Towards a Universal Stanford Dependencies parallel treebank

Bosco Cristina;Sanguinetti Manuela

2014-01-01

---

Abstract

The paper describes the process conceived to convert a multilingual parallel treebank, namely ParTUT, into an annotated resource that conforms to the representation format and specifications of the Universal Stanford Dependencies (USD). The main goal of this work is to create, taking an already existing resource as the starting point, a fully parallel treebank that is featured by a widely-known and used representation format, i.e. that of the Stanford Dependencies, and in particular its cross-linguistic variant, namely the Universal Stanford Dependencies, in order to provide a useful resource for a number of NLP tasks, including those that have typically benefitted from such representation format, such as Information Extraction and statistical parsing, but also translation-related tasks (by virtue of the parallel annotation).