REG in Context

Referring Expression Generation (REG) in Context maps a representation of an intended referent given within a full textual context to a fully realised surface string that is appropriate for the given context.  The representation of the intended referent may be a list of possible referring expressions (RE) for the referent, and/or a set of semantic and syntactic properties.

In contrast to REG from ASs, a list of distractors is not provided. REG in Context systems typically start by analysing the surrounding text in order to extract information that will help decide what kind of RE to output for the intended referent. Because REs are generated within a discourse context, other REs in the text for the same and other referents will have to be taken into account, so the task could be considered to be the generation of chains of REs, rather than individual REs.

REG in Context resources on genchal-repository

• 1. Legacy resources from the GREC Shared Tasks (2008-2010):

     1. Selection of REs for references to the main subject of Wikipedia articles: GREC-MSR

     2. Regeneration of REs for references to people entities in Wikipedia articles about people:

        1. Recognising all references to people in raw text (named entity recognition task): GREC-NER

        2. Selection of REs or references to people entities: GREC-NEG

        3. Regeneration of REs for references to people entities in raw text: GREC-Full