Grammatical Role Embeddings for Enhancements of Relation Density in the Princeton WordNet

In this paper we present an approach to train subatom embeddings for verbs. For each verb we learn not just one embedding, but several. One for the verb itself and embeddings for each grammatical role of this verb. For example, for the verb ‘to give’ we learn four embeddings: one for the lemma ‘give’, one for the subject, one for the direct object and one for the indirect object of it. We are exploiting these grammatical role embeddings in order to add new syntagmatic relations to WordNet. The evaluation of the quality of the new relations is done extrinsically via Knowledge-based Word Sense Disambiguation.