Verena Henrich

Using part---whole relations for automatic deduction of compound-internal relations in GermaNet

This paper provides a deduction-based approach for automatically classifying compound-internal relations in GermaNet, the German version of the Princeton WordNet for English. More specifically, meronymic relations between simplex and compound nouns provide the necessary input to the deduction patterns that involve different types of compound-internal relations. The scope of these deductions extends to all four meronymic relations modeled in version 6.0 of GermaNet: component, member, substance, and portion. This deduction-based approach provides an effective method for automatically enriching the set of semantic relations included in GermaNet.

Extending the TüBa-D/Z Treebank with GermaNet Sense Annotation

This paper describes the manual construction of a sense-annotated corpus for German with the goal of providing a gold standard for word sense disambiguation. The underlying textual resource, the TuBa-D/Z treebank, is a German newspaper corpus already manually enriched with high-quality, manual annotations at various levels of grammar. The sense inventory used for tagging word senses is taken from GermaNet [8,9], the German counterpart of the Princeton WordNet for English [6]. With the sense annotation for a selected set of 109 words (30 nouns and 79 verbs) occurring together more than 15 500 times in the TuBa-D/Z, the treebank currently represents the largest manually sense-annotated corpus available for GermaNet.

Aligning GermaNet Senses with Wiktionary Sense Definitions

Sense definitions are a crucial component for wordnets and enhance the usability of wordnets for a wide variety of NLP applications. Many wordnets for languages other than English – including the German wordnet GermaNet – lack comprehensive coverage of such definitions. The purpose of this paper is to automatically align sense descriptions from the web-based dictionary Wiktionary to lexical units in GermaNet in order to extend GermaNet with sense descriptions. An alignment algorithm based on word overlaps is developed and different setups of the algorithm are compared. This algorithm yields as the best result an accuracy of 93.8 % and an F1-score of 84.3, which confirms the viability of the proposed method for automatically enriching GermaNet. This best result crucially involves the use of coordinated relations as a novel concept for calculating sense alignment.

Letter Sequence Labeling for Compound Splitting.

For languages such as German where compounds occur frequently and are written as single tokens, a wide variety of NLP applications benefits from recognizing and splitting compounds. As the traditional word frequency-based approach to compound splitting has several drawbacks, this paper introduces a letter sequence labeling approach, which can utilize rich word form features to build discriminative learning models that are optimized for splitting. Experiments show that the proposed method significantly outperforms state-ofthe-art compound splitters.