CodeSwitch-Reddit: Exploration of Written Multilingual Discourse in Online Discussion Forums

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4776–4786, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4776 CodeSwitch-Reddit: Exploration of Written Multilingual Discourse in Online Discussion Forums Ella Rabinovich Masih Sultani Suzanne Stevenson Dept. of Computer Science, University of Toronto, Canada {ella,masih,suzanne}@cs.toronto.edu Abstract In contrast to many decades of research on oral code-switching, the study of written multilin- gual productions has only recently enjoyed a surge of interest. Many open questions remain regarding the sociolinguistic underpinnings of written code-switching, and progress has been limited by a lack of suitable resources. We introduce a novel, large, and diverse dataset of written code-switched productions, curated from topical threads of multiple bilingual com- munities on the Reddit discussion platform, and explore questions that were mainly ad- dressed in the context of spoken language thus far. We investigate whether findings in oral code-switching concerning content and style, as well as speaker proficiency, are carried over into written code-switching in discussion fo- rums. The released dataset can further facili- tate a range of research and practical activities. 1 Introduction Multilingual communities adopt various commu- nicative strategies that navigate among multiple languages. One of the most notable of such strate- gies is code-switching (CS) – when a bilingual mixes two or more languages within a discourse, or even within a single utterance, as in: εγώ θα το ήθελα, θα ήταν το guilty pleasure μου [ I would like it, it would be my guilty pleasure ] The sociolinguistic underpinnings of code- switching as an oral conversational strategy have been investigated extensively for many decades. The purposes of oral CS have been shown to range from practical considerations of domain-specific lexical knowledge, to indications of register and nuanced meanings (see Gardner-Chloros, 2009, for a recent review). Oral code-switching is thus known to be a multifaceted component of success- ful communication in a multilingual community, interacting with factors of speaker proficiency, and both style and content of utterances. By contrast, the analysis of written code- switching has only recently enjoyed a surge of in- terest, and remains seriously under-studied (Sebba et al., 2012). Written text often differs greatly from conversation in its levels of both spontane- ity and formality, and findings thus far have dif- fered in their conclusions regarding the extent to which various genres of written text reflect the same communicative functions of CS as observed in oral conversation (e.g., McClure, 2001; Chan, 2009; Gardner-Chloros and Weston, 2015). The growing popularity of social media and on- line discussion platforms poses both opportuni- ties and new research questions regarding writ- ten code-switching. Global online forums, in which English is a lingua franca, not only draw on but create wide-reaching multilingual commu- nities. The resulting communications lead to a wealth of data that potentially includes a large amount of code-switching across multiple lan- guage pairs (e.g., Solorio et al., 2014; Diab et al., 2016; Aguilar et al., 2018b). Moreover, …

Content truncated to 500 words