< Back to previous page
Project
Statistical Relational Learning of Natural Language
While understanding natural language is easy for humans, it is complex for
computers. The main reasons for this can be found in the structural nature and
the inherent ambiguity of natural language. Correctly interpreting language
therefore requires one to take into account the necessary context. In order
to perform natural language understandingby means of machine learning
techniques, an appropriate representation is required that takes into account
this relational information and integrates the necessary background knowledge
into the learning process.
Statistical relational learning is well-suited to represent this structural
information, and to incorporate the necessary context and background knowledge
for natural language understanding. Furthermore, its inherent probabilistic
nature offers opportunities todeal with linguistic ambiguity. This thesis
investigates the promiseof statistical relational learning for natural language
processing and provides evidence for the utility of this approach.
As a first contribution, we demonstrate the expressiveness and interpretability
of the relational representation on two natural language learning problems.
Furthermore, we explore the importance of the declarative approach for the
inclusion of contextual information, and analyze the influence of the relational
representation by a comparison of severalmachine learning techniques. A second
contribution is the extension of the graph kernelbased relational learning
framework kLog with a natural language processing module, in order to obtain
a full relationlearning framework for natural language learning. As a third
contribution, we introduce relational regularization and feature ranking in order
to assess the importance of the relational features. Finally, we extend rule
learning to a probabilistic setting and explore its application in the context of
machine reading.
computers. The main reasons for this can be found in the structural nature and
the inherent ambiguity of natural language. Correctly interpreting language
therefore requires one to take into account the necessary context. In order
to perform natural language understandingby means of machine learning
techniques, an appropriate representation is required that takes into account
this relational information and integrates the necessary background knowledge
into the learning process.
Statistical relational learning is well-suited to represent this structural
information, and to incorporate the necessary context and background knowledge
for natural language understanding. Furthermore, its inherent probabilistic
nature offers opportunities todeal with linguistic ambiguity. This thesis
investigates the promiseof statistical relational learning for natural language
processing and provides evidence for the utility of this approach.
As a first contribution, we demonstrate the expressiveness and interpretability
of the relational representation on two natural language learning problems.
Furthermore, we explore the importance of the declarative approach for the
inclusion of contextual information, and analyze the influence of the relational
representation by a comparison of severalmachine learning techniques. A second
contribution is the extension of the graph kernelbased relational learning
framework kLog with a natural language processing module, in order to obtain
a full relationlearning framework for natural language learning. As a third
contribution, we introduce relational regularization and feature ranking in order
to assess the importance of the relational features. Finally, we extend rule
learning to a probabilistic setting and explore its application in the context of
machine reading.
Date:8 Sep 2009 → 16 Dec 2014
Keywords:Natural language, Relational learning
Disciplines:Applied mathematics in specific fields
Project type:PhD project