The paper describes the implementation and evaluation of an automatic neologism detection prototype, the "Word trawler". In two different experiments involving newspaper texts three detection techniques are tested, namely primitive filtering, statistical "weirdness" by comparison with a reference corpus predating the analysis corpus, and neology markers. It is found that a combination of these techniques results in the highest precision (approximately 40%). However, neology markers drastically reduce recall and should only be used when ample data is available. The authors finally suggest that diachronic frequency profiling could be used to further reduce system "noise", such as occasional isms and spelling errors.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados