Automatic construction of a bilingual lexicon
Within the Twenty-One project a new statistics based method for the automatic extraction of bilingual lexicons was developed. The method inputs a bilingual corpus (that is: a text and its translation in another language) and outputs a bilingual lexicon. The method has advantages over algorithms that have been published before, because it is based on a symmetric translation model. The resulting bilingual lexicon can be used to translate in both directions between a language pair.Example results
For the purpose of Cross-Language Information Retrieval a domain specific bilingual lexicon was automatically derived from the aligned Dutch, English, French and German version of Agenda 21. Below you find some prelimanary results.- English to Dutch
- Dutch to English
- English to German
- German to English
- Dutch to German
- German to Dutch
- English to French
- French to English
Agenda 21
The name of the project Twenty-One refers to the United Nations conference on ecology and sustainable development in Rio de Janeiro in 1992. An important result of this conference is a document titled Agenda 21. Agenda 21 captures the application domain of the Twenty-One system and is available in all the major European languages.- Contents of Dutch corpus
- Contents of English corpus (under construction)
- Contents of French corpus (under construction)
- Contents of German corpus (under construction)
Word alignment software
Twente word alignment software version 0.9b can be down loaded from this web site.Other sites
- ARCADE (Evaluation of parallel text alignment systems)
- Parallel text processing bibliography (by Jean Véronis and Marie-Dominique Mahimon)
- Dan Melamed's home page (with tools and manually annotated Blinker corpus)
- Parallel Corpora around the World (at Uppsala University)
- Parallel Corpus Project (University of Maryland)
Last modified January 1999 by Djoerd Hiemstra