Welcome to the Alpino Treebank website. The Alpino treebank contains syntactically annotated Dutch sentences. The treebank (more than 150,000 words) includes the full cdbl (newspaper) part of the Eindhoven corpus. You will also find here a number of tools to browse and search the treebank.

Original CDROM Version

The content of the CDROM which appeared in November 2002.

Treebanks: hand-corrected

• Set of 18 sentences used in the Battle of the Parsers during the 2001 LOT Winterschool.
• Set of 425 sentences from the CGN annotation guidelines.
• Set of 330 sentences from the CGN Leuven Yellow Pages document.
• Set of 450 questions from CLEF 2003
• Set of 700 questions from CLEF 2004
• Set of 200 questions from CLEF 2005
• Set of 200 questions from CLEF 2006
• Set of 200 questions from CLEF 2007
• Set of 1000 quiz questions
• Set of 1000 sentences constructed during the development of the Alpino Grammar and Lexicon.
• Set of 1000 sentences also constructed during the development of the Alpino Grammar and Lexicon.
• Set of growing number of sentences also constructed during the development of the Alpino Grammar and Lexicon.
• Set of 7153 sentences which constitute the cdbl (Newspaper) part of the Eindhoven Corpus.
• Set of 500+ sentences from the EANS (hope nobody minds)

Treebanks: not corrected

• Set of over 4 million sentences from the AD and NRC 1994 and 1995 volumes. This is a newer version of the data that was used for our submission to the CLEF 2005 monolingual Dutch QA competition.
• All sentences from the CGN corpus, automatically annotated by Alpino. Quality of this material is not overwhelming. Note that we have also converted the manual syntactic annotations from CGN into the Alpino/D-Coi/Lassy XML format. These are available as www.let.rug.nl/~vannoord/trees/Treebank/CGN/
• Set of 30 sentences from the 18.00 hr news broadcast from May 10 2005, as used by the Adams Appel television broadcast.
• All uncorrected corpora. Unordered.

Formats

• XML. The XML format is understood by a number of utilities described below. This format is suitable for machine processing. As of December 2005, for the bigger corpora the .xml files are collected together in compressed format (dictzip). We have new tools capable of efficiently retrieving individual .xml files from these archives.
• PDF. The PDF format is for visual human inspection - note that in some cases the leaf nodes are abbreviated.
• SVG. Recently, we also provide the dependency structures in SVG format. If your browser does not understand this format, perhaps you need to install a separate SVG viewer.

Utilities

A number of scripts and programs is supplied to browse and search through the treebanks. For more recent versions and various new tools, please download the Alpino distribution from http://www.let.rug.nl/~vannoord/alp/Alpino/

• dtview Graphical display of CGN dependency structures. This script takes file name arguments (typically with a .xml extension) and displays the CGN structures found in those files. The script assumes you have Tcl/Tk installed on your machine.
• dt_search Simple wrapper around the xmlmatch program. In order to use this script, you first need to compile and install the xmlmatch program (there are installation instructions in the source directory). The script enables efficient and powerful search in CGN dependency structures. Examples are provided in a separate readme file.
• dtv Simple wrapper to combine the dt_search and dtview programs. The script takes a query and a set of file names. The files which match the given query are then displayed by means of dtview (the part(s) of the tree that matched the query will be highlighted).
• Thistle is an editor for linguistic datastructures by Jo Calder. We have used Thistle for editing CGN dependency structures. The files without an extension are in the SGML format understood by Thistle. The file cgn_dt.spec contains the specification of our implementation of the CGN structures - it is required by Thistle. In order to use thistle you need to install the wrapper scripts (the program is in Java and already compiled). Please cd to directory thistle-2-0-1 for further instructions. Note that this version of Thistle contains a number of small improvements and bug fixes based on the thistle-2-0-alpha release. See the Thistle homepage.

Documentation

An attempt to describe a number of differences between the CGN annotation practice and ours is given in this document, which is heavily out of date. The good news is that the number of differences has been reduced heavily recently.

Publications

• Robert Malouf, Gertjan van Noord. Wide Coverage Parsing with Stochastic Attribute Value Grammars. In: IJCNLP-04 Workshop Beyond Shallow Analyses - Formalisms and statistical modeling for deep analyses. [pdf,web page]
• Chapter 5. The Alpino Dependency Treebank. In: Leonoor van der Beek, Gosse Bouma, Jan Daciuk, Tanja Gaustad, Robert Malouf, Gertjan van Noord, Robbert Prins, Begoña Villada, Algorithms for Linguistic Processing NWO PIONIER Progress Report. Groningen 2002. [pdf]
• Leonoor van der Beek, Gosse Bouma, Robert Malouf, Gertjan van Noord. The Alpino Dependency Treebank. In: Computational Linguistics in the Netherlands CLIN 2001. Rodopi 2002. [pdf]
• Leonoor van der Beek, Gosse Bouma, and Gertjan van Noord. Een brede computationele grammatica voor het Nederlands. Nederlandse Taalkunde, 2002. [pdf]
• Gosse Bouma and Geert Kloosterman. Querying dependency treebanks in XML. In Proceedings of the Third international conference on Language Resources and Evaluation (LREC), Gran Canaria, 2002. [pdf]
• Gosse Bouma, Gertjan van Noord, Robert Malouf. Alpino: Wide Coverage Computational Analysis of Dutch. In: Computational Linguistics in the Netherlands CLIN 2000. Rodopi 2001. [pdf]

Alpino Demo

In the context of the treebanking efforts, we are constructing a natural language understanding system for Dutch: Alpino. This ever-growing system is built on top of Hdrug; it contains a wide-coverage HPSG for Dutch, a large-scale lexicon, a parser, a disambiguation component using a log-linear (maximum entropy) model, etc. There is an experimental web-demo.

Who to blame

• Leonoor van der Beek (annotation)
• Gosse Bouma (annotation)
• Jan Daciuk (tools)
• Geert Kloosterman (tools)
• Robert Malouf (tools)
• Gertjan van Noord (annotation, tools)
• Robbert Prins (art work, tools)

Further information: Algorithms for Linguistic Processing Homepage.

Feedback

Based on the number of errors that we have found ourselves during the last few months, it is certain that there are still many errors in the treebank. We appreciate your feedback if you find errors. Please send a polite email to: vannoord@let.rug.nl

Hugo Brandt Corstius was keynote speaker at the 13th CLIN meeting (29 November 2002 in Groningen). After his presentation, the first cdrom was officially handed to him.