Automatic Detection of Language and Annotation Model Information in CoNLL Corpora

We introduce AnnoHub, an on-going effort to automatically complement existing language resources with metadata about the languages they cover and the annotation schemes (tagsets) that they apply, to provide a web interface for their curation and evaluation by means of domain experts, and to publish them as a RDF dataset and as part of the (Linguistic) Linked Open Data (LLOD) cloud. In this paper, we focus on tabular formats with tab-separated values (TSV), a de-facto standard for annotated corpora as popularized as part of the CoNLL Shared Tasks. By extension, other formats for which a converter to CoNLL and/or TSV formats does exist, can be processed analoguously. We describe our implementation and its evaluation against a sample of 93 corpora from the Universal Dependencies, v.2.3.