Semantic Annotation of Anaphoric Links in Language

This paper attempts to integrate several existing coreference annotation schemes into an extended annotation scheme AS(ana). The proposed AS(ana) allows some other types of the anaphor-antecedent relation, called 'anaphoric link', than the canonical type of coreference that implies the referential identity between an anaphor and its antecedent. The structure of AS(ana) itself is very simple, consisting of a single entity type for mentions and a single anaphoric relation, each of which is characterized by a small set of attribute-value specifications. Constrained by these specifications, AS(ana) supports a two-step annotation procedure: For a given text T, (1) identification of a set of mentions M in the text T that refer to something in the universe of discourse referents as its markables, (2a) identification of a set of pairs of the mentions in M that are anaphorically related, and (2b) specification of the type of such a relation.