arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2101.09079cs.CL
收藏

评价离散发现:一个句子压缩的案例研究

Evaluation Discrepancy Discovery: A Sentence Compression Case-study

Bloomberg L.P.(彭博有限合伙企业);Ubiquitous Knowledge Processing Lab (UKP Lab)(普适知识处理实验室(UKP实验室));Technical University of Darmstadt(达姆施塔特工业大学)

Yevgeniy Puzikov

摘要:

可靠的评估协议对于可重复的NLP研究至关重要。在这项工作中,我们表明,有时既不是度量,也不是传统的人的评价是不够的,得出有关系统性能的结论。使用句子压缩作为示例任务,我们演示了系统如何与完善的数据集进行博弈,以获得最先进的结果。与以前的工作中报告的结果相比,显示了人类判断和度量分数之间的相关性,我们对最先进的系统输出的手动分析表明,高度量分数可能只表明更好地拟合数据,而不是更好的输出,正如人类所感知的那样。

英文摘要:

Reliable evaluation protocols are of utmost importance for reproducible NLP research. In this work, we show that sometimes neither metric nor conventional human evaluation is sufficient to draw conclusions about system performance. Using sentence compression as an example task, we demonstrate how a system can game a well-established dataset to achieve state-of-the-art results. In contrast with the results reported in previous work that showed correlation between human judgements and metric scores, our manual analysis of state-of-the-art system outputs demonstrates that high metric scores may only indicate a better fit to the data, but not better outputs, as perceived by humans.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑