GOLD and Discourse: Domain- and Community-Specific Extensions

Corpora annotated for discourse–related phenomena have become an important source for the empirical study of whole texts and a source of training data for the automated parsing of whole texts. By discourse–related phenomena, we refer to various textual relations (e.g., anaphoric and rhetorical relations) that hold among textual units (e.g., whole texts, adjacency pairs, and discourse segments). One of the main problems associated with corpus linguistics—and for the markup of linguistic data in general—has been the lack of interoperability between disparately marked up resources. While it is theoretically possible to standardize the content and structure of markup, the richness of the data itself suggests that no one standard scheme would suffice for all data, all of the time. In fact one of the major contributions of the E–MELD project1 has been to show that annotation elements need not be standardized as such. Rather, by following certain parameters of “best–practice” (Bird & Simons 2003) the stage can be set for wide–spread interoperability among disparate corpora. For example, one of the key strategies of the E–MELD project has been to suggest that all markup elements used in annotating linguistic data (including discourse–related corpora) should be mapped to a semantic resource that defines the meaning of each element. Such a semantic resource for descriptive discourse categories does not exist at present. To rectify this situation, we propose a discourse–specific extension to the General Ontology for Linguistic Description (GOLD), as introduced by Farrar & Langendoen (2003) and explicated in Farrar (forthcoming).