Neural Dependency Parsing of Morphologically-Rich Languages

Language and Computation Courses

Advanced Course

Neural Dependency Parsing of Morphologically-Rich Languages,
Erhard Hinrichs (University of Tübingen, Germany) and Daniël de Kok (University of Tübingen, Germany)

This course will introduce parsing of morphologically-rich languages using neural transition-based dependency parsers. Until recently, most work on neural transition-based dependency parsing was conducted on English. However, with the recent introduction of the Universal Dependency annotation scheme and corresponding treebanks for 50 languages (De Marneffe et al., 2014; Nivre et al., 2016), it seems timely to explore neural transition-based dependency parsing for other languages. In this course, we will focus on morphologically-rich languages and will draw on our own dependency parsing research as one use case of this kind. More generally, we will discuss how information about the macrostructure of different clause types (de Kok and Hinrichs, 2016) and about word-level morphology (de Kok, 2015; Ballesteros et al., 2015) can substantially improve parsing accuracy for such languages. We will further show how recurrent neural networks can be used to model such information.