Transfer Learning from Semantic Role Labeling to Event Argument Extraction with Template-based Slot Querying

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2627 - 2647 December 7-11, 2022 ©2022 Association for Computational Linguistics Transfer Learning from Semantic Role Labeling to Event Argument Extraction with Template-based Slot Querying Zhisong Zhang, Emma Strubell, Eduard Hovy Language Technologies Institute, Carnegie Mellon University zhisongz@cs.cmu.edu, strubell@cmu.edu, hovy@cmu.edu Abstract In this work, we investigate transfer learning from semantic role labeling (SRL) to event argument extraction (EAE), considering their similar argument structures. We view the ex- traction task as a role querying problem, unify- ing various methods into a single framework. There are key discrepancies on role labels and distant arguments between semantic role and event argument annotations. To mitigate these discrepancies, we specify natural language-like queries to tackle the label mismatch problem and devise argument augmentation to recover distant arguments. We show that SRL annota- tions can serve as a valuable resource for EAE, and a template-based slot querying strategy is especially effective for facilitating the transfer. In extensive evaluations on two English EAE benchmarks, our proposed model obtains im- pressive zero-shot results by leveraging SRL annotations, reaching nearly 80% of the fully- supervised scores. It further provides benefits in low-resource cases, where few EAE anno- tations are available. Moreover, we show that our approach generalizes to cross-domain and multilingual scenarios. 1 Introduction Event argument extraction (EAE) is a key compo- nent in the task of event extraction (Ahn, 2006) that aims to identify the arguments that serve as roles for event frames. While recent developments in neural network models have enabled impressive improvements on this task in the fully-supervised setting (Wang et al., 2019b; Pouran Ben Veyseh et al., 2020; Ma et al., 2020; Li et al., 2021b), EAE remains challenging when abundant annotations are not available. In particular, event schemes are usually specific to the target scenarios. For ex- ample, events in biomedical domains, like GENE- EXPRESSION in GENIA (Kim et al., 2008), can be quite different than the ones in ACE (LDC, 2005), such as ATTACK and CONTACT. It is costly and He went to the store and bought a book . TransferOwnership buy.01 Transport go.02 Artifact Destination Buyer Artifact Place Arg0 Arg4 Arg0 Arg1 Event: (ACE) SRL: (PropBank) Figure 1: Example annotations with ACE events (above) and PropBank semantic frames (below). Brown and red rectangles indicate predicate and argument words, respectively. Green lines denote argument links. inefficient to annotate large amounts of data for every new application. Compared with the specific schemes in EAE, semantic role labeling (SRL; Gildea and Juraf- sky, 2002; Palmer et al., 2010) extracts predicate- argument structures with more general and broad- coverage frame ontologies. SRL also enjoys rich and carefully annotated resources, such as Prop- Bank (Palmer et al., 2005) and FrameNet (Baker et al., 1998), covering a wide range of semantic frame types. As shown in the example in Figure 1, SRL closely resembles EAE: they both specify se- mantic frames triggered by predicate words and aim at finding arguments for participating roles. Therefore, it …

Content truncated to 500 words