Finding common ground: towards a surface realisation shared task

Abstract

In many areas of NLP reuse of utility tools such as parsers and POS taggers is now common, but this is still rare in NLG. The subfield of surface realisation has perhaps come closest, but at present we still lack a basis on which different surface realisers could be compared, chiefly because of the wide variety of different input representations used by different realisers. This paper outlines an idea for a shared task in surface realisation, where inputs are provided in a common-ground representation formalism which participants map to the types of input required by their system. These inputs are derived from existing annotated corpora developed for language analysis (parsing etc.). Outputs (realisations) are evaluated by automatic comparison against the human-authored text in the corpora as well as by human assessors.