Dynamic-oracle transition-based parsing with calibrated probabilistic output

Research output: Contribution to conferencePaperpeer-review

Abstract

We adapt the dynamic-oracle training method of Goldberg and Nivre (2012; 2013) to train classifiers that produce probabilistic output. Evaluation of an Arc-Eager parser on 6 languages shows that the AdaGrad-RDA based training procedure results in models that provide the same high level of accuracy as the averaged-perceptron trained models, while being sparser and providing well-calibrated probabilistic output.

Original languageEnglish
Pages82-90
Number of pages9
StatePublished - 2013
Event13th International Conference on Parsing Technologies, IWPT 2013 - Nara, Japan
Duration: 27 Nov 201329 Nov 2013

Conference

Conference13th International Conference on Parsing Technologies, IWPT 2013
Country/TerritoryJapan
CityNara
Period27/11/1329/11/13

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence
  • Human-Computer Interaction
  • Linguistics and Language

Fingerprint

Dive into the research topics of 'Dynamic-oracle transition-based parsing with calibrated probabilistic output'. Together they form a unique fingerprint.

Cite this