A Meta-framework for Spatiotemporal Quantity Extraction from Text

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2736 - 2749 May 22-27, 2022 c⃝2022 Association for Computational Linguistics A Meta-framework for Spatiotemporal Quantity Extraction from Text Qiang Ning∗ Amazon qning@amazon.com Ben Zhou University of Pennsylvania xyzhou@seas.upenn.edu Hao Wu Hooray Data haowu@hooray.ai Haoruo Peng Newsbreak haoruo.peng@newsbreak.com Chuchu Fan MIT chuchu@mit.edu Matt Gardner∗ Microsoft Semantic Machines mattgardner@microsoft.com Abstract News events are often associated with quanti- ties (e.g., the number of COVID-19 patients or the number of arrests in a protest), and it is often important to extract their type, time, and location from unstructured text in order to analyze these quantity events. This pa- per thus formulates the NLP problem of spa- tiotemporal quantity extraction, and proposes the first meta-framework for solving it. This meta-framework contains a formalism that de- composes the problem into several informa- tion extraction tasks, a shareable crowdsourc- ing pipeline, and transformer-based baseline models. We demonstrate the meta-framework in three domains—the COVID-19 pandemic, Black Lives Matter protests, and 2020 Cali- fornia wildfires—to show that the formalism is general and extensible, the crowdsourcing pipeline facilitates fast and high-quality data annotation, and the baseline system can handle spatiotemporal quantity extraction well enough to be practically useful. We release all re- sources for future research on this topic.1 1 Introduction Events are often associated with quantities – how many COVID-19 patients are on ventilators, how many people are injured during protests, or how large is the extent of a wildfire. We often need to figure out the type of an event, and where and when it happened for these quantities for coher- ent discussion of public policy on sociopolitical events in rapidly evolving situations: “19 deaths” is different from “19 recoveries;” “19 deaths in a small city yesterday” apparently describes a more severe situation than “19 deaths in the whole coun- try last month.” However, until dedicated channels are established, these quantities are typically first reported on social media and local news articles, which then have to slowly make their way to some ∗Work started while at the Allen Institute for AI 1https://github.com/steqe DCT: Thursday, 08/27/2020 Title: Study Sessions, Dinners: 104 New USC Student Coronavirus Cases Text: LOS ANGELES , CA -- The number of coronavirus cases confirmed among USC students continued rising Thursday, with the university announcing [104] new cases over the past four days… Recognition: 104 Type: Confirmed cases Spatial Grounding: US à California à Los Angeles à USC Temporal Grounding: [08/23/2020, 08/26/2020] DCT: Monday, 06/01/2020 Title: Black Lives Matter: 16 Organizations That Are Bailing Out Protestors Text: …Police officers have arrested [thousands] of demonstrators… Recognition: thousands Type: Arrests Spatial Grounding: US Temporal Grounding: Overall quantity ending on 06/01/2020 Figure 1: Given document creation time (DCT), title, and text, the STEQE problem is to do quantity recogni- tion, typing, spatial grounding, and temporal grounding according to the proposed formalism (Sec. 2). Above are two examples from our COVID-19 dataset and BLM protest dataset. aggregate location for decision-makers to use. …

Content truncated to 500 words