This paper presents a novel approach of incorporating fine-grained treebanking decisions made by human annotators as discriminative features for automatic parse disambiguation. To our best knowledge, this is the first work that exploits treebanking decisions for this task. The advantage of this approach is that use of human judgements is made. The paper presents comparative analyses of the performance of discriminative models built using treebanking decisions and state-of-the-art features. We also highlight how differently these features scale when these models are tested on out-of-domain data. We show that, features extracted using treebanking decisions are more efficient, informative and robust compared to traditional features.
Using Treebanking Discriminants as Parse Disambiguation Features
Chowdhury, Faisal Mahbub;
2009-01-01
Abstract
This paper presents a novel approach of incorporating fine-grained treebanking decisions made by human annotators as discriminative features for automatic parse disambiguation. To our best knowledge, this is the first work that exploits treebanking decisions for this task. The advantage of this approach is that use of human judgements is made. The paper presents comparative analyses of the performance of discriminative models built using treebanking decisions and state-of-the-art features. We also highlight how differently these features scale when these models are tested on out-of-domain data. We show that, features extracted using treebanking decisions are more efficient, informative and robust compared to traditional features.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
