cross lingual ccg induction
play

Cross-lingual CCG Induction Kilian Evang @texttheater University - PowerPoint PPT Presentation

Introduction Derivation Projection Experiments Conclusions References Cross-lingual CCG Induction Kilian Evang @texttheater University of D usseldorf 2019-06-04 NAACL-HLT 1 / 23 Introduction Derivation Projection Experiments


  1. Introduction Derivation Projection Experiments Conclusions References Cross-lingual CCG Induction Kilian Evang @texttheater University of D¨ usseldorf 2019-06-04 NAACL-HLT 1 / 23

  2. Introduction Derivation Projection Experiments Conclusions References Outline Introduction Derivation Projection Experiments Conclusions 2 / 23

  3. Introduction Derivation Projection Experiments Conclusions References Outline Introduction Derivation Projection Experiments Conclusions 3 / 23

  4. Introduction Derivation Projection Experiments Conclusions References Combinatory Categorial Grammar We sang NP S \ NP < 0 S 3 / 23

  5. Introduction Derivation Projection Experiments Conclusions References Combinatory Categorial Grammar We saw the car that John bought NP (S \ NP) / NP NP / N N (N \ N) / (S / NP) N (S \ NP) / NP ∗ NP T > S / (S \ NP) > 1 S / NP > 0 N \ N < 0 N > 0 > 0 NP NP > 0 S \ NP < 0 S 4 / 23

  6. Introduction Derivation Projection Experiments Conclusions References Appeal • coordination • universal rules • syntax-semantics interface 5 / 23

  7. Introduction Derivation Projection Experiments Conclusions References Most CCG Parsers • trained on large treebanks, or • hand-crafted 6 / 23

  8. Introduction Derivation Projection Experiments Conclusions References David Blackwell, CC-BY-NC What about low-resource languages? 7 / 23

  9. Introduction Derivation Projection Experiments Conclusions References Unsupervised CCG Induction? target-language text + magic = target-language CCG parser (Bisk and Hockenmaier, 2013; Bisk et al., 2015) 8 / 23

  10. Introduction Derivation Projection Experiments Conclusions References Cross-lingual CCG Induction? English CCG parser + parallel corpus + magic = target-language CCG parser 9 / 23

  11. Introduction Derivation Projection Experiments Conclusions References Cross-lingual CCG Induction via Derivation Projection parallel corpus + English CCG derivations + word alignments + derivation projection = target-language CCG derivations = target-language training data 10 / 23

  12. Introduction Derivation Projection Experiments Conclusions References Outline Introduction Derivation Projection Experiments Conclusions 11 / 23

  13. Introduction Derivation Projection Experiments Conclusions References Derivation Projection • project lexical categories along word alignments • n:1 alignment → merge • word order difference → flip slash 11 / 23

  14. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli 12 / 23

  15. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli N 4 / N 5 12 / 23

  16. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli N 4 / N 5 N 5 12 / 23

  17. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli S 2 / NP 3 N 4 / N 5 N 5 12 / 23

  18. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli S 2 / NP 3 N 4 / N 5 N 5 > 0 N 4 12 / 23

  19. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli S 2 / NP 3 N 4 / N 5 N 5 > 0 N 4 ∗ NP 3 12 / 23

  20. Introduction Derivation Projection Experiments Conclusions References Example 1/3 S 2 < 0 S 2 \ NP 1 > 0 NP 3 ∗ N 4 > 0 NP 1 (S 2 \ NP 1 ) / NP 3 N 4 / N 5 N 5 He had three sons Aveva tre figli S 2 / NP 3 N 4 / N 5 N 5 > 0 N 4 ∗ NP 3 > 0 S 2 12 / 23

  21. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa 13 / 23

  22. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 13 / 23

  23. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 13 / 23

  24. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 (N 2 \ N 3 ) / (N 4 \ N 5 ) 13 / 23

  25. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 (N 2 \ N 3 ) / (N 4 \ N 5 ) N 4 \ N 5 13 / 23

  26. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 (N 2 \ N 3 ) / (N 4 \ N 5 ) N 4 \ N 5 > 0 N 2 \ N 3 13 / 23

  27. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 (N 2 \ N 3 ) / (N 4 \ N 5 ) N 4 \ N 5 > 0 N 2 \ N 3 < 0 N 2 13 / 23

  28. Introduction Derivation Projection Experiments Conclusions References Example 2/3 NP 1 > 0 N 2 > 0 N 2 / N 3 > 0 NP 1 / N 2 (N 2 / N 3 ) / (N 4 / N 5 ) N 4 / N 5 N 3 a very decorative plant una pianta molto decorativa NP 1 / N 2 N 3 (N 2 \ N 3 ) / (N 4 \ N 5 ) N 4 \ N 5 > 0 N 2 \ N 3 < 0 N 2 > 0 NP 1 13 / 23

  29. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht 14 / 23

  30. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht NP 8 14 / 23

  31. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht NP 8 (S 5 \ NP 6 ) \ (S 3 \ NP 4 ) 14 / 23

  32. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht (S 3 \ NP 4 ) / NP 8 NP 8 (S 5 \ NP 6 ) \ (S 3 \ NP 4 ) 14 / 23

  33. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht (S 3 \ NP 4 ) / NP 8 NP 8 (S 5 \ NP 6 ) \ (S 3 \ NP 4 ) > 0 S 3 \ NP 4 14 / 23

  34. Introduction Derivation Projection Experiments Conclusions References Example 3/3 S 5 \ NP 6 > 0 S 3 \ NP 4 > 0 (S 5 \ NP 6 ) / (S 3 \ NP 4 ) (S 3 \ NP 4 ) / PR 7 < 1 > 0 × (S 1 \ NP 2 ) / (S 3 \ NP 4 ) (S 5 \ NP 6 ) \ (S 1 \ NP 2 ) ((S 3 \ NP 4 ) / PR 7 ) / NP 8 NP 8 PR 7 Do n’t mess it up Versau es nicht (S 3 \ NP 4 ) / NP 8 NP 8 (S 5 \ NP 6 ) \ (S 3 \ NP 4 ) > 0 S 3 \ NP 4 < 0 S 5 \ NP 6 14 / 23

  35. Introduction Derivation Projection Experiments Conclusions References Outline Introduction Derivation Projection Experiments Conclusions 15 / 23

  36. Introduction Derivation Projection Experiments Conclusions References Training • Parallel corpus: tatoeba.org • English parser: EasyCCG trained on CCGrebank • Word aligments: GIZA++ • Target-language parser: EasyCCG trained on projected derivations 15 / 23

Recommend


More recommend