Borrowing or Codeswitching? Annotating for Finer-Grained Distinctions in Language Mixing

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3195–3201 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3195 Borrowing or Codeswitching? Annotating for Finer-Grained Distinctions in Language Mixing Elena ´Alvarez Mellado1, Constantine Lignos2 1NLP & IR Group, School of Computer Science, UNED 2Michtom School of Computer Science, Brandeis University elena.alvarez@lsi.uned.es, lignos@brandeis.edu Abstract We present a new corpus of Twitter data annotated for codeswitching and borrowing between Spanish and English. The corpus contains 9,500 tweets annotated at the token level with codeswitches, borrowings, and named entities. This corpus differs from prior corpora of codeswitching in that we attempt to clearly define and annotate the boundary between codeswitching and borrowing and do not treat common “internet-speak” (lol, etc.) as a codeswitch when used in an otherwise monolingual context. The result is a corpus that enables the study and modeling of Spanish-English borrowing and codeswitching on Twitter in one dataset. We present baseline scores for modeling the labels of this corpus using Transformer-based language models. The annotation itself is released with a CC BY 4.0 license, while the text it applies to is distributed in compliance with the Twitter terms of service. Keywords: codeswitching, borrowing, corpora 1. Challenges in Annotating Codeswitching and Borrowing Over the last decade, several initiatives have drawn at- tention to codeswitching as a challenging NLP task, such as various shared tasks (Aguilar et al., 2018; Molina et al., 2016; Solorio et al., 2014) and a recent benchmark (Aguilar et al., 2020). Prior work on creating datasets of codeswitching has framed the task of language identification for codeswitched utterances as token-level annotation, where every token receives a language identification tag (Maharjan et al., 2015, among others). For exam- ple, in a collection of English-Spanish codeswitched tweets, tokens in Spanish will be labeled with a lan- guage identification tag for Spanish, and tokens in En- glish will be labeled with a language identification tag for English. Further labels have been proposed and adopted to the repertoire of tags in codeswitching datasets, such as ambiguous for words whose language is difficult to de- termine even in context, other for tokens in languages other than the main languages under study, mixed for intralexical codeswitching (words that combine mor- phemes from different languages), NE for named en- tities, and none for punctuation marks, emoji, Twitter mentions, etc. This repertoire of labels has become the usual tagset in codeswitching shared tasks (Aguilar et al., 2018; Molina et al., 2016; Solorio et al., 2014). However, this tagset does not take into account that not all other- language inclusions are necessarily a codeswitch. In the prior work we examined, we were not able to identify explicit, published definitions or guidelines on what should constitute a codeswitch—or perhaps more crucially, what is not a codeswitch. After all, lan- guage mixing can happen in cases that are not neces- sarily codeswitching. Given an utterance that is mostly monolingual, should we assume that any token from another language is a …

Content truncated to 500 words