id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
cord-028436-ahmpe981	Azad, Sushmita	Strategies for Deploying Unreliable AI Graders in High-Transparency High-Stakes Exams	2020-06-09		.txt	text/plain	3937	229	60	deployed in a low-stakes homework context for which we had two goals: 1) we wanted students to improve their ability to provide natural language descriptions of code, so we provided both immediate correct/incorrect feedback and example correct answers as shown in Fig. 1(B) and 2) we wanted to collect additional training data which could be used to train improved NLP-based AI graders. Second, we could provide students an appeal system where they could, after they are shown the correct answer, request a manual re-grade for an EiPE question, if they believed the AI grader had scored them incorrectly. 5. Students' perception of the grading accuracy of our NLP-based AI grader was lower than that of deterministically-correct auto-graders for true/false, multiple-choice, and programming questions, but only to a modest degree.	./cache/cord-028436-ahmpe981.txt	./txt/cord-028436-ahmpe981.txt