Introduction to Data Driven Dependency Parsing
At ESSLLI 2007 in Dublin, August 13-17th
Desription: Syntactic dependency representations of sentences have a long history in theoretical linguistics. Recently, they have found renewed interest in the computational parsing community due to their efficient computational properties and their ability to naturally model non-nested constructions, which is important in freer-word order languages such as Czech, Dutch, and German. This interest has led to a rapid growth in multilingual data sets and new parsing techniques. One modern approach to building dependency parsers, called data-driven dependency parsing, is to learn good and bad parsing decisions solely from labeled data, without the intervention of an underlying grammar. This course will cover:
- Dependency parsing (history, definitions, motivation, etc.)
- Grammar-driven versus data-driven parsing
- Brief intro to learning algorithms: Generative, Memory-based, SVM, etc.
- Greedy parsing algorithms: Covington, Nivre, Yamada and Matsumoto
- Graph-based algorithms: MST algorithms (Eisner, McDonald et al)
- Other approaches
- Empirical results and applications
- Available software
Slides (subject to change from Aug 13-17)
- Lecture 1 (Intro to dependency parsing)
- Lecture 2 (Intro to data-driven methods and ML)
- Lecture 3 (Transition-based dependency parsing)
- Lecture 4 (Graph-based dependency parsing)
- Lecture 5 (Other approaches, empirical eval, available software, treebanks, etc.)
Coarse Reader
Links