CNLP-NITS-PP at MixMT 2022: Hinglish-English Code-Mixed Machine Translation

Proceedings of the Seventh Conference on Machine Translation (WMT), pages 1158 - 1161 December 7-8, 2022 ©2022 Association for Computational Linguistics CNLP-NITS-PP at MixMT 2022: Hinglish–English Code-Mixed Machine Translation Sahinur Rahman Laskar1, Rahul Singh1, Shyambabu Pandey1, Riyanka Manna2 Partha Pakray1, Sivaji Bandyopadhyay1 1Department of Computer Science and Engineering, National Institute of Technology, Silchar, India 2Department of Computer Science and Engineering, Adamas University, Kolkata, India {sahinurlaskar.nits, rahuljan, babushyampandey, riyankamanna16}@gmail.com {parthapakray,sivaji.cse.ju}@gmail.com Abstract The mixing of two or more languages in speech or text is known as code-mixing. In this form of communication, users mix words and phrases from multiple languages. Code-mixing is very common in the context of Indian languages due to the presence of multilingual societies. The probability of the existence of code-mixed sen- tences in almost all Indian languages since in India English is the dominant language for so- cial media textual communication platforms. We have participated in the WMT22 shared task of code-mixed machine translation with the team name: CNLP-NITS-PP. In this task, we have prepared a synthetic Hinglish–English parallel corpus using transliteration of origi- nal Hindi sentences to tackle the limitation of the parallel corpus, where, we mainly consid- ered sentences that have named-entity (proper noun) from the available English-Hindi parallel corpus. With the addition of synthetic bi-text data to the original parallel corpus (train set), our transformer-based neural machine transla- tion models have attained recall-oriented un- derstudy for gisting evaluation (ROUGE-L) scores of 0.23815, 0.33729, and word error rate (WER) scores of 0.95458, 0.88451 at Sub- Task-1 (English-to-Hinglish) and Sub-Task-2 (Hinglish-to-English) for test set results respec- tively. 1 Introduction The mixing of alternating words from two different language vocabulary without misinterpreting the context of the sentence is known as code-switching or code-mixing (Poulisse, 1998). This style of com- munication is one of the most frequent in multilin- gual communities, such as India. English is exten- sively mixed with local languages, such as Hindi, and Bengali, which causes code-mixed English- Hindi: Hinglish and English-Bengali: Binglish languages (Sailaja, 2011). Code-mixing is not ob- served in formal literature such as books but is com- monly used on social media platforms such as Face- book and Twitter. The WMT22 organizes shared task code-mixed machine translation for English- to-Hinglish and Hinglish-to-English, where the main challenge is low-resource availability of par- allel corpus. We have participated in the same task and to mitigate the issue of data scarcity, a synthetic Hinglish-English parallel corpus is pre- pared (as discussed in Section 3.1). In this work, the transformer-based neural machine translation (NMT) technique (Vaswani et al., 2017; Laskar et al., 2022) is utilized to build NMT models for both directions (English-to-Hinglish, Hinglish-to- English) of code-mixed MT. 2 Related Work In recent times, many significant NLP studies have included the study of code-mixed languages. The EMNLP 2022 seventh conference on machine translation (WMT22) has put forward several tasks directed to meet new challenges in the field of NLP for code-mixed Indian languages. The competition has attracted many researchers to follow up with these tasks, which have …

Content truncated to 500 words