Elamur Research⌕
☰
home / papers / 2026 / Improving Language Identification for Code-Switched Speech: The Pivotal Role of Accented English

Improving Language Identification for Code-Switched Speech: The Pivotal Role of Accented English

Paper · 2026 · Adyasha Patra, Dhiraj Kumar Sah and Preethi Jyothi

Summary

Spoken language identification (LID) systems are usually built for monolingual audio and struggle with code-switched speech. When English is mixed into a Hindi or Bengali utterance, pretrained models almost never detect it. The authors attribute this to English being spoken with the accent of the matrix language.

They adapt the MMS-LID model (wav2vec 2.0-based) using only small sets of matrix-language-accented English. Two parameter-efficient methods are compared: Adapters, and LoRA applied to the self-attention query, key and value projections. Evaluation covers Hindi-, Bengali-, Arabic- and Mandarin-English code-switched test sets, plus monolingual sets to check for forgetting. The paper also proposes LangRank, a ranking-based metric that reveals spurious English predictions which Exact Match misses.

LoRA on accented English gives the best balance: it recognises English in mixed utterances without wrongly predicting it in monolingual ones. Adapters, Whisper, and LoRA trained directly on code-switched data all fall short on this. The approach matters because accented English is far easier to collect than code-switched corpora. Reliable utterance-level LID is also needed to route speech to language-specific expert ASR models.

Key points

  • Base model. MMS-LID has 48 Transformer encoder layers and outputs 126 language classes. It was finetuned on 50–500 accented English samples per accent, with as few as 80 samples sufficing for the lightweight setup.
  • Code-switched test data. MUCS (Hindi-English, Bengali-English), ZAEBUC-Spoken (Arabic-English) and ASCEND (Mandarin-English).
  • Accented English training data. NISP (Hindi accent), Svarah (Bengali accent) and L2-Arctic (Arabic and Mandarin accents). The monolingual control sets are 2000 samples per language from Mozilla Common Voice v13.
  • Main result. Without finetuning, English detection in Hindi-English and Bengali-English speech is close to zero. LoRA has the smallest average Euclidean distance to the oracle on the LangRank (LR_en) trade-off plot (d_avg = 0.31). Adapters and Whisper score high on Exact Match but over-predict English in monolingual speech.
  • Limitations.
    • The method requires knowing the matrix language in advance.
    • It needs accent-matched data and does not transfer to mismatched accents.
    • It is tested only with English as the embedded language.
    • Remaining errors come from phonetically close distractor languages (e.g., Urdu, Punjabi) and from utterances with a low code-mixed index (CMI).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Adyasha Patra, Dhiraj Kumar Sah and Preethi Jyothi. 2026. Improving Language Identification for Code-Switched Speech: The Pivotal Role of Accented English. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.242

@inproceedings{patra2026improving,
  title = {{Improving Language Identification for Code-Switched Speech: The Pivotal Role of Accented English}},
  author = {Adyasha Patra, Dhiraj Kumar Sah and Preethi Jyothi},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.242},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.242}
}

References

43 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Maha Tufail Agro, Atharva Kulkarni, Karima Kadaoui, Zeerak Talat, and Hanan Aldarmaki. 2025. Code-switching in end-to-end automatic speech recognition: A systematic literature review. Preprint, arXiv:2507.07741. find
  2. R. Ardila, M. Branson, K. Davis, M. Henretty, M. Kohler, J. Meyer, R. Morais, L. Saunders, F. M. Tyers, and G. Weber. 2020. Common voice: A massively-multilingual speech corpus. In Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4211–4215.
  3. Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Preprint, arXiv:2006.11477.
  4. Laurie Burchell, Alexandra Birch, Robert Thompson, and Kenneth Heafield. 2024. Code-switched language identification is harder than you think. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 646–658, St. Ju… find
  5. Keyu Chen, Yuan Pang, and Zi Yang. 2024. Parameter-efficient fine-tuning with adapters. Preprint, arXiv:2405.05493. — [arXiv:2405.05493](https://arxiv.org/abs/2405.05493) find
  6. Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna. 2022. Fleurs: Few-shot learning evaluation of universal representations of speech. Preprint, arXiv:2205.12446.
  7. Anuj Diwan, Rakesh Vaideeswaran, Sanket Shah, Ankita Singh, Srinivasa Raghavan, Shreya Khare, Vinit Unni, Saurabh Vyas, Akash Rajpuria, Chiranjeevi Yarra, Ashish Mittal, Prasanta Kumar Ghosh, Preethi Jyothi, Kalika Bali, Vivek Seshadri, Sunayana Sitaram, Samarth Bharadwaj, Jai Nanavati, Raoul Nanava… find
  8. Mucs 2021: Multilingual and code-switching asr challenges for low resource indian languages. In Interspeech 2021, interspeech_2021. ISCA. find
  9. Hugging Face. 2024. Trainer class. Accessed: 2024-10-16. find
  10. Penelope Gardner-Chloros. 2009. Cambridge University Press. find
  11. Nizar Habash and David Palfreyman. 2022. ZAEBUC: An annotated Arabic-English bilingual writer corpus. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 79–88, Marseille, France. European Language Resources Association. find
  12. Injy Hamed, Fadhl Eryani, David Palfreyman, and Nizar Habash. 2024. Zaebuc-spoken: A multilingual multidialectal arabic-english speech corpus. Preprint, arXiv:2403.18182.
  13. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. Preprint, arXiv:2106.09685.
  14. Hukai Huang, Shenghui Lu, Yahui Shan, He Qu, Fengrun Zhang, Wenhao Guan, Qingyang Hong, and Lin Li. 2024. Dynamic language group-based moe: Enhancing code-switching speech recognition with hierarchical routing. Preprint, arXiv:2407.18581. — [arXiv:2407.18581](https://arxiv.org/abs/2407.18581) find
  15. Tahir Javed, Sakshi Joshi, Vignesh Nagarajan, Sai Sundaresan, Janki Nawale, Abhigyan Raman, Kaushal Bhogale, Pratyush Kumar, and Mitesh M. Khapra. 2023. Svarah: Evaluating english asr systems on indian accents. Preprint, arXiv:2305.15760.
  16. Tahir Javed, Janki Atul Nawale, Eldho Ittan George, Sakshi Joshi, Kaushal Santosh Bhogale, Deovrat Mehendale, Ishvinder Virender Sethi, Aparna Ananthanarayanan, Hafsah Faquih, Pratiti Palit, Sneha Ravishankar, Saranya Sukumaran, Tripura Panchagnula, Sunjay Murali, Kunal Sharad Gandhi, Ambujavalli R,… find
  17. Shareef Babu Kalluri, Deepu Vijayasenan, Sriram Ganapathy, Ragesh Rajan M, and Prashant Krishnan. 2020. Nisp: A multi-lingual multi-accent dataset for speaker profiling. Preprint, arXiv:2007.06021. — [arXiv:2007.06021](https://arxiv.org/abs/2007.06021) find
  18. Amir Hossein Kargaran, François Yvon, and Hinrich Schuetze. 2024. MaskLID: Code-switching language identification through iterative masking. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 459–469, Bangkok, Thailand. Associat… find
  19. Shuyue Stella Li, Cihan Xiao, Tianjian Li, and Bismarck Odoom. 2023. Simple yet effective code-switching language identification with multitask pre-training and transfer learning. Preprint, arXiv:2305.19759. — [arXiv:2305.19759](https://arxiv.org/abs/2305.19759) find
  20. Ignacio Lopez-Moreno, Javier Gonzalez-Dominguez, David Martinez, Oldřich Plchot, Joaquin Gonzalez-Rodriguez, and Pedro J. Moreno. 2016. On the use of deep feedforward neural networks for automatic language identification. Computer Speech & Language, 40:46–59. find
  21. Ignacio Lopez-Moreno, Javier Gonzalez-Dominguez, Oldrich Plchot, David Martinez, Joaquin Gonzalez-Rodriguez, and Pedro Moreno. 2014. Automatic language identification using deep neural networks. In 2014 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 5337– 53… find
  22. Holy Lovenia, Samuel Cahyawijaya, Genta Indra Winata, Peng Xu, Xu Yan, Zihan Liu, Rita Frieske, Tiezheng Yu, Wenliang Dai, Elham J Barezi, et al. 2022. Ascend: A spontaneous chinese-english dataset for code-switching in multi-turn conversation. In Proceedings of the 13th Language Resources and Evalu… find
  23. Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. 2022. Peft: State-of-the-art parameter-efficient fine-tuning methods. https://github.com/huggingface/peft. ↗
  24. Carol Myers-Scotton. 2017. Code-Switching, chapter 13. John Wiley & Sons, Ltd. find
  25. Yuting Nie, WeiQiang Zhang, Zhe Ji, and GuiXin Shi. 2022. Language code-switching detection based on bert-lid. In 2022 16th IEEE International Conference on Signal Processing (ICSP), volume 1, pages 36–40. find
  26. Chad Nilep. 2006. “code switching” in sociocultural linguistics. Colorado Research in Linguistics, 19. find
  27. Douglas O’Shaughnessy. 2024. Spoken language identification: An overview of past and present research trends. Speech Communication, page 103167. find
  28. Mario Piergallini, Rouzbeh Shirvani, Gauri Shankar Gautam, and Mohamed Chouikha. 2016. Word-level language identification and predicting codeswitching points in swahili-english language data. In Proceedings of the second workshop on computational approaches to code switching, pages 21–29. find
  29. Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, and Michael Auli. 2023. Scaling speech technology to 1,000+ languages. arXiv.
  30. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Robust speech recognition via large-scale weak supervision. arXiv preprint.
  31. P. Rao, M. Pandya, K. Sabu, K. Kumar, and N. Bondale. 2018. A study of lexical and prosodic cues to segmentation in a hindi-english code-switched discourse. In Proceedings of Interspeech, Hyderabad, India. find
  32. Mirco Ravanelli, Titouan Parcollet, Peter Plantinga, Aku Rouhe, Samuele Cornell, Loren Lugosch, Cem Subakan, Nauman Dawalatabad, Abdelwahab Heba, Jianyuan Zhong, Ju-Chieh Chou, Sung-Lin Yeh, Szu-Wei Fu, Chien-Feng Liao, Elena Rastorgueva, François Grondin, William Aris, Hwidong Na, Yan Gao, Renato D… — [arXiv:2106.04624](https://arxiv.org/abs/2106.04624) find
  33. Ramon Sanabria, Nikolay Bogoychev, Nina Markl, Andrea Carmantini, Ondrej Klejch, and Peter Bell. 2023. The edinburgh international accents of english corpus: Towards the democratization of english asr. Preprint, arXiv:2303.18110.
  34. Thamar Solorio and Yang Liu. 2008. Learning to predict code-switching points. In Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 973–981.
  35. Irshad Ahmad Thukroo, Rumaan Bashir, and Kaiser J. Giri. 2022. A review into deep learning techniques for spoken language identification. Multimedia Tools and Applications, 81(22):32593–32624. find
  36. Rong Tong, Bin Ma, Donglai Zhu, Haizhou Li, and Eng Siong Chng. 2006. Integrating acoustic, prosodic and phonotactic features for spoken language identification. In 2006 IEEE International Conference on Acoustics Speech and Signal Processing Proceedings, volume 1, pages I–I. IEEE. find
  37. Jörgen Valk and Tanel Alumäe. 2021. VoxLingua107: a dataset for spoken language recognition. In Proc. IEEE SLT Workshop. find
  38. Wenxuan Wang, Guodong Ma, Yuke Li, and Binbin Du. 2023. Language-routing mixture of experts for multilingual and code-switching speech recognition. Preprint, arXiv:2307.05956. — [arXiv:2307.05956](https://arxiv.org/abs/2307.05956) find
  39. Steven H. Weinberger. 2015. Speech accent archive. https://accent.gmu.edu/. Accessed: 2026-01-22. ↗
  40. Genta Winata, Alham Fikri Aji, Zheng Xin Yong, and Thamar Solorio. 2023. The decades progress on code-switching research in NLP: A systematic survey on trends and challenges. In Findings of the Association for Computational Linguistics: ACL 2023, pages 2936–2978, Toronto, Canada. Association for Com… find
  41. Genta Winata, Shijie Wu, Mayank Kulkarni, Thamar Solorio, and Daniel Preotiuc-Pietro. 2022. Cross-lingual few-shot learning on unseen languages. In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conferen… find
  42. Guanlong Zhao, Sinem Sonsaat, Alif Silpachai, Ivana Lucic, Evgeny Chukharev-Hudilainen, John Levis, *Figure 4: Block structure of an encoder layer in the MMS-LID model, adapted using LoRA finetuning.* The LoRA-based model comprises approximately 1.6 million learnable parameters, which are fine-tuned… find
  43. Marc A Zissman and Kay M Berkling. 2001. Automatic language identification. speech communication, 35(1-2):115–124. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/patra-2026-improving-language-identification-code-switched/