Lemma selection in domain specific computational lexica - some specific problems
Proceedings of the Third International Conference on Language Resources and Evaluation (LREC 2002)
Abstract
This paper describes the lemma selection process of a Danish computational lexicon, the STO project, for domain specific language and focuses on some specific problems encountered during the lemma selection process. After a short introduction to the STO project and an explanation of why the lemmas are selected from a corpus and not chosen from existing dictionaries, the lemma selection process for domain specific language is described in detail. The purpose is to make the lemma selection process as automatic as possible but a manual examination of the final candidate lemma lists is inevitable. The lemmas found in the corpora are compared to a list of lemmas of general language, sorting out lemmas already encoded in the database. Words that have already been encoded as general language words but that are also found with another meaning and perhaps another syntactic behaviour in a specific domain should be kept on a list and the paper describes how this is done. The recognition of borrowed words the spelling of which have not been established constitutes a big problem to the automatic lemma selection process. The paper gives some examples of this problem and describes how the STO project tries to solve it.