Some Languages are More Equal than Others: Probing Deeper into the Linguistic Disparity in the NLP World

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 823–848 November 20–23, 2022. ©2022 Association for Computational Linguistics 823 Some Languages are More Equal than Others: Probing Deeper into the Linguistic Disparity in the NLP World Surangika Ranathunga and Nisansa de Silva Department of Computer Science and Engineering University of Moratuwa, Katubedda 10400, Sri Lanka {surangika, nisansaDds}@cse.mrt.ac.lk Abstract Linguistic disparity in the NLP world is a prob- lem that has been widely acknowledged re- cently. However, different facets of this prob- lem, or the reasons behind this disparity are seldom discussed within the NLP community. This paper provides a comprehensive analy- sis of the disparity that exists within the lan- guages of the world. We show that simply cat- egorising languages considering data availabil- ity may not be always correct. Using an exist- ing language categorisation based on speaker population and vitality, we analyse the dis- tribution of language data resources, amount of NLP/CL research, inclusion in multilingual web-based platforms and the inclusion in pre- trained multilingual models. We show that many languages do not get covered in these re- sources or platforms, and even within the lan- guages belonging to the same language group, there is wide disparity. We analyse the impact of family, geographical location, GDP and the speaker population of languages and provide possible reasons for this disparity, along with some suggestions to overcome the same. 1 Introduction Even after more than fifty years since the inception of the fields of Computational Linguistics (CL) and Natural Language Processing (NLP), we still ob- serve a significant bias favouring the so-called high- resource languages in the field. Conversely, this means that the majority of the 6500+ languages in the world, which have been classified as low- resource, have received limited to no attention. This resource poverty is not merely an academic or theoretical issue. It impacts the lives and the well-being of people concerned in a very present The paper title is inspired by the quote “All animals are equal, but some animals are more equal than others” by Or- well (1945) which satirically alludes to disparities that exist in places which, ostensibly are supposed to be homogeneous. In this paper, we discuss how the same phenomenon is observed in the broadly used language categorisation systems. and practical manner, and deprives the speakers of low-resource languages from reaping the benefits of NLP in areas such as healthcare (Perez-Rosas et al., 2020), disaster response (Ray Chowdhury et al., 2019), law (Ratnayaka et al., 2020), and edu- cation (Taghipour and Ng, 2016). This digital divide between high-resource and low-resource languages (LRLs)1 has been brought into the spotlight by many scholars in the field (Ben- der, 2019; Cains, 2019; Joshi et al., 2020; Anas- tasopoulos et al., 2020). Consequently, there have been efforts to build data sets covering low-resource languages (Conneau et al., 2018; Ebrahimi et al., 2022), benchmarks (Hu et …

Content truncated to 500 words