SyMCoM - Syntactic Measure of Code Mixing A Study Of English-Hindi Code-Mixing

Findings of the Association for Computational Linguistics: ACL 2022, pages 472 - 480 May 22-27, 2022 c⃝2022 Association for Computational Linguistics SyMCoM - Syntactic Measure of Code Mixing A Study Of English-Hindi Code-Mixing Prashant Kodali† Anmol Goel† Monojit Choudhury‡ Manish Shrivastava† Ponnurangam Kumaraguru† †International Institute of Information Technology Hyderabad ‡Microsoft Research, India {prashant.kodali, anmol.goel}@research.iiit.ac.in monojitc@microsoft.com, {m.shrivastava, pk.guru}@iiit.ac.in Abstract Code mixing is the linguistic phenomenon where bilingual speakers tend to switch be- tween two or more languages in conversa- tions. Recent work on code-mixing in com- putational settings has leveraged social media code mixed texts to train NLP models. For cap- turing the variety of code mixing in, and across corpus, Language ID (LID) tags based mea- sures (CMI) have been proposed. Syntactical variety/patterns of code-mixing and their rela- tionship vis-a-vis computational model’s per- formance is under explored. In this work, we in- vestigate a collection of English(en)-Hindi(hi) code-mixed datasets from a syntactic lens to propose, SyMCoM, an indicator of syntactic variety in code-mixed text, with intuitive theo- retical bounds. We train SoTA en-hi PoS tagger, accuracy of 93.4%, to reliably compute PoS tags on a corpus, and demonstrate the utility of SyMCoM by applying it on various syntacti- cal categories on a collection of datasets, and compare datasets using the measure. 1 Introduction Code-mixing refers to mixing of linguistic units and structures from multiple languages in a sin- gle utterance and/or conversation (Myers-Scotton, 1997). The complexity of code-mixing can be in- tuitively understood as the degree of structural in- terleaving between the languages at the level of the lexicon and morpho-syntax (Myers-Scotton, 1997), and also at the level of pragmatic and socio- linguistic functions of code-mixing in a linguistic community (Begum et al., 2016; Annamalai, 2001; Malhotra, 1980). It is an important notion that is linguistically well-studied and provides insights into cognitive and cultural aspects of human lan- guage. Additionally, quantification of this com- plexity has recently attracted attention of computa- tional linguists because studies have shown that the performance of the same model can widely vary on different code-mixed corpora. As a result, dif- Figure 1: Two sentences, having same language patterns, but the syntactic nature of the switched units are differ- ent - VERB is switched in Ex 1, NOUNS are switched in Ex 2. ferent metrics of complexity of code-mixing have been proposed such as CMI, Ratio-based measures time-course measures and memory-based measures (Guzmán et al., 2017; Gambäck and Das, 2016). But as Srivastava and Singh (2021) points out, these metrics are limited, partly because they are primar- ily based on language switch patterns at the token level, being completely agnostic to structural fea- tures. For instance, Figure 1 shows en−hi code-mixed sentences with the same language tag distribution, but in Example (1), verb is switched, while in (2) nouns are switched. The former seems much more complex and difficult to process cognitively and computationally, than the latter, where switching seems to be an extension of noun-borrowing (Bali et al., 2014). Motivated by such cases, in this …

Content truncated to 500 words