The DOP Approach Comes to Age

Dr. Rens Bod, University of Amsterdam

 

 

The DOP approach of using all subtrees from an annotated corpus for statistical parsing has recently gained renewed interest (e.g. Collins & Duffy 2002; Johnson 2002; Neumann 2003; Joshi & Sarkar 2003). While the computational complexity of using all subtrees deterred its wider use for some time, several new algorithms have recently been developed that allow for fast processing. In this presentation, I will give an overview of Data-Oriented Parsing and go into the latest developments, among which HPSG-DOP, TAG-DOP and Data-Oriented Translation. I will also generalize DOP to "Data-Oriented Perception" and discuss some experiments with musical corpora, suggesting an interesting parallel between linguistic and musical processing.