Output details
11 - Computer Science and Informatics
University of Edinburgh
Inducing Compact but Accurate Tree-Substitution Grammars
<22> Originality: The first paper to formalize estimation of a tree-substitution grammar as Bayesian inference.
Significance: Method has been adopted for a wide range of applications including machine translation (Cohn and Blunsom, 2009), dependency parsing (Blunsom and Cohn, 2010), sentence compression (Yamangil and Shieber, 2010), and multiword expression identification (Green et al., 2011). Also cited in a recent textbook on Linguistic Structure Prediction (Smith, 2011).
Rigour: Avoids the statistical and computational problems of previous methods (e.g., inconsistency, overfitting); experiments show parsing results comparable to state-of-the-art parsers. Above cited papers show the method can also be applied successfully to other tasks.