When errors become the rule: A survey of Transformation-Based Learning
(2011) In LU-DV-EX DATM01 20111Computer Science
- Abstract
- Transformation-based learning (TBL) is a machine learning method for sequential classification, invented by Eric Brill (1993, 1995). It is widely used within
natural language processing (but surprisingly little in other areas).
TBL is a simple yet flexible paradigm, which achieves competitive or even state-of-the-art performance in several areas and does not overtrain easily. It is especially successful at catching local, fixed-distance dependencies. The learned representation -- an ordered list of transformation rules -- is compact and efficient, with clear, declarative semantics. Individual rules are interpretable and often meaningful to humans.
The present thesis has two main parts. First and foremost, we offer a survey of the... (More) - Transformation-based learning (TBL) is a machine learning method for sequential classification, invented by Eric Brill (1993, 1995). It is widely used within
natural language processing (but surprisingly little in other areas).
TBL is a simple yet flexible paradigm, which achieves competitive or even state-of-the-art performance in several areas and does not overtrain easily. It is especially successful at catching local, fixed-distance dependencies. The learned representation -- an ordered list of transformation rules -- is compact and efficient, with clear, declarative semantics. Individual rules are interpretable and often meaningful to humans.
The present thesis has two main parts. First and foremost, we offer a survey of the most important theoretical work on TBL. It is intended to be informal but relatively comprehensive, addressing a perceived gap in the literature. Second, in a more practical part, we describe a recursive, parallelizable rephrasing, well suited for declarative languages, of a fast imperative learning algorithm proposed by Ngai and Florian (2001). We implement and test this algorithm in the functional language Haskell. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/2007414
- author
- Uneson, Marcus LU
- supervisor
- organization
- course
- DATM01 20111
- year
- 2011
- type
- M1 - University Diploma
- subject
- keywords
- computer science, computational linguistics, machine learning, sequential classification, brill tagging, transformation-based learning
- publication/series
- LU-DV-EX
- report number
- 2011-157
- ISSN
- 1651-6389
- language
- English
- id
- 2007414
- date added to LUP
- 2011-12-08 11:16:29
- date last changed
- 2011-12-08 11:16:29
@misc{2007414,
abstract = {{Transformation-based learning (TBL) is a machine learning method for sequential classification, invented by Eric Brill (1993, 1995). It is widely used within
natural language processing (but surprisingly little in other areas).
TBL is a simple yet flexible paradigm, which achieves competitive or even state-of-the-art performance in several areas and does not overtrain easily. It is especially successful at catching local, fixed-distance dependencies. The learned representation -- an ordered list of transformation rules -- is compact and efficient, with clear, declarative semantics. Individual rules are interpretable and often meaningful to humans.
The present thesis has two main parts. First and foremost, we offer a survey of the most important theoretical work on TBL. It is intended to be informal but relatively comprehensive, addressing a perceived gap in the literature. Second, in a more practical part, we describe a recursive, parallelizable rephrasing, well suited for declarative languages, of a fast imperative learning algorithm proposed by Ngai and Florian (2001). We implement and test this algorithm in the functional language Haskell.}},
author = {{Uneson, Marcus}},
issn = {{1651-6389}},
language = {{eng}},
note = {{Student Paper}},
series = {{LU-DV-EX}},
title = {{When errors become the rule: A survey of Transformation-Based Learning}},
year = {{2011}},
}