Aspectual Classifications: Use of Raters' Associations and Co-occurrences of Verbs for Aspectual Classification in German

Michael Richter, Jürgen Hermes, Claes Neuefeind

Abstract: The present study examines the results of experiments on the automatic classification of German verbs into five aspectual classes [1]: An experiment within an unsupervised framework based on associations of raters [1] and a couple of experiments within a distributional framework, i.e. in window-based and in a subcategorization-frame-based approach [2]. We compare the predictive power of raters’ associations against two types of verbal cooccurrences: i. pure, unstructured co-occurrences and ii. linguistically motivated, well defined co-occurrences which we denote as informed distributional framework. We observed substantial (unsupervised) and excellent (supervised) agreements with a Gold Standard classification.