Vine Parsing and Minimum Risk Reranking for Speed and Precision - - PowerPoint PPT Presentation

vine parsing and minimum risk reranking
SMART_READER_LITE
LIVE PREVIEW

Vine Parsing and Minimum Risk Reranking for Speed and Precision - - PowerPoint PPT Presentation

Vine Parsing and Minimum Risk Reranking for Speed and Precision Markus Dreyer David A. Smith Noah A. Smith Johns Hopkins University Design Goals Speed Precision CoNLL-2006 M. Dreyer, D. A. Smith, N. A. Smith Vine Parsing and Minimum


slide-1
SLIDE 1

Vine Parsing and Minimum Risk Reranking

for Speed and Precision

Markus Dreyer David A. Smith Noah A. Smith

Johns Hopkins University

slide-2
SLIDE 2

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Design Goals

Speed Precision

slide-3
SLIDE 3

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Reranker

Labeler

Unlabeled

Parser

Charniak and Johnson (2005) Collins (2000)

U×L-best labeled parses U-best unlabeled parses

S m

  • t

h e d M L E S m

  • t

h e d M L E L

  • g
  • l

i n e a r

slide-4
SLIDE 4

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Unlabeled

Parser

Split-Head assumption Projectivity Vine Grammar

Eisner and N. Smith (2005)

slide-5
SLIDE 5

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

slide-6
SLIDE 6

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

slide-7
SLIDE 7

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

slide-8
SLIDE 8

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

slide-9
SLIDE 9

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

slide-10
SLIDE 10

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

D i f f e r e n t b

  • u

n d s f

  • r

l e f t

a n d right c h i l d r e n

C h

  • s

e b

  • u

n d s : 9 %

  • f
  • r

i g i n a l d e p e n d e n c i e s u n t

  • u

c h e d S p e e d

  • A

c c u r a c y T r a d e

  • f

f

O(nBleft

2 +nBright 2

)

slide-11
SLIDE 11

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Reranker

Deterministic annealing Minimum Risk Training

  • D. Smith and Eisner (2006)
slide-12
SLIDE 12

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Labeled

M e d i a n 6 7 . 6

Unlabeled

M e d i a n 7 7 . 5 Japanese (86.0), Portuguese (82.4), Bulgarian (82.0), Swedish (79.5), Chinese (77.6) Japanese (82.9), Portuguese (75.3), Bulgarian (74.8), Chinese (71.6), German (71.0)

slide-13
SLIDE 13

CoNLL-2006  M. Dreyer, D. A. Smith, N. A. Smith  Vine Parsing and Minimum Risk Reranking

Summary

 Parsing constraints

Linear-time

inference and decoding  Minimum Risk reranking  High precision, mediocre recall

Reranker

Labeler

Unlabeled

Parser

Future Work

 Better estimation  Better labeler (label bigrams)  More fine-grained parsing constraints (length bounds given head)