2 Hits in 6.5 sec

CaPE: Contrastive Parameter Ensembling for Reducing Hallucination in Abstractive Summarization [article]

Prafulla Kumar Choubey, Alexander R. Fabbri, Jesse Vig, Chien-Sheng Wu, Wenhao Liu, Nazneen Fatema Rajani
2022 arXiv   pre-print
Hallucination is a known issue for neural abstractive summarization models. Recent work suggests that the degree of hallucination may depend on errors in the training data.  ...  The improvement in factual performance does not degrade the performance on other metrics of informativeness such as ROUGE.  ...  Conclusion We present Contrastive Parameter Ensembling (CaPE) to reduce content hallucinations in abstractive summarization models.  ... 
arXiv:2110.07166v2 fatcat:orq6yjorjnh3fjgmgavfgcmoje


1934 The American Journal of Cancer  
The animals were nephrotomized and a mixture of two parts of tar and one of cement placed in the pelvis of one kidney.  ...  These reports have been summarized in two excellent collective reviews of the recent literature of tar cancer, published independently, one by Watson (Cancer Review 7: 445, 1932) and the second, and more  ...  The conjunotival lesions occur chiefly in the palpebral folds, but may be mofe widespread.  ... 
doi:10.1158/ajc.1934.141 fatcat:7zjhv5ig6zgolnri3rme4t4ddq