Source PaperYearLineSentence
P14-2064 2014 102
Our results align with the literature suggesting that difficult cases in training data can be disruptive (Beigman and Beigman Klebanov, 2009; Schwartz et al, 2011; Rehbein and Ruppenhofer, 2011; Reidsma andCarletta, 2008); yet we also show that investigat ing their impact on the learning outcomes in some detail can provide insight about the task at hand.The main contribution of this paper is there fore in providing additional empirical evidence insupport of the argument put forward in the literature regarding the need to pay attention to prob lematic, disagreeable instances in annotated data ? both from the linguistic and from the machine learning perspectives
P14-2064 2014 14
(2011) showed that judgments of readability of the same texts by different groups of experts are sufficiently systematically different to hampercross-expert generalization of readability classi fiers trained on annotations from different groups.Rehbein and Ruppenhofer (2011) discuss the ne gative impact of systematic simulated annotation inconsistencies on active learning performance on a word-sense disambiguation task.In this paper, we address the task of classify ing words in a text as semantically new or old