Lexical Normalization for Code-switched Data and its Effect on POS Tagging

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 2352–2365 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 2352 Lexical Normalization for Code-switched Data and its Effect on POS Tagging Rob van der Goot IT University of Copenhagen robv@itu.dk ¨Ozlem C¸ etino˘glu IMS, University of Stuttgart ozlem@ims.uni-stuttgart.de Abstract Lexical normalization, the translation of non- canonical data to standard language, has shown to improve the performance of many natural language processing tasks on social media. Yet, using multiple languages in one ut- terance, also called code-switching (CS), is fre- quently overlooked by these normalization sys- tems, despite its common use in social media. In this paper, we propose three normalization models specifically designed to handle code- switched data which we evaluate for two lan- guage pairs: Indonesian-English (Id-En) and Turkish-German (Tr-De). For the latter, we in- troduce novel normalization layers and their corresponding language ID and POS tags for the dataset, and evaluate the downstream ef- fect of normalization on POS tagging. Results show that our CS-tailored normalization mod- els outperform Id-En state of the art and Tr-De monolingual models, and lead to 5.4% relative performance increase for POS tagging as com- pared to unnormalized input. 1 1 Introduction Social media provide an invaluable source of in- formation for natural language processing (NLP) systems. Its informative and spontaneous nature leads to many interesting phenomena, like non- standard words, spelling errors and abbreviations. One particularly challenging and interesting phe- nomenon is the use of multiple languages within the same utterance, which is also called code- switching (CS) (Gumperz, 1982; Myers-Scotton, 1995; Toribio and Bullock, 2012). Because most NLP models are designed to pro- cess canonical and monolingual data, their perfor- mance drops enormously when having to process 1Source code is available at: https://bitbucket. org/robvanderg/csmonoise. The Turkish-German data is available at: https://github.com/ozlemcek/ TrDeNormData social media data (Eisenstein, 2013). One solution to this problem is lexical normalization: the trans- lation of non-standard (e.g. social media) text to its canonical form (Han and Baldwin, 2011). Pre- vious work has shown that by standardizing the data, we can improve the robustness of NLP sys- tems (Derczynski et al., 2013; Zhang et al., 2013; van der Goot and van Noord, 2017). Nevertheless these systems overlook code-switching. (1) shows a code-switched tweet (upper) and its normaliza- tion annotation (lower), taken from an Indonesian- English CS corpus (Barik et al., 2019) (Indonesian in bold). This example demonstrates that CS com- plicates normalization, because it can be unclear in which language to normalize (e.g., ak is normalized to aku ‘I’ in Indonesian. English-only normaliza- tion systems would probably normalize it to ok). (1) ak aku .. luv love u you :(:( till till die die ‘I love you till (I) die’ Recently, there has been an increasing interest in the automatic processing of CS data, however, there has not been much work on its lexical normaliza- tion. To the best of our knowledge, only Adouane et al. (2019) focus entirely on …

Content truncated to 500 words