id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_2qq6wxmizbfornkrysxkp6vpqu	Egoitz Laparra	From Characters to Time Intervals: New Paradigms for Evaluation and Neural Parsing of Time Normalizations	2018	14	.pdf	application/pdf	9069	825	66	times from different annotation schemes by measuring overlap of intervals on the timeline. Figure 1: Annotation of the expression Saturdays since March 6 following the SCATE schema. (2004), Bethard and Parker (2016) proposed Semantically Compositional Annotation of Time Expressions (SCATE). propose a new evaluation metric to compare time normalizations annotated in both the ISO 8601 format of We apply the interval-based evaluation metric introduced in Section 3 to the AQUAINT and TimeBank datasets, treating the TimeML annotations as the SCATE annotations cover different time intervals that TimeML has a recall of only 92% of the time intervals identified by SCATE in the AQUAINT corpus The SCATE annotation represents the full expression and, consequently, produces the correct time interval [1986-01Unsurprisingly, TimeML has a lower recall of the time intervals from the SCATE annotations Our 2-Softmax model (Figure 3) splits the output space of labels into two sets: non-operators and	./cache/work_2qq6wxmizbfornkrysxkp6vpqu.pdf	./txt/work_2qq6wxmizbfornkrysxkp6vpqu.txt