Sense-Tagging at the Cycle-Level Using GLDB

Resumé

This report describes a large-scale attempt to identify automatically the appropriate sense for content words taken from Swedish open-source texts. Sense-tagging, 'the process of assigning the appropriate sense from some kind of lexicon to the (content) words in a text', is a difficult and demanding task in Natural Language Processing and researchers have been engaged in finding a suitable solution to the problem for a very long time. The usefulness of automatically assigning each word in unrestricted text with its most likely sense is necessary for a great spectrum of applications. The sense-tagger described here has been tested both on a random sample of content words, as well as on a large population of a single ambiguous entry. In the first case, the achieved precision was 84,21 %, and in the second 82,75% respectively. Evaluation was made against manually sense-annotated texts.