Prompt for Extraction? PAIE: Prompting Argument Interaction for Event Argument Extraction

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6759 - 6774 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Prompt for Extraction? PAIE: Prompting Argument Interaction for Event Argument Extraction Yubo Ma1†∗, Zehao Wang2†∗, Yixin Cao‡3 Mukai Li4†, Meiqi Chen5†, Kun Wang4, Jing Shao4 1 S-Lab, Nanyang Technological University 2 KU Leuven 3 Singapore Management University 4 SenseTime Research 5 Peking University yubo001@e.ntu.edu.sg, zehao.wang@esat.kuleuven.be Abstract In this paper, we propose an effective yet ef- ficient model PAIE for both sentence-level and document-level Event Argument Extrac- tion (EAE), which also generalizes well when there is a lack of training data. On the one hand, PAIE utilizes prompt tuning for extrac- tive objectives to take the best advantages of Pre-trained Language Models (PLMs). It intro- duces two span selectors based on the prompt to select start/end tokens among input texts for each role. On the other hand, it captures argu- ment interactions via multi-role prompts and conducts joint optimization with optimal span assignments via a bipartite matching loss. Also, with a flexible prompt design, PAIE can ex- tract multiple arguments with the same role instead of conventional heuristic threshold tun- ing. We have conducted extensive experiments on three benchmarks, including both sentence- and document-level EAE. The results present promising improvements from PAIE (3.5% and 2.3% F1 gains in average on three benchmarks, for PAIE-base and PAIE-large respectively). Further analysis demonstrates the efficiency, generalization to few-shot settings, and effec- tiveness of different extractive prompt tuning strategies. Our code is available at https: //github.com/mayubo2333/PAIE. 1 Introduction Understanding text by identifying the event and arguments has been a long-standing goal in Natural Language Processing (NLP) (Sundheim, 1992). As shown in Fig. 1, we can quickly understand that the document is talking about a Sell event, with four in- volved arguments, i.e., Vivendi (Seller), Universal Studios (Artifact), parks (Artifact), and company (Artifact), where the argument roles are in brackets. Since event detection has achieved great success in *Equal Contribution. †Work was done when Yubo, Zehao, Mukai and Meiqi were intern researchers at SenseTime Research. ‡Corresponding Author. Dan Sanchez reports : The Saudis Go Full ISIS In Their US - Backed Takfiri War on the Shia Saudi Arabia has perpetrated a mass <t> execution </t> that puts ISIS’s beach beheadings to shame. Forty-seven heads rolled on Saturday. One of them belonged to Nimr al - Nimr, a revered Shi’ite cleric who had been sentenced to death for sermons in which he criticized the government (especially for its persecution of the country ’s Shi’ite minority). Cash - strapped Vivendi wants to <t> sell </t> Universal Studios, its Universal theme parks and television production company. Defendant Executor Crime Seller Event type: justice.judicialconsequences.execute Event type: Transaction.Transfer-Ownership Artifact Artifact Artifact Sentence-level EAE Document-level EAE Figure 1: Examples of (top) sentence-level and (bot- tom) document-level event argument extraction. Trigger words are included in special tokens <t> and </t>. Un- derlined words denote arguments and arcs denote roles. recent years (Wang et al., 2021), …

Content truncated to 500 words