Event Extraction in Video Transcripts

Proceedings of the 29th International Conference on Computational Linguistics, pages 7156–7165 October 12–17, 2022. 7156 Event Extraction in Video Transcripts Amir Pouran Ben Veyseh1, Viet Dac Lai1, Franck Dernoncourt2, and Thien Huu Nguyen1 1Department of Computer Science, University of Oregon, OR, USA 2Adobe Research, Seattle, WA, USA {apouranb, vietl, thien}@cs.uoregon.edu {franck.dernoncourt}@adobe.com Abstract Event extraction (EE) is one of the fundamental tasks for information extraction whose goal is to identify mentions of events and their partici- pants in text. Due to its importance, different methods and datasets have been introduced for EE. However, existing EE datasets are limited to formally written documents such as news articles or scientific papers. As such, the chal- lenges of EE in informal and noisy texts are not adequately studied. In particular, video tran- scripts constitute an important domain that can benefit tremendously from EE systems (e.g., video retrieval), but has not been studied in EE literature due to the lack of necessary datasets. To address this limitation, we propose the first large-scale EE dataset obtained for transcripts of streamed videos on the video hosting plat- form Behance to promote future research in this area. In addition, we extensively evalu- ate existing state-of-the-art EE methods on our new dataset. We demonstrate that such sys- tems cannot achieve adequate performance on the proposed dataset, revealing challenges and opportunities for further research effort. 1 Introduction Event Extraction is an important task in the full pipeline of Information Extraction. In EE, the goal is to identify mentions/trigger words of events, and their participants and attributes of interest. For in- stance, in the sentence “Joe Biden was born on November 20, 1942”, an event of Birth is men- tioned. An event mention consists of two important components: (1) Trigger: the word(s) that most clearly refer to the occurrence of the event (e.g, “born” in the above example); and (2) Argument: the entity mentions involved in the event with some role (e.g., “Joe Biden” with the role of Entity) or attributes of the event (e.g., time and location). Due to its importance, various methods and an- notated datasets have been proposed for EE (Ahn, 2006; Nguyen and Grishman, 2015; Yang et al., 2019; Wang et al., 2020). Also, with the prolifer- ation of EE methods, datasets for EE have been diversified to cover different domains and settings, e.g., multiple domains (Walker et al., 2006), mul- tiple languages, (Mitamura et al., 2016), literary texts (Sims et al., 2019), cybersecurity texts (Man Duc Trong et al., 2020), and events in long docu- ments (Ebner et al., 2020). However, despite all progress thus far, most of the available datasets for EE are restricted to the domains of formally written texts, e.g., news, reports, scientific papers, or books. As such, the challenges for EE in other domains with informal and noisy texts are less explored. One of such domains that has not been studied be- fore for EE involves video transcripts obtained by automatic speech recognition (ASR) tools. Since the such transcripts might be noisy, …

Content truncated to 500 words