Improving a supervised CCG parser

The central topic of this thesis is the task of syntactic parsing with Combinatory Categorial Grammar (CCG). We focus on pipeline approaches that have allowed researchers to develop efficient and accurate parsers trained on articles taken from the Wall Street Journal (WSJ). We present three approaches to improving the state-of-the-art in CCG parsing. First, we test novel supertagger-parser combinations to identify the parsing models and algorithms that benefit the most from recent gains in supertagger accuracy. Second, we attempt to lessen the future burdens of assembling a state-of-the-art CCG parsing pipeline by showing that a part-of-speech (POS) tagger is not required to achieve optimal performance. Finally, we discuss the deficiencies of current parsing algorithms and propose a solution that promises improvements in accuracy – particularly for difficult dependencies – while preserving efficiency and optimality guarantees.

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

2016

Author dc:creator

• Musa, Ryan A

Contributors dc:contributor

• Hockenmaier, Julia

• Natural language processing
• combinatory categorial grammar
• parsing

Statement dc:rights

• Copyright 2016 Ryan Musa

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/90677

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/90677