Search

This paper presents an ongoing work on the multilingual event-type ontology SynSemClass, where multilingual verbal synonymy is formalized in terms of syntactic and semantic properties. In the ontology, verbs are grouped into synonym classes, both monolingually and cross-lingually. Specifically, verbs are considered to belong to the same class if they both express the same meaning in a specific context, and their valency frame can be mapped to the set of roles defined for a particular class. SynSemClass is built following a bottom-up approach where translational equivalents are automatically extracted from parallel corpora and annotated by human annotators. The task of the annotators consists in mapping the valency frame of a particular verb with the set of roles defined for the class where the verb is included as a potential class member, establishing links to external resources, and selecting relevant examples. The Spanish part of the ontology currently contains 257 classes enriched with Spanish synonyms. The resulting resource provides finegrained syntactic and semantic information on multilingual verbal synonyms and links to other existing monolingual and multilingual resources.

This article presents a study which utilized English training data to classify Czech documents in cross-language text classification. The dataset is a collection of manually transcribed, spontaneous, conversational speech in English and Czech. Indexing proceeds on the English documents by first checking if the term is already present in the probabilistic dictionary. If it is, the term's frequency is incremented. Precision was calculated over the five and ten highest ranked thesaurus labels as well as the five highest concept labels alone.