Oral to Web: Digitizing 'Zero Resource'Languages of Bangladesh

We present the Multilingual Cloud Corpus, the first national-scale, parallel, multimodal linguistic dataset of Bangladesh’s ethnic and indigenous languages. Despite being home to approximately 40 minority languages spanning four language families, Bangladesh has lacked a systematic, cross-family digital corpus for these predominantly oral, computationally “zero resource” varieties, 14 of which are classified as endangered. Our corpus comprises 85792 structured textual entries, each containing a Bengali stimulus text, an English translation, and an IPA transcription, together with approximately 107 hours of transcribed audio recordings, covering 42 language varieties from the Tibeto-Burman, Indo-European, Austro-Asiatic, and Dravidian families, plus two genetically unclassified languages. The data were collected through systematic fieldwork over 90 days across nine districts of Bangladesh, involving 16 data collectors, 77 speakers, and 43 validators, following a predefined elicitation template of 2224 unique items organized at three levels of linguistic granularity: isolated lexical items (475 words across 22 semantic domains), grammatical constructions (887 sentences across 21 categories including verbal conjugation paradigms), and directed speech (862 prompts across 46 conversational scenarios). Post-field processing included IPA transcription by 10 linguists with independent adjudication by 6 reviewers. The complete dataset is publicly accessible through the Multilingual Cloud platform (multiling.cloud), providing searchable access to annotated audio and textual data for all documented varieties. We describe the corpus design, fieldwork methodology, dataset structure, and per-language coverage, and discuss implications for endangered language documentation, low-resource NLP, and digital preservation in linguistically diverse developing countries.

💡 Research Summary

The paper presents the Multilingual Cloud Corpus, the first national‑scale, parallel, multimodal linguistic resource for Bangladesh’s ethnic and indigenous languages. The corpus contains 85,792 structured textual entries—each pairing a Bengali stimulus sentence, an English translation, and an IPA transcription—and roughly 107 hours of transcribed audio recordings. Data were collected over 90 days in nine districts, involving 77 speakers and 43 validators, and cover 42 language varieties from the Tibeto‑Burman, Indo‑European, Austro‑Asiatic, and Dravidian families, plus two unclassified languages. A pre‑designed elicitation template of 2,224 items was used, organized into three granularity levels: 475 isolated lexical items across 22 semantic domains, 887 grammatical constructions covering 21 categories (including verb conjugation paradigms), and 862 directed‑speech prompts spanning 46 conversational scenarios. Post‑field processing involved IPA transcription by ten linguists, with independent adjudication by six reviewers to ensure high quality. The dataset is openly accessible via the Multilingual Cloud platform (multiling.cloud), offering searchable audio and text for all documented varieties. The authors situate their work within three research strands: (i) the ethnolinguistic taxonomy of Bangladesh’s minorities, (ii) endangered‑language documentation practices, and (iii) the scarcity of computational resources for low‑resource South Asian languages. They argue that the corpus fills a critical gap by providing a systematically comparable, parallel resource that can support typological studies, low‑resource machine translation, and speech‑recognition research. The paper also discusses methodological strengths—such as rigorous pre‑field preparation, community engagement, and a two‑stage validation pipeline—as well as limitations, including reliance on a single Bengali stimulus (potentially obscuring cultural nuances), limited naturalness of elicited speech, and sparse data for critically endangered languages like Laleng and Thar. Future directions propose expanding natural conversation recordings, incorporating community‑driven data maintenance, and developing multilingual translation layers (local language ↔ Bengali ↔ English). Overall, the Multilingual Cloud Corpus represents a pioneering effort to digitize “zero‑resource” oral languages at a national level, offering a foundational asset for both linguistic preservation and the development of NLP technologies for some of the world’s most under‑documented languages.