Criteria for Useful Automatic Romanization in South Asian Languages

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 680–688 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 680 PLOD: An Abbreviation Detection Dataset for Scientific Documents Leonardo Zilio1, Hadeel Saadany1, Prashant Sharma2, Diptesh Kanojia1,3, Constantin Or˘asan1,3 1Centre for Translation Studies, University of Surrey, United Kingdom. 2Hitachi CRL, Japan. 3Surrey Institute for People-centred AI, University of Surrey, United Kingdom. 1,3{l.zilio, h.saadany, d.kanojia, c.orasan}@surrey.ac.uk, 2prashaantsharmaa@gmail.com Abstract The detection and extraction of abbreviations from unstructured texts can help to improve the performance of Natural Language Processing tasks, such as machine translation and information retrieval. However, in terms of publicly available datasets, there is not enough data for training deep-neural-networks-based models to the point of generalising well over data. This paper presents PLOD, a large-scale dataset for abbreviation detection and extraction that contains 160k+ segments automatically annotated with abbreviations and their long forms. We performed manual validation over a set of instances and a complete automatic validation for this dataset. We then used it to generate several baseline models for detecting abbreviations and long forms. The best models achieved an F1-score of 0.92 for abbreviations and 0.89 for detecting their corresponding long forms. We release this dataset along with our code and all the models publicly in this Github Repository. Keywords: abbreviation detection, PLOS journals, abbreviation dataset 1. Introduction A pervasive characteristic of scientific reports and re- search papers is their frequent use of abbreviations (Wu et al., 2011). For submitting to various journals, au- thors are also required to submit an abbreviation glos- sary, i.e., a list of short forms with their expanded long forms. Such a glossary is essential for the reader to un- derstand the domain-specific terminology used in the reported work. From a linguistic point of view, there are different ty- pologies of abbreviations, and often authors disagree in relation to a common classification system (Fabijani´c, 2015). As Tchiotashvili et al. (2021) explain, there are authors advocating for the distinction between ini- tialisms and acronyms1, and others that defend a sepa- ration between shortening abbreviations and initial ab- breviations. Considering that a typology of abbrevia- tions is not our focus in this paper, we will use the terms “short form”, “abbreviation” and “abbreviated token” interchangeably and as umbrella terms to denote any token(s) that can be expanded into a longer token or into a sequence of tokens that corresponds to its long form. As such, unless explicitly stated, we will not differentiate between, for instance, abbreviations and acronyms, and nor does our dataset. From a Natural Language Processing (NLP) point of view, abbreviations are problematic for automatic pro- cessing, and the presence of short forms might hinder 1According to those authors, initialisms and acronyms are both short forms that are created using the initial letter(s) from a sequence of tokens, and are differentiated according to how they are pronounced. the machine processing of unstructured text. For ex- ample, a machine translation system may not provide a suitable translation for such …

Content truncated to 500 words