论文信息 - Voting and Stacking in Data-Driven Dependency Parsing

Voting and Stacking in Data-Driven Dependency Parsing

We compare the techniques of voting and stacking for system combination in datadriven dependency parsing, using a set of eight different transition-based parsers as component systems. Experimental results show that both methods lead to significant improvements over the best component system, and that voting gives the highest overall accuracy. We also investigate different weighting schemes for voting.

Joakim Nivre | Mark Fishel | Joakim Nivre | Mark Fishel

[1] Daniel Zeman,et al. Improving Parsing Accuracy by Combining Diverse Dependency Parsers , 2005, IWPT.

[2] Joakim Nivre,et al. Integrating Graph-Based and Transition-Based Dependency Parsers , 2008, ACL.

[3] Richard Johansson,et al. The CoNLL 2008 Shared Task on Joint Parsing of Syntactic and Semantic Dependencies , 2008, CoNLL.

[4] Alon Lavie,et al. Parser Combination by Reparsing , 2006, NAACL.

[5] Joakim Nivre,et al. An Efficient Algorithm for Projective Dependency Parsing , 2003, IWPT.

[6] Fernando Pereira,et al. Non-Projective Dependency Parsing using Spanning Tree Algorithms , 2005, HLT.

[7] Joakim Nivre,et al. MaltParser: A Data-Driven Parser-Generator for Dependency Parsing , 2006, LREC.

[8] Michael A. Covington,et al. A Fundamental Algorithm for Dependency Parsing , 2004 .

[9] Johan Eklund,et al. Mixing and Blending Syntactic and Semantic Dependencies , 2008, CoNLL.

[10] Joakim Nivre,et al. Single Malt or Blended? A Study in Multilingual Parser Optimization , 2007, EMNLP.

[11] Sabine Buchholz,et al. CoNLL-X Shared Task on Multilingual Dependency Parsing , 2006, CoNLL.