German Semantic Verb Classes - 
Theoretical, Computational and Cognitive Approaches

Classifications have always achieved wide attention in linguistic
research. From a theoretic point of view, they explore the relevant
set of objects and their features; from a computational point of view,
they generalise over objects and therefore support Natural Language
Processing algorithms in order to overcome the lexical bottleneck.

The focus of this talk is on German semantic verb classes -
classifications of German verbs with respect to their semantic
properties. On the one hand, the presentation focuses on the variety
of theoretical classifications: which are the motivations, which are
the goals, what distinguishes multiple verb senses, and which verb
features are relevant? On the other hand, the talk describes the
automatic acquisition of a computational classification: A statistical
grammar model serves as source for lexical descriptions of German
verbs, and the empirical distributions are used as the basis for
cluster analysis. The results demonstrate the potential and the limits
of the learning methodology, referring to both linguistic and
technical aspects.

The final part of the talk discusses the relation between the
theoretical and computational approaches: Since existing
classifications strongly differ, the comparison of the classes
provides insight into relevant features for clustering algorithms. In
addition, human intuition on verb associations - as obtained by a web
experiment - indicate the kinds of semantic relations associated with
a verb, and point to cognitively plausible verb properties.