作者: Nina Dethlefs , Simon Keizer , Xingkun Liu , Helen Hastie , Heriberto Cuayahuitl
DOI:
关键词:
摘要: In the past 10 years, only around 15% of published conference papers include some kind extrinsic evaluation an NLG component in end-to-end system. These types evaluations are costly to set-up and run, so is it worth it? Is there anything be gained over above intrinsic quality measures obtained off-line experiments? this paper, we describe a case study evaluating two variants surface realiser show that significant differences both measures. would need factored into future iterations therefore, conclude worthwhile.