Semantic role labeling using rich morphological features

In this thesis, I examine the impact of morphological features on semantic role labeling (SRL) in Modern Standard Arabic. This study provides the first rigorous examination of the effect of individual morphological attributes on the automatic classification of predicate-argument roles in Arabic. I obtain a classification accuracy comparable to the best-published previous results using fewer features and a linear kernel SVM. This provides evidence that there is a strong morpho-semantic interaction in Arabic and that careful, language-dependent feature engineering may provide substantial improvements to morphologically rich languages. I demonstrate that the incorporation of rich-morphological features substantially outperform the established lexico-syntactic featuresets of standard SRL systems. A principled feature selection algorithm for SRL is also introduced that significantly decreases generalization error, leading to state-of-the-art argument role classification results.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2013

Author dc:creator

• Lundgren, David

Contributors dc:contributor

• Girju, Roxana

• semantic role labeling (SRL)
• Arabic
• morphology
• semantics

Statement dc:rights

• Copyright 2013 David Lundgren

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/44091

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/44091