chinese semantic role labeling with dependency-driven constituent parse tree structure

20
计计计计计计计计计计 Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure Hongling Wang, Bukang Wang Guodong Zhou NLP Lab, School of Computer Scienc e & Technology, Soochow University

Upload: tanisha-dickerson

Post on 01-Jan-2016

16 views

Category:

Documents


0 download

DESCRIPTION

Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure. Hongling Wang, Bukang Wang Guodong Zhou NLP Lab, School of Computer Science & Technology, Soochow University. Motivation Related work Tree kernel-based nominal SRL on D-CPT Experimentation and Results - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

计算机科学与技术学院

Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Hongling Wang, Bukang Wang Guodong Zhou

NLP Lab, School of Computer Science & Technology, Soochow University

Page 2: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Outlines

Motivation Related work Tree kernel-based nominal SRL on D-CPT Experimentation and Results Conclusion

Page 3: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Motivation(1)

Two kinds of method for SRL Feature-based VS. Tree Kernel-based

The latter has the potential in better capturing structured knowledge in the parse tree structure.

There are only a few studies employing tree kernel-based methods for SRL and most of them focus on the CPT structure.

Page 4: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Motivation(2)

DPT (Dependency Parse Tree) captures the dependency relationship between individual words, while CPT (Constituent Parse Tree) concerns with the phrase structure in a sentence.

CPT and DPT may behave quite differently in capturing different aspects of syntactic phenomena.

We explore a tree kernel-based methods using a new syntactic parse tree structure, called dependency-driven constituent parse tree (D-CPT) for Chinese nominal SRL.

Page 5: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Related Work

Some studies use tree kernel-based methods for verbal SRL Moschitti et al., Che et al., Zhang et al.

There are a few related studies in other NLP tasks employing DPT in tree kernel-based methods and achieve comparable performance to the ones on CPT. semantic relation extraction and co-reference resolution

To our knowledge, there are no reported studies on tree kernel-based methods for SRL from the DPT structure perspective.

Page 6: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

D-CPT Structure

The new D-CPT structure benefits from the advantages of both DPT and CPT since D-CPT not only keeps the dependency relationship information in DPT but also retains the basic structure of CPT.

This is done by transforming the DPT structure to a new CPT-style structure, using dependency types instead of phrase labels in the traditional CPT structure.

Page 7: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Example of DPT with nominal predicate and its related arguments annotated

Page 8: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Example of achieving D-CPT structure from DPT structure

Page 9: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Extraction schemes

1) Shortest path tree (SPT) This extraction scheme only includes the nodes

occurring in the shortest path connecting the predicate and the argument candidate, via the nearest commonly-governing node.

2) SV-SPT This extraction scheme includes the support verb

information in nominal SRL.3) H-SV-SPT

Both of the head argument and the support verb information are included in nominal SRL.

Page 10: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Extraction schemes

Page 11: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Kernels

In order to capture the complementary nature between feature-based methods and tree kernel-based methods, we combine them via a composite kernel.

Our composite kernel is combined by linearly interpolating a convolution tree kernel KT over a parse tree structure and a feature-based linear kernel KL as follow:

Page 12: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Experiment setting

Corpus: Chinese NomBank Training data: 648 files (chtb_081 to 899.fid) Test data: 72 files (chtb_001 to 040.fid and chtb_900 t

o 931.fid) Development data: 40 files (chtb_041 to 080.fid)

Tools Classifier: SVM-light toolkit with the convolution tree k

ernel function SVMlight–TK C (SVM) and λ (tree kernel) are fine-tuned to 4.0 and

0.5 respectively

Page 13: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on golden parse trees (1)

Page 14: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on golden parse trees (2)

Page 15: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on automatic parse trees

Page 16: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on Comparison experments(1)

Page 17: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on Comparison experments(2)

Page 18: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Results on the CoNLL-2009 Chinese corpus

Page 19: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Conclusion

This paper systematically explores a tree kernel-based method on a novel D-CPT

We propose a simple strategy, which transforms the DPT structure into a CPT-style structure .

Several extraction schemes are designed to extract various kinds of necessary information for nominal SRL and verbal SRL

Evaluation shows the effectiveness of D-CPT both on Chinese NomBank and CoNLL-2009 corpus.

Page 20: Chinese Semantic Role Labeling with Dependency-driven Constituent Parse Tree Structure

Thanks!

Q & C?