Presentation is loading. Please wait.

Presentation is loading. Please wait.

Semantic Role Chunking Combining Complementary Syntactic Views Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, Daniel Jurafsky  Center for.

Similar presentations


Presentation on theme: "Semantic Role Chunking Combining Complementary Syntactic Views Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, Daniel Jurafsky  Center for."— Presentation transcript:

1 Semantic Role Chunking Combining Complementary Syntactic Views Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, Daniel Jurafsky  Center for Spoken Language Research Department of Computer Science University of Colorado at Boulder  Department of Linguistics Stanford University

2 Different Syntactic Views Hypothesis: Different views make different errors Two views: Phrase structure based (Charniak, Collins) Chunk based

3 Constituents from Charniak parse tree Charniak Parse Tree Constituent Views John kicked the ball. Collins Parse Tree Constituents from Collins parse tree

4 Chunk View Salomon will buy sufficient shares to cover its entire position O O O B-A2 I-A2 O B-V B-A1 I-A1 Chunk using an IOB representation [Ramshaw & Marcus, 1995] Yamcha [Kudo & Matsumoto, 2001] Bottom up as opposed to top down Flat representation Uses flat syntactic chunks [Hacioglu & Ward 2003]

5 Algorithm Generate Charniak and Collins parse based features Add few features from one to the other Generate semantic IOB tags using these views Use them as features Generate the final semantic role label set using a phrase-based chunking paradigm

6 Architecture ChunkerCharniakCollinsWords Features IOB Semantic Role Labels IOB Phrases

7 Illustration Train Model 1 2R B B B B B BB I I I I I I I I I I II I I I I I O O O O O O O O O O O O O O B B O O

8 Features Semantic IOB tags for Charniak and Collins based semantic role labels [Pradhan et al., 2005] Phrase level chunk features [Hacioglu et al., 2004]

9 Active Learning Randomly selelected 10k examples and trained a N ULL vs A RGUMENT classifier Classified remaining examples using this classifier Added misclassified examples to the seed set Iterated Final data amounted to about a third of the total

10 Combination Results Test: Section 24 of PropBank Train: Sections 02-21 of PropBank ID + Class ASSERT Charniak System P R F 1 80 75 77 ASSERT Collins 79 74 76 ASSERT Combined 81 76 78

11 Results Section 24 Submitted System System P R F 1 80.9 75.4 78.0 Section 23 P R F 1 81.9 73.3 77.4 Brown P R F 1 73.7 61.5 67.1 Bug fixed System 81.9 75.1 78.382.9 74.7 78.674.5 63.3 68.4 ID + Class

12 Thank You Arda AQUAINT program contract OCG4423B NSF grant IS-9978025

13 Software ASSERT (Automatic Statistical SEmantic Role Tagger) Publicly downloadable at http://oak.colorado.edu/asserthttp://oak.colorado.edu/assert Downloaded by more than 50 research groups

14 Null Filtering Removed constituents with P(N ULL ) > 0.9 Removed phrases with P(N ULL ) > 0.8 after incorporating context

15 Analysis Active learning using confidence threshold Constituent level instead of Sentence level N-Best Charniak parses

16 Features (Constituent)

17

18 Features (Phrase)

19

20 Representation

21 Features

22

23 But analysts reckon underlying support for sterling has been eroded by the chancellor 's failure to announce any new policy measures in his Mansion House speech last Thursday Minipar-based Semantic Labeling Rule-based dependency parser


Download ppt "Semantic Role Chunking Combining Complementary Syntactic Views Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, Daniel Jurafsky  Center for."

Similar presentations


Ads by Google