This page provides language resources and tools created by the DFKI LT RE group. You are welcome to send questions of any kind to: firstname.lastname@example.org
Please also check out the download page of our sar-graph website!
Recon is a Java-based tool for the annotation of relations among textual elements and semantic concepts.
The “Celebrity” corpus consists of 150 news articles annotated with three semantic relations of the biographic domain. The corpus is provided in two formats, a CoNLL-like format (plain-text files with tabular-separated values) and an XML-based format. Files in the XML-based format can be loaded with the Recon tool (see above).
The “CockrACE” corpus consists of 140 news articles annotated with mentions of entities and their coreference links, as well as relation mentions for the evaluation of relation extraction (RE) experiments. Three semantic relations have been annotated, each of them dealing with people's family relationships (marriages, brother/sister, parent/child).
Differences to corpus “Celebrity” include that “CockrACE” offers a full-document annotation of entities and their relations (not just an annotation of sentences with a sufficient amount of essential relation arguments) and the manual resolution of coreferences.
This annotation effort is ongoing. We are providing here a snapshot of the (more-or-less) raw annotation. The files can be loaded with the Recon tool (see above).