A Multi-Format Transfer Learning Model for Event Argument Extraction via Variational Information Bottleneck

Proceedings of the 29th International Conference on Computational Linguistics, pages 1990–2000 October 12–17, 2022. 1990 A Multi-Format Transfer Learning Model for Event Argument Extraction via Variational Information Bottleneck Jie Zhou1†*, Qi Zhang2†, Qin Chen2, Qi Zhang1, Liang He2, and Xuanjing Huang1 1School of Computer Science, Fudan Univerisity, Shanghai, China 2School of Computer Science and Technology, East China Normal University, Shanghai, China {jie_zhou, qz, xjhuang}@fudan.edu.cn qzhang@stu.ecnu.edu.cn; {qchen, lhe}@cs.ecnu.edu.cn Abstract Event argument extraction (EAE) aims to ex- tract arguments with given roles from texts, which have been widely studied in natural lan- guage processing. Most previous works have achieved good performance in specific EAE datasets with dedicated neural architectures. Whereas, these architectures are usually dif- ficult to adapt to new datasets/scenarios with various annotation schemas or formats. Fur- thermore, they rely on large-scale labeled data for training, which is unavailable due to the high labelling cost in most cases. In this paper, we propose a multi-format transfer learning model with variational information bottleneck, which makes use of the information especially the common knowledge in existing datasets for EAE in new datasets. Specifically, we intro- duce a shared-specific prompt framework to learn both format-shared and format-specific knowledge from datasets with different formats. In order to further absorb the common knowl- edge for EAE and eliminate the irrelevant noise, we integrate variational information bottleneck into our architecture to refine the shared repre- sentation. We conduct extensive experiments on three benchmark datasets, and obtain new state-of-the-art performance on EAE. 1 Introduction Event Extraction (EE) has received widespread at- tention in recent years, which aims to obtain struc- tured information (e.g., trigger, event types, argu- ments, and argument roles) from large unstructured text corpora (Lu et al., 2021; Zhang et al., 2022). Event argument extraction (EAE) plays a crucial role in EE. Recently, deep learning-based mod- els have obtained tremendous success in this task. However, these methods rely on a large-scale la- beled dataset for training, which is time-consuming and labor-intensive due to the complexity of event extraction. *Corresponding authors; †Equal contribution. ACE2005 Example Conflict.Attack And to the south , British[Place] forces[Target] continue their attack[trigger] on targets around Basra . trigger attack Target forces Place Basra Attacker - Victim - Instrument - WIKIEVENTS Example Conflict.Attack.DetonateExplode For example , Ms . Davis has identified a man whose photo matches that of a \" John Doe # 2 \" sought immediately after the Murrah Building[Target] attack[trigger]. He appears to be a Palestinian by the name of Hussain Hashem Al Hussaini ... trigger attack Target Murrah Building Place - Attacker - ExplosiveDevice - Instrument - Figure 1: An example with a different format. In this paper, we aim to answer the question “Can we transfer the knowledge from the exist- ing complex event extraction datasets with differ- ent formats?". There are several event extraction datasets, such as ACE 2005 (Doddington et al., 2004), RAMS (Ebner et al., 2020), and WikiEvents (Li et al., 2021). These datasets contain abun- dant event types and semantic roles that may pos- sess overlap …

Content truncated to 500 words