Developing Language Resources and NLP Tools for the North Korean Language

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5595–5600 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5595 Developing Language Resources and NLP Tools for the North Korean Language Arda Akdemir†, Yeojoo Jeon§, Tetsuo Shibuya† †Institute of Medical Science, The University of Tokyo § Graduate Schools for Law and Politics, The University of Tokyo †{aakdemir, tshibuya}@hgc.jp §jeon.yeojoo@gmail.com Abstract Since the division of Korea, the two Korean languages have diverged significantly over the last 70 years. However, due to the lack of linguistic source of the North Korean language, there is no DPRK-based language model. Consequently, scholars rely on the Korean language model by utilizing South Korean linguistic data. In this paper, we first present a large-scale dataset for the North Korean language. We use the dataset to train a BERT-based language model, DPRK-BERT. Second, we annotate a subset of this dataset for the sentiment analysis task. Finally, we compare the performance of different language models for masked language modeling and sentiment analysis tasks. Keywords: North Korean, Language Model, Sentiment Analysis, Less-resourced languages, Language Resources 1. Introduction It is now common sense in the Natural Language Pro- cessing (NLP) community that high-performing Ma- chine Learning (ML) models require large amount of relevant datasets. However, such a large corpus is not available for many less-resourced languages. An ex- treme example is the Democratic People’s Republic of Korea (DPRK) language. Due to the current circum- stances, there are almost no digital resources available online for the DPRK Korean language. This resource scarcity makes it very difficult, if not impossible, to develop statistical models for the DPRK Korean lan- guage. There are many publicly available labeled datasets for the Republic of Korea (ROK) language (Jang et al., 2013; Lim et al., 2019; Ham et al., 2020). More- over, it is relatively easy to obtain large unlabeled corpora through web crawling, or Wikipedia dumps for the RoK Korean language. These resources have greatly fostered Korean NLP research. For example, there are various high-performing deep pre-trained lan- guage models (PLMs) for the ROK language such as KLUE-BERT (Park et al., 2021), and KR-BERT (Lee et al., 2020). These models are applied to various down- stream tasks to achieve state-of-the-art results (Lee et al., 2020). Yet, the training data of these models do not include DPRK-specific text. Over the years, the Ko- rean language used in the countries diverged signifi- cantly from each other. Table 1 highlights some of the important differences in the written Korean language between the two countries. 1 As a result of these dif- ferences, we can not directly apply the RoK PLMs to analyze the DPRK data. The current NLP research on the DPRK language is highly limited in volume and scope because of a lack 1These differences are obtained fromKwon Jae Il (2015). of NLP tools and resources. Most NLP research on the language tries to mitigate this issue either by relying on domain experts’ manual readings of the DPRK articles …

Content truncated to 500 words