Joachim Nivre and Sandra Kuebler

Dependency-based methods for syntactic parsing are becoming more and
more popular in the computational linguistics community, as evidenced
e.g. by the fact that dependency parsing has been chosen as the shared
task at CoNLL-06. The aim of this tutorial is to give an overview of
the state of the art in dependency parsing, including computational
methods for dependency analysis as well as available resources for
different languages in terms of parsers and syntactically annotated
data resources. After an introduction, in which the basic terms will
be defined, the three main parsing methods for dependency parsing will
be presented: dependency parsing based on dynamic programming
techniques, dependency parsing as constraint satisfaction, and
dependency parsing with deterministic parsing algorithms combined with
machine learning techniques. The next section will give an overview of
existing implementations and treebanks, followed by a discussion of
the pros and cons of dependency parsing and an outlook on the expected
developments in this area.

This tutorial is designed for researchers working on syntactic
analysis or related topics within other traditions and for application
developers who may be interested in using dependency parsers in their
systems. Acquaintance with basic parsing algorithms will be useful,
but not mandatory.

TUTORIAL OUTLINE

1. Introduction
• basic concepts of dependency grammar and dependency parsing
• define dependency graphs
• basic constraints on dependency graphs
• define dependency parsing as used in tutorial
2. Parsing methods I
• dynamic programming techniques
• constraint satisfaction
• deterministic parsing algorithms
3. Practical Issues
• implementations of dependency parsers
• treebanks
• wiki for dependency parsing
4. Pros and cons of dependency-based methods
5. Outlook

JOAKIM NIVRE's research in recent years has focused on data-driven
methods for syntactic parsing, in particular dependency-based parsing,
as well as on the treebanks needed to develop and evaluate such
parsers. His areas of expertise include algorithms for dependency
parsing and the use of machine learning for syntactic parsing. He is
the main developer of the MaltParser system, which has been used to
develop dependency parsers for a range of different languages.

SANDRA KUEBLER's research in recent years has focused on machine
learning approaches to syntactic parsing as well as on the creation of
treebanks, which are targeted for training and evaluation of parsers
as well as for linguistic research. Her areas of expertise include the
symbolic and probabilistic approaches to syntactic parsing,
dependency-based evaluation of parsers, the conversion of
constituent-based treebanks into dependency representations, and the
evaluation of treebanks concerning their suitability for parsers.