Hengam: An Adversarially Trained Transformer for Persian Temporal Tagging

Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1013–1024 November 20–23, 2022. ©2022 Association for Computational Linguistics 1013 Hengam: An Adversarially Trained Transformer for Persian Temporal Tagging Sajad Mirzababaei⋆ Amir Hossein Kargaran⋆ Hinrich Schütze Ehsaneddin Asgari Center for Information and Language Processing, LMU Munich, Germany Computer Engineering Department, Sharif University of Technology, IR NLP Expert Center, Data:Lab, Volkswagen AG, Munich, Germany asgari@berkeley.edu and inquiries@cislmu.org Abstract Many NLP main tasks benefit from an accurate understanding of temporal expressions, e.g., text summarization, question answering, and information retrieval. This paper introduces Hengam, an adversarially trained transformer for Persian temporal tagging outperforming state-of-the-art approaches on a diverse and manually created dataset. We create Hengam in the following concrete steps: (1) we develop HengamTagger, an extensible rule-based tool that can extract temporal expressions from a set of diverse language-specific patterns for any language of interest. (2) We apply HengamTag- ger to annotate temporal tags in a large and di- verse Persian text collection (covering both for- mal and informal contexts) to be used as weakly labeled data. (3) We introduce an adversarially trained transformer model on HengamCorpus that can generalize over the HengamTagger’s rules. We create HengamGold, the first high- quality gold standard for Persian temporal tag- ging. Our trained adversarial HengamTrans- former not only achieves the best performance in terms of the F1-score (a type F1-Score of 95.42 and a partial F1-Score of 91.60) but also successfully deals with language ambiguities and incorrect spellings. Our code, data, and models are publicly available at https:// github.com/kargaranamir/Hengam. 1 Introduction A wide array of natural language processing (NLP) applications relies on accurately identifying of events and their respective occurrence times. Text summarization (Christensen et al., 2013; Aslam et al., 2015; Ghodratnama et al., 2021), question answering (Llorens et al., 2015; Bast and Hauss- mann, 2015; Jia et al., 2018, 2021), and information retrieval tasks requiring to classify information in a chronological order (Kanhabua and Nejdl, 2013) are all examples of such applications. In order to ⋆ The first two authors contributed equally and their au- thorships were determined randomly. address these needs in the last decades, there has been an increased interest in temporal information extraction systems and developing their appropri- ate corpora and evaluation frameworks. TempEval challenges are, for instance, great examples of such efforts held as a part of SemEval workshops focus- ing on temporal information extraction (Verhagen et al., 2007, 2010; UzZaman et al., 2013). The study of temporal expressions in English and other languages has been an ongoing re- search track in the last decade, spanning renowned rule-based efforts such as HeidelTime (Strötgen and Gertz, 2010) and SUTime (Chang and Man- ning, 2012) to learning-based approaches, e.g., a transformer-based “BERT got a Date” (Almasian et al., 2021). The majority of efforts in this area have been rule-based, which is suffering from (i) a relatively low recall, as finite rules are usually …

Content truncated to 500 words