What Are You Trying to Do? Semantic Typing of Event Processes

Proceedings of the 24th Conference on Computational Natural Language Learning, pages 531–542 Online, November 19-20, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 531 “What Are You Trying to Do?” Semantic Typing of Event Processes Muhao Chen1,2, Hongming Zhang3∗, Haoyu Wang1, Dan Roth1 1Department of Computer and Information Science, UPenn 2Information Sciences Institute, USC 3Department of Computer Science and Engineering, HKUST muhaoche@usc.edu; hzhangal@cse.ust.hk; {why16gzi, danroth}@seas.upenn.edu Abstract This paper studies a new cognitively motivated semantic typing task, multi-axis event process typing, that, given an event process, attempts to infer free-form type labels describing (i) the type of action made by the process and (ii) the type of object the process seeks to affect. This task is inspired by computational and cogni- tive studies of event understanding, which sug- gest that understanding processes of events is often directed by recognizing the goals, plans or intentions of the protagonist(s). We develop a large dataset containing over 60k event pro- cesses, featuring ultra fine-grained typing on both the action and object type axes with very large (103 ∼ 104) label vocabularies. We then propose a hybrid learning framework, P2GT, which addresses the challenging typing prob- lem with indirect supervision from glosses1 and a joint learning-to-rank framework. As our experiments indicate, P2GT supports identify- ing the intent of processes, as well as the fine semantic type of the affected object. It also demonstrates the capability of handling few- shot cases, and strong generalizability on out- of-domain processes.2 1 Introduction Events are the fundamental building blocks of natu- ral languages. To help machines understand events, extensive research effort has been devoted to in- ducing how events described in text are procedu- rally connected (Ning et al., 2017; Radinsky et al., 2012), and how they form event processes3 (Pi- chotta and Mooney, 2014; Berant et al., 2014; Jin- dal and Roth, 2013). Consequently, such proto- typical schematic sequences of events have found ∗ This work was done when the author was visiting the University of Pennsylvania. 1A gloss provides a sense definition for a lexeme. 2The contributed learning resources, software and a system demonstration are available at http://cogcomp.org/ page/publication_view/915. 3A.k.a. event chains (Chambers and Jurafsky, 2008). important use cases including storyline construc- tion (Do et al., 2012; Radinsky and Horvitz, 2013), narrative cloze (Chaturvedi et al., 2017; Lee and Goldwasser, 2019), biological process comprehen- sion (Berant et al., 2014) and diagnostic prediction (Zhang et al., 2020b). Nonetheless, understanding an event process is not just about inducing temporal relations between events or inferring missing steps in an event se- quence. As suggested by cognitive studies (Zacks et al., 2001; Zacks and Tversky, 2001; Kurby and Zacks, 2008), a process of events is defined more by the goals, plans, intentions, or traits of its per- former, rather than by physical characteristics. For example, a series of events digging a hole, putting in some seeds, filling with soil and watering the soil, occurs in a specific sequence since these steps are directed towards the central goal of planting a plant by …

Content truncated to 500 words