Welcome to Treeler

Treeler is an open-source c++ library of structure prediction methods, focusing on Natural Language Processing tasks like tagging and parsing. It is released under the  GNU GPL.

Treeler implements a framework for linear structure prediction. The main features of the framework are:

  • Structured prediction models take the form of factored predictors. A central component behind a structured prediction model is a factorization or decomposition of structures into "parts". The model is then defined according to this factorization.
  • The library provides learning algorithms that are generic with respect to the particular factorization employed by a model. So far, the library implements learning algorithms for classification that have been ported to structure prediction. This includes Perceptron, log-linear models, and max-margin methods.
  • The library provides standard factored models for multiclass classification, sequence tagging and dependency parsing and semantic role labeling.

For an overview of Treeler, check the  video presentation at WAPA'2011, or check the slides attached at the bottom of this page.

Authors

Treeler was developed at  the TALP Research Centre of the UPC. The main authors are  Xavier Carreras,  Xavier Lluís, and  Lluís Padró, with contributions from  Andreu Mayo,  Pranava Swaroop Madhyastha, and Manuel Bertran (from  CLiC - Centre de Llenguatge i Computacio at University of Barcelona within the actions of the Araknion project).

Treeler originates from the  EGSTRA library, which was originally written by Xavier Carreras and Terry Koo at  MIT/CSAIL during 2006-2009. We thank Michael Collins and Amir Globerson for their contribution and support to the development of EGSTRA.

Getting Treeler

To get treeler, download it directly from our svn repository:

svn co http://devel.cpl.upc.edu/treeler/svn/trunk

The  FreeLing project now includes statistical dependency parsers and semantic role labeling methods for English, Spanish and Catalan, using Treeler components. To simply use such methods, we strongly recommend to use  FreeLing.

Acknowledgements

Pascal2 logo (supporting institution in 2011)

Treeler obtained financial support from the  Pascal2 Harvest Programme, the  XLike EU project, the  KNOW2 Spanish project (TIN2009-14715-C04-01) and the  Araknion Spanish project (FFI-2010-11474-E). We are grateful for their support.

Attachments