关于概率神经摘要中的不确定性校准和选择性生成：一项基准研究（英文版）

英文标题：On Uncertainty Calibration and Selective Generation in Probabilistic Neural Summarization: A Benchmark Study中文摘要：我们研究了深度学习方法在文本摘要中存在的局限性，提出使用概率深度学习方法来提高模型生成与不确定性质量， 并在实践中采用适当的方法来避免摘要生成的低质量问题。英文摘要：Modern deep models for summarization attains impressive benchmarkperformance, but they are prone to generating miscalibrated predictiveuncertainty. This means that they assign high confidence to low-qualitypredictions, leading to compromised reliability and trustworthiness inreal-world application