMI for each link, then average them

  21. Argument-Argument Association calculate the PMI for each possible pairings of the arguments (across the two events), then average them

  22. Cause-Effect Association (CEA) CEA score: predicts whether the two events are causally related The police arrested him because he killed someone.

  23. Overview (Event Causality) • Event causality: • Interaction between event causality and discourse relations • Event predicates: verbs, nominals • Cause-Effect Association (CEA) • Discourse and Causality: • Discourse relations • Constraints for joint inference with CEA • Experiments: • Settings • Evaluation • Analysis • Conclusion

  24. Discourse and Causality Interaction [ … … … ] connective [ … … … ] • Interaction between: • Discourse relation evoked by the connective c • Relations between ep (event pairs that crosses the two text spans) e e e e causal? not-causal?

  25. Penn Discourse Treebank (PDTB) Relations • Discourse relations: • Comparison: • Concession, Contrast, Pragmatic-concession, Pragmatic-contrast • Contingency: • Cause, Condition, Pragmatic-cause, Pragmatic-condition • Expansion: • Alternative, Conjunction, Exception, Instantiation, List, Restatement • Temporal: • Asynchronous, Synchronous

  26. Discourse Relations Contrast: [According to the survey, x% of Chinese Internet users prefer Google] whereas [y% prefer Baidu]. Cause: [The first priority is search and rescue] because [many people are trapped under the rubble]. • Comparison: • Highlights differences between the situations described in the text spans: • Negative evidence for causality • Contingency: • The situation described in one text span causally influences the situation in the other: • Positive evidence

  27. Discourse Relations Conjunction: [Over the past decade, x women were killed] and [y went missing]. Synchrony: [He was sitting at his home] when [the whole world started to shake]. • Expansion: • Providing additional information, illustrating alternative situations, etc.: • Negative evidence, except for Conjunction (which connects arbitrary pieces of text spans) • Temporal: • Temporal precedence of the (cause) event over the (effect) event is a necessary, but not sufficient requisite for causality

  28. Discourse and Causality Interaction [ … … … ] connective [ … … … ] 2 1 3 Cause, Condition, Temporal, Asynchronous, Synchrony, Conjunction ei ei ei ej ej ej Cause, Condition If we have a (crossing) ep which is causal At least one (crossing) ep is causal Comparison, Concession, Contrast, Pragmatic-concession, Pragmatic-contrast, Expansion, Alternative, Exception, Instantiation, List, Restatement e e e e No (crossing) ep is casual

  29. Joint Inference: Discourse & Distributional Causality Probability that connective c is predicted with discourse relation dr CEA prediction that event pair ep takes on the causal or not-causal label er discourse relation indicator variable event pair causality indicator variable Objective function:

  30. Constraints [ … … … ] connective [ … … … ] 1 ei ej Cause, Condition At least one (crossing) ep is causal e e e e If the connective is predicted with a “Cause” discourse relation, then the CEA system should predict that at least one of the (crossing) event pair is causally related

  31. Constraints [ … … … ] connective [ … … … ] 2 Cause, Condition, Temporal, Asynchronous, Synchrony, Conjunction ei ej If we have a (crossing) ep which is causal e e e e If a (crossing) event pair is predicted by CEA as causally related, then the associated connective should be predicted as having discourse relation; “Cause”, “Condition”, …, “Conjunction”

  32. Constraints [ … … … ] connective [ … … … ] 3 ei ej Comparison, Concession, Contrast, Pragmatic-concession, Pragmatic-contrast, Expansion, Alternative, Exception, Instantiation, List, Restatement e e e e No (crossing) ep is casual {“Comparison”,”Concession”…} If the connective is predicted with discourse relation “Comparison”, “Concession”, …, “Restatement”; no (crossing) event pair is causally related

  33. Overview (Event Causality) • Event causality: • Interaction between event causality and discourse relations • Event predicates: verbs, nominals • Cause-Effect Association (CEA) • Discourse and Causality: • Discourse relations • Constraints for joint inference • Experiments: • Settings • Evaluation • Analysis • Conclusion

  34. Experimental Settings • To collect the distributional statistics for measuring CEA: 760K documents in the English Gigaword corpus • 25 CNN documents from first three months of 2010: • 20 documents for evaluation • 5 documents for development

  35. Annotation for Causal Event Pairs • Annotation guidelines: • The Cause event should temporally precede the Effect event; the Effect event occurs because the Cause event occurs

  36. Annotation for Causal Event Pairs Document R (relatedness) C (causality) … Si-1 Si Si+1 … … R (relatedness) • Drawing links between event predicates: • Event arguments are not annotated, but annotators are free to look at the entire document text • Annotators are not restricted to a fixed sentence window size

  37. Annotation for Causal Event Pairs • Annotators overlap on 10 evaluation documents. Agreement ratio: • 0.67 for C+R • 0.58 for C

  38. Performance on Extracting Causality

  39. Performance on Extracting Causality and Relatedness

  40. Analysis of CEA mistakes • 50 (randomly selected) false-positives (precision errors): • 56%: CEA assigns a high score to event pairs that are not causal • 22%: involves events containing pronouns (“he”, “it”, etc.) as arguments • 50 false-negatives (recall errors): • 23%: CEA assigns a low score to causal event pairs • 19%: involving nominal predicates that are not in our list of event evoking noun types • 17%: involving nominal predicates without any argument (less information for CEA) • 15%: involves events containing pronouns as arguments

  41. Conclusion (Event Causality) Developed a minimally supervised approach to identify event causality Use distributional scores and discourse relations to jointly identify event causality

More Related