An Integrated Representation of Linguistic and Social Functions of Code-Switching

An Integrated Representation of Linguistic and Social Functions of Code-Switching Silvana Hartmann, Monojit Choudhury, Kalika Bali Microsoft Research Lab India Bangalore, India nlp@silvanahartmann.de,{monojitc,kalikab}@microsoft.com Abstract We present an integrated representation of code-switching (CS) functions, i.e., a representation that includes various CS phenomena (intra- /inter-sentential) and modalities (written/spoken), and aims to derive CS functions from local and global properties of the code-switched discourse. By applying it to several English/Hindi CS datasets, we show that our model contributes i) to the standardization and re-use of CS data collections by creating a resource footprint, and ii) to the study of CS functions by creating a systematic description and hierarchy of reported functions together with the (local and social) properties that may affect them. At the same time, the model provides a flexible framework to add emerging functions, supporting theoretical studies as well as the automatic detection of CS functions. Keywords: Code-Switching, Multilingual, Bilingual, Pragmatics, Discourse 1. Introduction Code-Switching (CS), or alternating between two or more languages in a single conversation, is a marked feature of multilingual communities. Linguists have studied this phe- nomenon in great detail and recently, with the rise of so- cial media, the processing and generating of CS content has gained attention in the NLP community as well. The amount and type of code-switching depends on a number of structural, functional and social factors (Begum et al., 2016). Linguistic studies in the past have focused on two aspects that may be summarized as the ”how” and the ”why” of code-switching (Poplack, 2015). The ”how” aims to explain the grammatical principles that underlie code-switching per- formance; the ”why” aims to explain the function of code- switching in discourse, defining social, pragmatic, and dis- course functions of code-switching, for instance addressee specification, emphasis, or marking quotations. These stud- ies are mostly based on limited recordings of conversations, concentrating on a small subset of functions, either linguis- tic, based on local properties of the discourse, or social and other global aspects. There are no large empirical studies on the interaction of linguistic and social aspects of CS func- tions. Further, there is no unified theory on the different CS functions and how they are expressed in discourse. Previ- ous theoretical work identifies lists of functions (see e.g., Abdul-Zahra (2010)), and recent work in NLP picks specific functions and aims to identify them in large collections of CS texts (Begum et al., 2016; Rudra et al., 2016). Thus, such studies use different levels of analysis, and often confuse observations (such as observing code-switched reiteration or translation) with CS functions (reiteration is often used for emphasis), as in the following example from Begum et al. (2016) that uses the translation of an utterance for emphasis: dimaag mein bhoosa bhara hai [gloss:up in their heads with fodder]. up in their heads with fodder. The task of defining a comprehensive set of functions is very difficult, if not elusive. The same switching phenomenon, in our example translation, is used for emphasis in some communities and for weakening request …

Content truncated to 500 words