Clustering of Motion Related Verbs based on Human Subjective Judgement of Similarity

Abstract

The process of clustering a given set of objects involves identifying hidden relations and underlying indicative patterns within data, and utilizing this information as a source of comparison for the objects, organizing them into meaningful groups. This thesis attempts to identify such of relations and patterns in order to cluster a set of Norwegian motion related verbs. The clustering process employed by this thesis utilizes data provided by a set of native Norwegian speakers, representing the ways in which each subject clustered the verbs, based on subjective measure of similarity. This information is employed by both a conventional clustering method, as well as an adaptation of a feature-drive method. As a source of similarity between the verbs, the criteria by which the human subjects discriminated the verbs, as well as the importance put on these criteria, are attempted identified and utilized as a part of the clustering process. Similarities between the verbs are also attempted visualized using multidimensional scaling. The clustering approached is evaluated by assessing how well the produced clusters represent the similarity between the verbs, based on the data provided by the human subjects.