Faithful Question Answering with Monte-Carlo Planning

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 351–363 July 9-14, 2023 ©2023 Association for Computational Linguistics SPEECH: Structured Prediction with Energy-Based Event-Centric Hyperspheres Shumin Deng♥, Shengyu Mao♠, Ningyu Zhang♠∗, Bryan Hooi♥∗ ♥National University of Singapore & NUS-NCS Joint Lab, Singapore ♠Zhejiang University & AZFT Joint Lab for Knowledge Engine, China {shumin,dcsbhk}@nus.edu.sg, {shengyu,zhangningyu}@zju.edu.cn Abstract Event-centric structured prediction involves predicting structured outputs of events. In most NLP cases, event structures are complex with manifold dependency, and it is challeng- ing to effectively represent these complicated structured events. To address these issues, we propose Structured Prediction with Energy- based Event-Centric Hyperspheres (SPEECH). SPEECH models complex dependency among event structured components with energy- based modeling, and represents event classes with simple but effective hyperspheres. Exper- iments on two unified-annotated event datasets indicate that SPEECH is predominant in event detection and event-relation extraction tasks. 1 Introduction Structured prediction (Taskar et al., 2005) is a task where the predicted outputs are complex structured components. This arises in many NLP tasks (Smith, 2011; Kreutzer et al., 2017; Wang et al., 2023) and supports various applications (Jagannatha and Yu, 2016; Kreutzer et al., 2021). In event-centric NLP tasks, there exists strong complex dependency be- tween the structured outputs, such as event detec- tion (ED) (Chen et al., 2015), event-relation extrac- tion (ERE) (Liu et al., 2020b), and event schema induction (Li et al., 2020). Thus, these tasks can also be revisited as event-centric structured predic- tion problems (Li et al., 2013). Event-centric structured prediction (ECSP) tasks require to consider manifold structures and de- pendency of events, including intra-/inter-sentence structures. For example, as seen in Figure 1, given a document containing some event mentions “David Warren shot and killed Henry Glover ... David was convicted and sentenced to 25 years and 9 months ...”, in ED task mainly considering intra-sentence structures, we need to identify event triggers (killed, convicted) from these tokens and categorize them ∗ Corresponding Author. …… [S1] Former NOPD police officer David Warren shot and killed Henry Glover. [S2] Five current and former officers of the NOPD were charged with Glover‘s death. [S3] David was convicted and sentenced to 25 years and 9 months in prison for shooting and killing Glover. ……. …… trigger …… …… …… non-trigger …… …… …… … …… …… …… … trigger …… ………… non-trigger …… …… …… …… …… … non-trigger …… …… …… trigger killing legal_rulings death ![S1], cause, [S3]! ![S2], before, [S3]! …… ED ERE Figure 1: Illustration of event-centric structured predic- tion tasks, with the examples of ED and ERE. into event classes (killing, legal_rulings); in ERE task mainly considering inter-sentence structures, we need to find the relationship between each event mention pair, such as event coreference, temporal, causal and subevent relations. As seen from Figure 1, the outputs of ECSP lie on a complex manifold and possess interdepen- dent structures, e.g., the long-range dependency of tokens, the association among triggers and event classes, and the dependency …

Content truncated to 500 words