Request Correction

Handling Subtle Sense Distinctions Through Wordnet Semantic Types

In this paper we challenge the question of whether there is value in having multiple layers of semantic information associated with corpus semantic annotation. In this context we introduce a semantic annotation experiment in which novice annotators were asked to assign sense tags to a set of polysemous corpus nouns, using Wordnet as their referential sense repository. Wordnet is a rich sense inventory that provides explicit information of the semantic types associated with every word sense. To measure the effect semantic types’ knowledge has on the sense assignment process, we carried out two annotation sessions. In the first session, annotators relied exclusively on Wordnet synsets to annotate corpus nouns, whereas in the second session the same pool of annotators examined Wordnet synsets in conjunction with their semantic types, prior assigning a sense tag. Comparing annotators’ performance in both sessions shows that when consulting semantic types, annotators assigned more salient senses to highly polysemous nouns, whereas for the same set of terms, when relying exclusively on Wordnet synsets, annotators tended to assign narrower senses, which whatsoever were more error-prone. Results indicate that semantic types have a potential in dealing with subtle sense distinctions in the course of corpus annotation.