NAME
Treex - Natural Language Processing framework
VERSION
version 0.07190
INTRODUCTION
Treex (formerly named TectoMT) is a highly modular, multi-purpose, multi-lingual, easily extendable Natural Language Processing framework.
Treex has the following features:
- There is a number of NLP tools already integrated in Treex, such as morphological tagger, lemmatizers, named entity recognizers, dependency parsers, constituency parsers, various kinds of dictionaries.
- Treex allows storing all data in an XML-based format, which simplifies data interchange with other frameworks.
- Treex is tightly coupled with the tree editor TrEd, which allows easy visualization of syntactic structures.
- Treex is language universal and supports processing multilingual parallel data.
- Treex facilitates distributed processing on a computer cluster.
- Treex architecture is inspired by the Prague Dependency Treebank, especially by its layered view on language (especially distinguishing surface syntax and deep syntax). Linguistic data structures used in Treex are highly similar to the PDT's data structures.
- Treex has been used for deep analysing large data, such as for Czech-English parallel treebank CzEng (millions of parallel sentence pairs).
- Treex has been intensively used for several years for developing a Czech-English machine translation system, which is currently the main, but not the only one application of Treex.
In a way, Treex is similar to GATE. However, in our opinion, Treex has a better support for deeply structured language data and for processing multilingual data.
COMPONENTS
Treex is divided into several distributions:
Treex::PML distribution
PML stands for Prague Markup Language (PML), which is an XML-based data format developed for interchange of linguistic data. Treex::PML comprises of a set of modules defining abstract Perl types (such as tree nodes) and related functionality (such as tree traversal), as well as procedures for storing the data structures into (and loading from) PML files. Treex::PML is a universal format, with only a few assumptions about linguistic data.
Treex::PML was developed by Petr Pajas, originally under the name Fslib as a part of the tree editor Tred, long before other components of TectoMT/Treex were created.
Treex::Core distribution
Treex::Core is an additional level of functionality added to Treex::PML. Most Treex::Core classes are descendants of Treex::PML classes.
Unlike Treex::PML, Treex::Core is not meant to be a universal library for linguistic data. Treex::Core predefines several quite specific types of linguistic data structures; this limitation allows Treex::Core to provide functionality designed specifically for these structures (such as resolving coordination, links between deep and surface syntax, clause segmentation, alignment of parallel data, etc.). Moreover, Treex::Core offers tools for distributed processing of Treex files, and for their visualization in TrEd. Like Treex::PML, Treex::Core is language universal.
Language specific distributions
Language specific components of Treex are planned to be released on CPAN within 2011.
HISTORY
Treex results from an active long-term NLP research carried out at the Institute of Formal and Applied Linguistics (Faculty of Mathematics and Physics, Charles University in Prague, Czech Republic). The following list surveys only selected technological shifts:
1996 - Michal Křen develops the Graph tree editor. Linguistic data structures are stored in a proprietary data format called fs-format.
1997 - an SGML-based format called CSTS (Czech sentence tree structure) is developed. In 2001, CSTS becomes the primary data format of the Prague Dependency Treebank 1.0.
2000 - Petr Pajas develops an alternative tree editor called TrEd (in Perl), which soon becomes the main tool used for various types of annotations realized in the department, and is used abroad as well. TrEd contained a library of modules for memory representation of linguistic structures and their storage, called Fslib.
2000-2005 - Numerous NLP modules were developed at the department, aimed either at exploiting the existing PDT data, or at facilitating new linguistic annotations. A new generic (language-universal, theory-universal) XML-based format called Prague Markup Language is developed. PML was the primary format of Jan Hajič et al.'s PDT 2.0 in 2006.
2005 - Zdeněk Žabokrtský starts the development of TectoMT, with the aim of creating a modular NLP system, primarily focused on Machine Translation with deep-syntactic (tectogrammatical) transfer.
2006 - First TectoMT-based English-Czech Machine Translation experiments are performed.
2006-2010 - TectoMT's architecture, as well as the translation quality, are gradually improved (several thousand svn revisions). Many NLP tools such as taggers and parsers for various languages are integrated.
2010 - Fslib separated from TrEd, and published as a CPAN module under the name Treex::PML.
2010-2011 - "TectoMT" is rebranded to "Treex" - "eXploit syntactic/semantic TREEs!". All core components are refactored using Moose.
2011 - The Treex::Core components are packed and go to CPAN. Some other (language specific) distributions are hoped to appear soon too.
REFERENCES
Selected TectoMT/Treex-related publications:
- Žabokrtský Zdeněk, Ptáček Jan, Pajas Petr: TectoMT: Highly Modular MT System with Tectogrammatics Used as Transfer Layer. In: ACL 2008 WMT: Proceedings of the Third Workshop on Statistical Machine Translation, Copyright © Association for Computational Linguistics, Columbus, OH, USA, ISBN 978-1-932432-09-1, pp. 167-170, 2008
- Žabokrtský Zdeněk, Bojar Ondřej: TectoMT, Developer's Guide. Technical report no. 2008/TR-2008-39, Copyright © Institute of Formal and Applied Linguistics, Faculty of Mathematics and Physics, Charles University in Prague, 50 pp., Dec 2008
- Popel Martin, Žabokrtský Zdeněk: TectoMT: Modular NLP Framework. In: Lecture Notes in Computer Science, Vol. 6233, Proceedings of the 7th International Conference on Advances in Natural Language Processing (IceTAL 2010), Copyright © Springer, Berlin / Heidelberg, ISBN 978-3-642-14769-2, ISSN 0302-9743, pp. 293-304, 2010
- Mareček David, Popel Martin, Žabokrtský Zdeněk: Maximum Entropy Translation Model in Dependency-Based MT Framework. In: Proceedings of the Joint Fifth Workshop on Statistical Machine Translation and MetricsMATR, Copyright © Association for Computational Linguistics, Uppsala, Sweden, ISBN 978-1-932432-71-8, pp. 201-201, 2010
AUTHOR
Treex is an open project, there is a number of people who have contributed.
Treex Cabal (not very original, is it?)
Those who are responsible for maintaining Treex::Core modules.
Zdeněk Žabokrtský <zabokrtsky@ufal.mff.cuni.cz>
Martin Popel <popel@ufal.mff.cuni.cz>
David Mareček <marecek@ufal.mff.cuni.cz>
Author of TrEd and Treex::PML
Petr Pajas
Some other contributors to Treex, both present and past, alphabetically
Ondřej Bojar
Václav Klimeš
Tomáš Kraut
Václav Novák
Jan Ptáček
Rudolf Rosa
Daniel Zeman
COPYRIGHT AND LICENSE
Copyright © 2011 by Institute of Formal and Applied Linguistics, Charles University in Prague
This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself.