Publikation

An Unsupervised Semantic Tagger Applied to German

Paul Buitelaar, Jan Alexandersson, Tilman Jäger, Stephan Lesch, Norbert Pfleger, Diana Raileanu, Tanja von den Berg, Kerstin Klöckner, Holger Neis, Hubert Schlarb

In: Proceedings of the 3rd Conference on Recent Advances in Natural Language Processing (RANLP'01), September 5-7. International Conference on Recent Advances in Natural Language Processing (RANLP-01) September 5-7 Tzigov Chark Bouvet Island 2001.

Abstrakt

We describe an unsupervised semantic tagger, applied to German, but which could be used with any language for which a corresponding "XNet" (WordNet, GermaNet, e tc.), POS tagger and morphological analyzer are available. Disambiguation is per formed by comparing co-occurrence weights on pairs of semantic classes (synsets from GermaNet). Precision is around 67% at a recall of around 65% (for all ambig uous words -- 81% for all words at a recall of 80%). Our results show the influe nce of context size and of semantic class frequency in the training corpus.

ranlp01.pdf (pdf, 68 KB)

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence