A Database of Attribution Relations

AbstractThe importance of attribution is becoming evident due to its relevance in particular for Opinion Analysis and Information Extraction applications. Attribution would allow to identify different perspectives on a given topic or retrieve the statements of a specific source of interest, but also to select more relevant and reliable information. However, the scarce and partial resources available to date to conduct attribution studies have determined that only a portion of attribution structures has been identified and addressed. This paper presents the collection and further annotation of a database of over 9800 attributions relations from the Penn Discourse TreeBank (PDTB). The aim is to build a large and complete resource that fills a key gap in the field and enables the training and testing of robust attribution extraction systems.