Evaluating Information Loss in Temporal Dependency Trees

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2148–2156 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2148 Evaluating Information Loss in Temporal Dependency Trees Mustafa Ocal & Mark A. Finlayson School of Computing and Information Sciences CASE Building, Room 362 11200 S.W. 8th Street, Miami, FL 33199 USA {mocal001, markaf}@fiu.edu Abstract Temporal Dependency Trees (TDTs) have emerged as an alternative to full temporal graphs for representing the temporal structure of texts, with a key advantage being that TDTs can be straightforwardly computed using adapted dependency parsers. Relative to temporal graphs, the tree form of TDTs naturally omits some fraction of temporal relationships, which intuitively should decrease the amount of temporal information available, potentially increasing temporal indeterminacy of the global ordering. We demonstrate a new method for quantifying this indeterminacy that relies on solving temporal constraint problems to extract timelines, and show that TDTs result in up to a 109% increase in temporal indeterminacy over their corresponding temporal graphs for the three corpora we examine. On average, the increase in indeterminacy is 32%, and we show that this increase is a result of the TDT representation eliminating on average only 2.4% of total temporal relations. This result suggests that small differences can have big effects in temporal graphs, and the use of TDTs must be balanced against their deficiencies, with tasks requiring an accurate global temporal ordering potentially calling for use of the full temporal graph. Keywords: Temporal Indeterminacy, TimeML,Temporal Information Extraction 1. Introduction Extracting a representation of the temporal information in a text is a useful yet challenging task within natural language processing. Representations of temporal information can facilitate question answering (Saquete et al., 2004), infor- mation extraction (Wu et al., 2005), and summarization (Liu et al., 2012), among many other tasks. Researchers have developed several formalisms for repre- senting temporal information expressed in text, each with specific advantages and disadvantages. Temporal repre- sentation languages that allow the representation of tempo- ral graphs, such as Allen’s temporal algebra (Allen, 1983) or the XML-based Temporal Markup Language (TimeML) (Sauri et al., 2006), are the most flexible schemes for mark- ing of events, temporal expressions, and temporal relations. Temporal graphs expressed in these languages can be used for inference or determining partial orders of events and times. Although some of these languages for temporal graphs are quite expressive (TimeML has 25 relation types) they have certain limitations. First, the generality of tem- poral graphs makes many operations on them computation- ally hard. Second, they are less than ideal for visualization purposes, as they are difficult for people to read and un- derstand. Third, they only explicitly provide partial, local orderings of events. In response to these limitations, Cheng et al. (2007) in- troduced the idea of the temporal dependency tree (TDT), a more computationally efficient representation where all events are arranged in a tree using only three temporal relation types: BEFORE, AFTER, and OVERLAPS. Both Kolomiyets et al. (2012) and Zhang …

Content truncated to 500 words