Tree Transduction Tools for Cdec

• Main Authors: Austin Matthews, Paul Baltescu, Phil Blunsom, Alon Lavie, Chris Dyer
• Format: Article
• Language: English
• Published: Sciendo 2014-09-01
• Series: Prague Bulletin of Mathematical Linguistics
• Online Access: https://doi.org/10.2478/pralin-2014-0011

We describe a collection of open source tools for learning tree-to-string and tree-to-tree transducers and the extensions to the cdec decoder that enable translation with these. Our modular, easy-to-extend tools extract rules from trees or forests aligned to strings and trees subject to different structural constraints. A fast, multithreaded implementation of the Cohn and Blunsom (2009) model for extracting compact tree-to-string rules is also included. The implementation of the tree composition algorithm used by cdec is described, and translation quality and decoding time results are presented. Our experimental results add to the body of evidence suggesting that tree transducers are a compelling option for translation, particularly when decoding speed and translation model size are important.