id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_oevkxmdnrjfp5akoxbszu576km	Keisuke Sakaguchi	Reassessing the Goals of Grammatical Error Correction: Fluency Instead of Grammaticality	2016	14	.pdf	application/pdf	8760	788	61	If the goal is to correct verb errors, the grammatical mistake in the original sentence has been addressed and we can move on. evaluation metrics computed over error-annotated We explore different methods for corpus annotation (with and without error codes, written by experts and non-experts) Fundamentally, this work reframes grammatical error correction as a fluency task. Table 2: An example sentence with expert and non-expert fluency edits. grammatical (following the NUCLE annotation instructions but without error coding), and two sets of Table 3: An example sentence with the original NUCLE correction and fluency and minimal edits written by experts Figure 2: Amount of changes made by different annotation sets compared to the original sentences. machine-translation metric BLEU because evaluating against our new non-coded annotations is similar Figure 3: Correlation of the human ranking with metric scores over different reference sets (Spearman's ρ). Human evaluation of grammatical error correction systems.	./cache/work_oevkxmdnrjfp5akoxbszu576km.pdf	./txt/work_oevkxmdnrjfp5akoxbszu576km.txt