Bangla NLP dataset. Bangla NER,POStag, text summarization, stopword, translate, sentiment analysis, wiki articles, root word, dataset etc.
72
15 commits
updated May 3, 2026
README
🇧🇩 Bangla NLP Dataset
A comprehensive, URL-validated collection of Bangla / Bengali NLP datasets, models, tools, and corpora — for researchers, students, and developers.
বাংলা ভাষার এনএলপি গবেষণা, শিক্ষা এবং প্রায়োগিক কাজের জন্য একটি যাচাইকৃত সম্পদ-সংগ্রহ।
🔄 Our sbnltk dataset is in LFS mode — clone the repository to download data.
🚀 All deep-learning-era datasets are linked below; we'll keep adding new releases.
📑 Table of Contents
- 📖 About
- 🎯 sbnltk Dataset List
- 🤖 Pre-trained Language Models
- 📚 Latest 2024–2026 Datasets
- 📊 Benchmarking and Evaluation
- 📰 News, Corpora & Pretraining
- 🔄 Machine Translation & Paraphrase
- 🎤 Speech (ASR / TTS / Emotion)
- 😊 Sentiment / Emotion / Sarcasm
- 🛡️ Hate Speech / Toxic / Cyberbullying
- 🕵️ Fake News & Misinformation
- 🏷️ NER / POS / Parsing
- ❓ Question Answering
- 📝 Text Summarization
- 🖊️ OCR, Handwriting & Document Layout
- ✋ Sign Language & Multimodal Vision
- 🗺️ Regional Dialects
- 🧪 NLI / Bias / Misc
- 🔧 NLP Tools & Libraries
- 📄 Research Papers
- 🇧🇩 Bangladesh Government Resources
- 🔗 Curated Aggregator Lists
- 💡 Motivation & Contribution
📖 About
This repository contains the sbnltk datasets used in the Bangla NLP toolkit sbnltk, and serves as a comprehensive, URL-validated catalogue of publicly available Bangla NLP resources contributed by the worldwide Bangla research community.
Validation note: Every link in this document was tested between 2025–2026. Resources that previously appeared here under fabricated GitHub paths (e.g.
github.com/poetry-bangla/corpus,github.com/medical-bangla/medical-translation, etc.) have been removed. If you find a dead link, please open an issue.
🎯 sbnltk Dataset List (DUMP & HUMAN Evaluated)
| Dataset | Description | Link |
|---|---|---|
| Number List | Bangla number list | 📥 Download |
| Root Word List | Bangla root word list | 📥 Download |
| Word List | Bangla word list (highest → lowest occurrence) | 📥 Download |
| Wiki Dump | Bangla wiki dump words | 📥 Download |
| POS Tag Static | Bangla POS-tag static dataset (single word) | 📥 Download |
| NER Static | Bangla NER static dataset (single word) | 📥 Download |
| Stop Words | Bangla stop-word list | 📥 Download |
| Dump POS Tag | Bangla dump POS-tag | 📥 Download |
| Question Classification | Bangla dump question classification dataset | 📥 Download |
| Sentiment Analysis | Bangla dump sentiment analysis | 📥 Download |
| Translation Dataset | Google translation dataset | 📥 Download |
| NER Enhanced | Existing NER dataset (modified + Date entity) | 📥 Download |
| News Articles | News article dataset | 📥 Download |
| POS Converted | POS-tag converted data | 📥 Download |
| POS Human Evaluated | POS-tag human-evaluated data | 📥 Download |
| NER Dump (Both) | Dump NER (active + passive) | 📥 Download |
| NER Dump (Active) | Dump NER (active only) | 📥 Download |
| Extractive Summarization | Extractive text summarization | 🔗 GitHub |
| Abstractive Summarization | Abstractive summarization (newspaper) | 📥 Drive · 📊 Kaggle |
| Text Classification | News article classification | 📥 Drive · 📊 Kaggle |
| Keywords Classification | Topic-keyword classification | 📥 Drive · 📊 Kaggle |
🤖 Pre-trained Language Models
BERT-style Encoders
| Model | Description | Params | Link |
|---|---|---|---|
| BanglaBERT | ELECTRA discriminator, SOTA Bangla NLU (BUET CSE NLP) | 110M | 🤗 HF · 🔗 GitHub |
| BanglaBERT (Small) | Lightweight variant | 13M | 🤗 HF |
| BanglaBERT (Large) | Large variant, top scores on BLUB | 335M | 🤗 HF |
| BanglishBERT | Bilingual (Bangla + English) | 110M | 🤗 HF |
| Bangla BERT Base (sagorsarker) | Popular community BERT | 110M | 🤗 HF |
| mBERT-Bengali-NER | Multilingual BERT fine-tuned for NER | — | 🤗 HF |
| mBERT-Bengali-TyDiQA-QA | mBERT fine-tuned for QA | — | 🤗 HF |
| sahajBERT | ALBERT-based collaborative training | 18M | 🤗 HF |
| MuRIL | Google multilingual (17 Indian) | 236M | 🤗 HF |
| IndicBERT | AI4Bharat (12 Indian) | — | 🤗 HF |
Generative / Seq2Seq Models
| Model | Description | Params | Link |
|---|---|---|---|
| BanglaT5 | T5-style seq2seq (BUET) | 247M | 🤗 HF |
| BanglaT5 (small) | Small T5 variant | 60M | 🤗 HF |
| BanglaT5 NMT bn↔en | Translation seq2seq | — | 🤗 bn→en · 🤗 en→bn |
| BanglaT5-Paraphrase | Paraphrase seq2seq | — | 🤗 HF |
| BanglaByT5 | Byte-level T5 | small | 📄 arXiv 2505.17102 |
| GPT-2 Bengali | Flax-community GPT-2 | 117M | 🤗 HF |
Bangla LLMs (2025)
| Model | Description | Params | Link |
|---|---|---|---|
| TigerLLM-1B-it | Bangla instruction-tuned LLM | 1B | 🤗 HF |
| TigerLLM-9B-it | Larger variant, beats GPT-3.5 on Bangla | 9B | 🤗 HF |
| TituLLMs (1B / 3B) | Family of Bangla LLMs with benchmarks | 1B / 3B | 📄 arXiv 2502.11187 |
| TigerLLM Paper | ACL 2025 short paper | — | 📄 arXiv 2503.10995 · 📄 ACL 2025 |
| BanglaLLaMA-3-8B-BnWiki-Instruct | Llama-3 fine-tuned on Bn Wiki | 8B | 🤗 HF |
| Bangla LLaMA (saiful9379) | LoRA-tuned LLaMA | — | 🔗 GitHub |
Speech Models
| Model | Description | Performance | Link |
|---|---|---|---|
| Wav2Vec2-Bengali (300M) | Self-supervised ASR | 17.8 % WER | 🤗 HF |
| Wav2Vec2-XLSR Bengali | XLSR fine-tune | — | 🤗 HF |
| BanglaConformer | Conformer ASR by Bengali.AI | — | 🤗 HF |
| BanglaASR | Whisper fine-tuned for Bengali | 14.73 % WER | 🤗 HF · 🔗 GitHub |
| Whisper (multilingual) | OpenAI base model — Bn supported | various sizes | 🤗 HF |
Word & Sentence Embeddings
| Resource | Description | Link |
|---|---|---|
| Bangla FastText (sagorsarker) | 20 M-token wiki-trained skipgram + CBOW | 🤗 HF |
| Bangla Word2Vec (sagorsarker) | 100-d Wikipedia embeddings | 🤗 HF |
| fastText 157-language Bengali | Facebook 300-d Wiki + CC | 🌐 fastText |
Spark NLP bengali_cc_300d | Production embedding | 🔗 Spark NLP |
| BanglaEmbed | Cross-lingual distilled sentence embeddings | 📄 arXiv 2411.15270 |
📚 Latest 2024–2026 Datasets
These are the most relevant new releases — cite the original authors when used.
| Dataset | Task | Size / Notes | Link |
|---|---|---|---|
| Bangla-Instruct | Instruction-tuning | 342 K instruction–response pairs | 🤗 HF |
| Bangla-TextBook | LM pretraining | 9.9 M tokens, 163 NCTB textbooks | 🤗 HF |
| BanglaSTEM | Technical-domain MT | 5 K Bn-En STEM sentence pairs | 📄 arXiv 2511.03498 |
| NCTB-QA | Educational QA | 87 805 QA pairs (grade 1–10) | 📄 arXiv 2603.05462 |
| BanglaQuAD | Open-domain QA | 30 808 QA pairs | 📄 arXiv 2410.10229 |
| ANCHOLIK-NER | Regional-dialect NER | 17 405 sentences, 5 regions | 📄 arXiv 2502.11198 |
| BanNERD (NAACL 2025) | NER, 10 classes / 29 domains | 85 K sentences, 991 K tokens | 🔗 GitHub |
| ONUBAD | Dialect→Standard MT | Chittagong / Sylhet / Barisal | 🔗 ScienceDirect |
| BanglaDial | Dialect text corpus | 60 729 entries × 11 dialects | 🔗 PMC |
| BIDWESH | Regional hate speech | Multi-region | 📄 arXiv 2507.16183 |
| BanglaTLit | Romanized→Bn back-transliteration | 42.7 K + 245.7 K pretrain | 📄 ACL 2024 |
| BanglishRev | E-commerce code-mix reviews | 1.74 M Daraz reviews | 📄 arXiv 2412.13161 |
| BengaliSent140 | Hate vs non-hate fusion | 140 K speeches | 📄 arXiv 2601.20129 · 🔗 IEEE DataPort |
| BLUCK | LLM cultural-knowledge benchmark | 2 366 MCQs / 23 categories | 📄 arXiv 2505.21092 |
| BNLI (refined) | NLI | Curated entail/contra/neutral | 📄 arXiv 2511.08813 |
| MultiBanAbs | Multi-domain abstractive sum. | Multi-corpus | 📄 arXiv 2511.19317 |
| MultiBanFakeDetect | Multimodal fake news | Text + image | 🔗 ScienceDirect |
| BanFakeNews-2.0 | Fake news (2024) | 47 K real + 13 K fake | 📊 Mendeley |
| BanglaHealth | Health-domain paraphrase | 200 K sentences | 🔗 ScienceDirect |
| BanglaCHQ-Summ | Consumer-health-question summary | 2 350 pairs (BLP-2023) | 🔗 GitHub |
| Bangla-MedER | Medical NER | 2 980 texts, 6 entity types | 📊 Mendeley |
| BanglaSarc3 | Sarcasm (ternary) | 12 089 FB comments | 🔗 ScienceDirect |
| VACASPATI | Bangla literature corpus | 11 M sentences / 115 M words | 📄 arXiv 2307.05083 |
| MixSarc | Code-mix sarcasm/humor/offence | Bn-En transliterated | 📄 arXiv 2602.21608 |
| EmoMix-3L | Code-mix emotion | 1 071 Bn-Hi-En instances | 🔗 GitHub |
| Bangla-ToCo | Context-aware toxic | 1 004 FB news comments | 🔗 ScienceDirect |
| BanglaDocAtlas | Document-layout, 8 classes | annotated complex docs | 🔗 IEEE |
| FoodBD | BD cuisine images, 67 categories | 3 523 polygon-annotated meals | 🔗 Springer 2025 |
| DeshiFoodBD | BD traditional food images | 5 425 images / 19 dishes | 🔗 Springer |
📊 Benchmarking and Evaluation
BLUB — Bangla Language Understanding Benchmark
The first comprehensive Bangla NLU benchmark, introduced with BanglaBERT (NAACL 2022).
| Task | Dataset | Metric | Best Model | Score |
|---|---|---|---|---|
| Sentiment Classification | SentNoB | Macro-F1 | BanglaBERT | 72.89 |
| Natural Language Inference | XNLI-bn / BNLI | Accuracy | BanglaBERT (Large) | 83.41 |
| Named Entity Recognition | MultiCoNER | Micro-F1 | BanglaBERT (Large) | 79.20 |
| Question Answering | SQuAD-bn / TyDiQA | EM / F1 | BanglaBERT (Large) | 76.10 / 81.50 |
📄 BLUB code & leaderboard: github.com/csebuetnlp/banglabert
BLUCK — Bangla LLM Cultural & Linguistic Benchmark (2025)
2 366 multiple-choice questions across 23 categories covering Bangladesh culture, history, and Bangla linguistics — designed to probe LLM cultural knowledge. 📄 arXiv 2505.21092
Recent Benchmark Datasets
| Dataset | Task | Size | Link |
|---|---|---|---|
| BanglaBook | Sentiment | 158 065 reviews | 🔗 GitHub |
| SentMix-3L / OffMix-3L | Code-mix sentiment / offence | ~1 K each | 📄 ACL |
| MultiCoNER (Bangla) | Multilingual complex NER | task | 🔗 multiconer.github.io |
📰 News, Corpora & Pretraining Data
| Dataset | Size | Link |
|---|---|---|
| Bangla2B+ (BanglaBERT pretraining corpus) | 27.5 GB / 110 sites | 🔗 GitHub |
| BanglaLM (data-mining corpus) | 14 GB | 📄 IEEE |
| BdNC – Bangladesh National Corpus ✅ | 40 GB / 3 B+ words | 🔗 corpus.bangla.gov.bd |
| VACASPATI literary corpus | 11 M sentences | 📄 arXiv 2307.05083 |
| CC-100 Bangla | 8.3 GB | 🔗 StatMT |
| OSCAR Bangla | 12 GB+ | 🔗 OSCAR |
| AI4Bharat IndicCorp | 9 B tokens incl. Bangla | 🔗 site |
| AI4Bharat IndicNLP corpus + catalog | meta-resource | 🔗 corpus · 🔗 catalog |
| Bangla Wikipedia Corpus (Kaggle) | wiki-text | 📊 Kaggle |
| Wikipedia bnwiki dumps | latest dumps | 🔗 dumps.wikimedia.org |
| Leipzig Bengali corpora (2021) | 1.65 M sentences | 🔗 corpora.uni-leipzig.de |
| Wiki Articles (Kaggle) | wiki snapshot | 📊 Kaggle |
| 40k News Articles | 40 K | 📊 Kaggle |
| Largest Bangla Newspaper | large multi-paper | 📊 Kaggle |
| bdNews24 corpus | bdnews24 articles | 📊 Kaggle |
| Bangladesh Protidin | Bangladesh Protidin news | 📊 Kaggle |
| csebuetnlp/xlsum | XL-Sum (Bangla subset) | 🤗 HF |
| csebuetnlp/dailydialogue_bn | Daily-dialogue translated | 🤗 HF |
| goru001/nlp-for-bengali | ULMFiT model + Wiki / news data | 🔗 GitHub |
| masiur/Bangla-Corpus | Open community corpus | 🔗 GitHub |
🔄 Machine Translation & Paraphrase
| Dataset | Description | Link |
|---|---|---|
| csebuetnlp/BanglaNMT | 2.38 M Bn-En pairs (133 MB) | 🤗 HF · 🔗 GitHub |
| AI4Bharat Samanantar | 49.6 M sentence pairs across Indic languages | 🤗 HF · 🔗 site |
| SUPara0.8M | Balanced En-Bn corpus | 🔗 IEEE DataPort |
| BanglaSTEM | 5 K STEM Bn-En pairs | 📄 arXiv 2511.03498 |
| WMT24 Bangla seed dataset | High-quality manual translation | 📄 ACL 2024 |
| BanglaParaphrase | 466 K paraphrase pairs (AACL 2022) | 🤗 HF · 🔗 GitHub |
| OPUS Collections | Multi-source parallel corpora | 🔗 OPUS |
| Bengali Visual Genome 1.0 | 29 K image-caption multimodal | 🔗 LINDAT |
| Google Dakshina | 12 South-Asian transliteration / parallel | 🔗 GitHub |
| BanglaTLit | Romanized → Bangla | 📄 ACL 2024 |
| bntranslit | Transliteration toolkit | 🔗 GitHub |
| Bengali Dictionary (Minhas Kamal) | Dictionary | 🔗 GitHub |
| TED2020 (Bangla) | TED multilingual | 📥 TSV |
🎤 Speech (ASR / TTS / Emotion)
ASR / Recognition
| Dataset | Description / Size | Link |
|---|---|---|
| OpenSLR-53 | Large Bengali ASR — 196 K utterances / 14.6 GB (Google) | 🔗 OpenSLR |
| OpenSLR (HF mirror) | All OpenSLR languages | 🤗 HF |
| OpenSLR-104 | Multilingual code-switching | 🔗 OpenSLR |
| Bengali Common Voice (Mozilla) | 399 h+ / 19 817 contributors (v9.0) | 🔗 Mozilla |
| Bengali.AI OOD-Speech | 1 177 h / 22 645 speakers — largest Bn ASR | 🔗 Bengali.AI |
| FLEURS Bangla | Cross-lingual 12 h | 🤗 HF |
| BanglaASR Dataset | Fine-tuned ASR | 🔗 GitHub |
| SUST-CSE-Speech / banspeech | TTS / ASR corpus | 🤗 HF |
| SKNahin / open-large-bengali-asr-data | Large open ASR | 🤗 HF |
| Bangla-Speech-Corpora (Bangla-Language-Processing) | Cleaned TTS-ready speech | 🔗 GitHub |
TTS
| Dataset | Description | Link |
|---|---|---|
| OpenSLR-37 | High-quality Google Bengali TTS | 🔗 OpenSLR |
| Bengali.AI TTS dataset | Studio-quality TTS | 🔗 site |
| bangla-tts (zabir-nabil) | Real-time multilingual synthesis | 🔗 GitHub |
Speech Emotion
| Dataset | Description | Link |
|---|---|---|
| SUBESCO | 7 K utterances / 7 emotions, gender-balanced | 📄 PLOS One |
| BanglaSER | 1 467 utterances / 34 speakers / 5 emotions | 🔗 ScienceDirect |
| KBES | Realistic speech-emotion w/ intensity | 🔗 ScienceDirect |
| BANSpEmo | Bangla emotional speech | 📄 arXiv 2312.14020 |
Speech-to-Text Toolkits
| Tool | Notes | Link |
|---|---|---|
| BanglaSpeech2Text | Whisper-FT offline ASR (mp3/mp4/wav) | 🔗 GitHub |
😊 Sentiment / Emotion / Sarcasm
| Dataset | Size / Notes | Link |
|---|---|---|
| BanglaBook | 158 K book reviews | 🔗 GitHub |
| SentNoB | Noisy social-media sentiment | 📊 Kaggle |
| EmoNoBa | 22 698 comments / 6 emotions (AACL 2022) | 📊 Kaggle · 📄 ACL |
| BanglaEmotion (shaoncsecu) | Emotion benchmark | 🔗 GitHub |
| MONOVAB | Multi-label emotion | 📄 paper |
| BAN-ABSA / BANGLA-ABSA | 9 009 aspect-level comments | 📄 arXiv 2012.00288 · 📊 Mendeley |
| BanglaSenti (lexicon) | 61 582 polarity words | 🔗 GitHub |
| banglanlp/bangla-sentiment-classification | Compiled benchmarking sets | 🔗 GitHub |
| Ayubur sentiment datasets | Multiple legacy datasets | 🔗 GitHub |
| BanglaSarc | 5 112 sarcasm samples | 📊 Kaggle |
| BanglaSarc3 | 12 089 ternary-class sarcasm | 🔗 ScienceDirect |
| BnSentMix | 20 K Bn-En code-mix sentiment | 📄 paper |
| SentMix-3L / OffMix-3L | Bn-En-Hi code-mix | 📄 ACL |
| Drama Review | Bengali drama reviews | 📊 Figshare |
| YouTube Sentiment / Emotion | YT comments | 📊 Kaggle |
| News Comments Sentiment | Bn news comments | 📊 Kaggle |
| News Headline Categories | Headline classification | 📊 Kaggle |
| Big News Classification | Large news classifier | 📊 Kaggle |
| News Article Classification (IndicNLP) | Indic news classifier | 📊 Kaggle |
| Bengali-Banglish Emotion | Mixed-script | 📊 Mendeley |
| E-commerce Sentiment + Emotion | 78 130 Daraz / Pickaboo reviews | 🔗 ScienceDirect |
| BanglishRev | 1.74 M Daraz code-mix reviews | 📄 arXiv 2412.13161 |
🛡️ Hate Speech / Toxic / Cyberbullying
| Dataset | Size / Notes | Link |
|---|---|---|
| Bengali-Hate-Speech (rezacsedu) | 6 418 / 5 categories | 🔗 GitHub · 📊 UCI |
| Bengali Hate Speech (naurosromim) | Annotated hate dataset | 📊 Kaggle |
| BIDWESH | Regional-based hate speech | 📄 arXiv 2507.16183 |
| BengaliSent140 | 140 K hate vs non-hate | 🔗 IEEE DataPort |
| ToxLex_bn | 1 959 bigrams from 2.2 M FB comments | 📊 Mendeley |
| Multi-Labeled Bengali Toxic | 16 073 / 7 labels | 🔗 GitHub |
| Bangla-ToCo | 1 004 context-aware toxic | 🔗 ScienceDirect |
| Bangla Multilabel Cyberbully | 12 557 (5 classes) | 📊 Mendeley |
| Bangla Social-Media Cyberbullying | YT/FB/IG/TikTok | 🔗 IEEE DataPort |
| Code-mixed Chaos (Banglish toxic) | 10 234 multi-label | 📊 Mendeley |
| BanglaMedia | 7 725 YT comments — 10 topics, 4 sentiments | 📊 Mendeley |
| VITD (BLP-2023 violence) | Violence-inciting text — 3 classes | 🔗 BLP |
🕵️ Fake News & Misinformation
| Dataset | Size | Link |
|---|---|---|
| BanFakeNews (LREC 2020) | 50 K news | 📊 Kaggle · 📄 arXiv 2004.08789 |
| BanFakeNews-2.0 (2024) | 47 K real + 13 K fake | 📊 Mendeley |
| MultiBanFakeDetect | Multimodal text + image | 🔗 ScienceDirect |
| Rowan1224/FakeNews | Code & data | 🔗 GitHub |
| DataCOVID19 | 14 571 COVID misinformation | 🔗 Springer |
🏷️ NER / POS / Parsing
| Dataset | Size / Notes | Link |
|---|---|---|
| B-NER (IEEE Access) | 22 144 sentences | 🔗 IEEE |
| BanNERD (NAACL 2025) | 85 K sentences / 991 K tokens / 10 classes / 29 domains | 🔗 GitHub |
| NER-Bangla-Dataset (MISabic) | 70 K sentences / 5 types | 🔗 GitHub |
| bnlp-resources NER | Train/dev/test splits | 🌐 banglanlp.github.io |
| ANCHOLIK-NER | Regional NER, 5 regions | 📄 arXiv 2502.11198 |
| celloscope_bangla_ner_dataset | 319 K NER | 🤗 HF |
| celloscope_bangla_ner_dataset (small) | 6.57 K | 🤗 HF |
| Bangla-MedER | Medical NER 2 980 / 6 types | 📊 Mendeley |
| Bangla NER (towhidahmedfoysal) | 400 K word-level | 📊 Kaggle |
| POS — 3 K sentences | abhishekgupta92 | 🔗 GitHub |
| POS — 100 K+ words (towhidahmedfoysal) | Word-level POS | 📊 Kaggle |
| UD_Bengali-BRU treebank | 14 UPOS tags, UD v2.9+ | 🌐 universaldependencies.org |
❓ Question Answering
| Dataset | Size | Link |
|---|---|---|
| csebuetnlp/squad_bn | 118 K train QA | 🤗 HF |
| BanglaRQA | 14 889 QA / 3 K passages (EMNLP 2022) | 🔗 GitHub |
| Bengali QA (Mayeesha) | SQuAD 2.0–style Bn QA | 📊 Kaggle |
| BanglaQuAD | 30 808 QA pairs | 📄 arXiv 2410.10229 |
| NCTB-QA | 87 805 educational QA | 📄 arXiv 2603.05462 |
| doctor_qa_bangla | 5.14 K medical QA | 🤗 HF |
| TyDiQA (Bengali subset) | Cross-lingual QA | 🤗 HF |
📝 Text Summarization
| Dataset | Size / Notes | Link |
|---|---|---|
| csebuetnlp/xlsum | XL-Sum (Bangla subset, BBC) | 🤗 HF |
| BanglaCHQ-Summ | 2 350 health-question summary pairs | 🔗 GitHub |
| MultiBanAbs | Multi-domain abstractive | 📄 arXiv 2511.19317 |
| BNLPC + NCTB | EACL 2021 unsupervised abstractive | 🔗 GitHub |
| BANSData (Prithwiraj) | News abstractive | 📊 Kaggle |
| Bengali Text Summarization (Hasan Moni) | Extractive + abstractive | 📊 Kaggle |
| BUSUM-BNLP | Multi-document update summarization | 📊 Kaggle |
| bnSum_gemma7b-it | Gemma-7B inst news summary system | 🔗 GitHub |
| Bangla-Text-summarization-Dataset (Abid) | Extractive | 🔗 GitHub |
| 3 Human-Evaluated articles (BNLPC) | Reference summaries | 🌐 BNLPC |
🖊️ OCR, Handwriting & Document Layout
| Dataset | Size / Notes | Link |
|---|---|---|
| NumtaDB | 85 K handwritten Bn-digit images | 📊 Kaggle |
| Bengali.AI Handwritten Grapheme Classification | Kaggle competition | 📊 Kaggle |
| Ekush | Handwritten Bn characters (largest) | 🌐 rabby.dev/ekush · 📊 Kaggle |
| BN-HTRd | 788 pages / 150 writers / 108 K words (HTR) | 📊 Mendeley |
| BanglaWriting | Multi-purpose offline HTR | 📄 paper |
| Bayanno | Multi-purpose handwriting | 📊 Mendeley |
| Bongabdo | Bn handwritten script | 📄 arXiv 2101.00204 |
| CMATERdb (1 / 2.1.2) | 5 K word imgs / 18 K Bn city names | 🔗 site |
| BaDLAD | 33 695 layout samples / 6 domains | 📄 paper |
| BanglaDocAtlas | 8-class complex Bn document layout | 🔗 IEEE |
| DL Sprint 2.0 (BUET CSE Fest 2023) | Layout segmentation | 📊 Kaggle |
| Bangla License Plate 2.5 K | 2 519 plate images | 🔗 Zenodo |
| BD-ALPDR | 725 high-res LP images | 🌐 site |
| Govt. Bangla OCR service | Free Bangla OCR | 🌐 ocr.bangla.gov.bd |
✋ Sign Language & Multimodal Vision
| Dataset | Size / Notes | Link |
|---|---|---|
| BdSLW60 | 60 sign words / 9 307 video trials | 📄 arXiv 2402.08635 |
| KU-BdSL | 30 classes / 38 consonants | 📊 Mendeley |
| BAUST Lipi | 18 K imgs / 36 alphabets | 📄 arXiv 2408.10518 |
| BDSL 49 | 29 490 imgs / 49 labels | 📄 arXiv 2208.06827 |
| BdSL36 | 4 M+ imgs / 36 cats | 📄 paper |
| Ego-SLD | Egocentric Bn sign-language video | 🔗 IEEE DataPort |
| Bengali Visual Genome 1.0 | Multimodal MT + captioning | 🔗 LINDAT |
| BAN-Cap | English-Bangla image-description | 📄 arXiv 2205.14462 |
| BNATURE | Bn image captioning | 📊 Kaggle |
| BanglaView | 31 783 imgs / 158 K captions | 📊 Mendeley |
| csebuetnlp/illusionVQA | Bn-aware VQA | 🤗 HF |
| DeshiFoodBD | 5 425 BD-cuisine images / 19 dishes | 🔗 Springer |
| FoodBD | 3 523 polygon-annotated meals (2025) | 🔗 Springer |
| BnLiT — Bangla Image-to-Text | Natural-language image-text | 📊 Kaggle |
🗺️ Regional Dialects
| Dataset | Coverage | Link |
|---|---|---|
| ANCHOLIK-NER | Barishal, Chittagong, Mymensingh, Noakhali, Sylhet | 📄 arXiv 2502.11198 |
| ONUBAD | Chittagong / Sylhet / Barisal → Standard Bn | 🔗 ScienceDirect |
| BanglaDial | 11 dialects / 60 729 entries | 🔗 PMC |
| BIDWESH | Regional hate speech | 📄 arXiv 2507.16183 |
| BanglaCHQ-Summ + dialect benchmarks | Sylheti / Chittagonian | 📄 ACL 2025 |
| Sylheti → Standard NMT | Sylheti corpus | 🔗 ScienceDirect |
🧪 NLI / Bias / Misc
| Dataset | Description | Link |
|---|---|---|
| csebuetnlp/xnli_bn | Bangla NLI translated from XNLI | 🤗 HF |
| BNLI (refined) | Curated entail/contra/neutral | 📄 arXiv 2511.08813 |
| csebuetnlp/BanglaSocialBias | Social-bias evaluation | 🤗 HF |
| csebuetnlp/BanglaContextualBias | Contextual-bias evaluation | 🤗 HF |
| csebuetnlp/CrossSum | Cross-lingual summarization | 🤗 HF |
| stopwords-iso/stopwords-bn | Stopword list | 🔗 GitHub |
| Bangla Plagiarism Dataset | 59.9 K | 🤗 HF |
| Banking 14-intents (en + bn + banglish) | 16.5 K intent samples | 🤗 HF |
| Massive intent (bn-BD) | 16.5 K | 🤗 HF |
| BanglaMusicStylo | 2 824 lyrics / 211 lyricists | 🔗 IEEE DataPort |
| Bangla Song Lyrics (genres + artists) | Bn song-lyric corpus | 📊 Kaggle |
| Bn Numbers w/ Words | Number-name dataset | 📊 Kaggle |
| likhonsheikh/BanglaNLP | 120 K parallel news pairs | 🤗 HF |
🔧 NLP Tools & Libraries
Python Libraries
| Library | Description | Link |
|---|---|---|
| BNLP (sagorbrur) | Comprehensive Bengali NLP toolkit | 🔗 GitHub · pip install bnlp_toolkit |
| BNLTK | Bangla NLP toolkit (tokenize / stem / POS) | 🔗 GitHub |
| sbnltk | This repo's toolkit (sentiment / NER / POS / sum) | 🔗 GitHub |
| bangla-stemmer | Lightweight Bn stemmer | 🔗 PyPI |
| bnunicode | Bijoy → Unicode normalization | 🔗 GitHub |
| Indic NLP Library | Multi-Indic processing / transliteration | 🔗 GitHub |
| bntranslit | Bengali transliteration | 🔗 GitHub |
| BanglaSpeech2Text | Bangla offline ASR | 🔗 GitHub |
| bangla-tts | Bangla TTS library | 🔗 GitHub |
| BanglaKit organisation | Tools, datasets, resources | 🔗 GitHub |
OCR / Vision
| Tool | Notes | Link |
|---|---|---|
| EasyOCR | Built-in Bangla support | 🔗 GitHub |
| Tesseract OCR | ben traineddata available | 🔗 GitHub |
| ocr.bangla.gov.bd | Govt-hosted Bangla OCR service | 🌐 site |
📄 Research Papers
Foundational
- BanglaBERT (NAACL Findings 2022) — 📄 ACL · 📄 arXiv 2101.00204
- BanglaNLG / BanglaT5 (EACL 2023) — 📄 arXiv 2205.11081
- BanglaParaphrase (AACL 2022) — 📄 arXiv 2210.05109
- Not Low-Resource Anymore: BanglaNMT (EMNLP 2020) — 📄 ACL
- MuRIL (Indic multilingual) — 📄 arXiv 2103.10730
- Bangla NLP Comprehensive Analysis — 📄 arXiv 2105.14875
LLMs & Generation (2024–2025)
- TigerLLM (ACL 2025) — 📄 arXiv 2503.10995
- TituLLMs — 📄 arXiv 2502.11187
- BanglaByT5 — 📄 arXiv 2505.17102
- BanglaLlama — 📄 arXiv 2410.21200
- WMT24 Bangla seed — 📄 ACL
QA / Reading Comprehension
- BanglaRQA (EMNLP 2022 Findings) — 📄 ACL
- BanglaQuAD — 📄 arXiv 2410.10229
- NCTB-QA — 📄 arXiv 2603.05462
NER
- B-NER (IEEE Access) — 🔗 IEEE
- ANCHOLIK-NER — 📄 arXiv 2502.11198
- BanNERD (NAACL 2025) — 🔗 GitHub
Summarization
- BanglaCHQ-Summ (BLP 2023) — 📄 ACL
- Bangla extractive (IEEE) — 🔗 IEEE
- MultiBanAbs — 📄 arXiv 2511.19317
Speech
- BanglaDialecto (2024) — 📄 arXiv 2411.10879
- Bengali Common Voice ASR — 📄 arXiv 2206.14053
- Bengali ASR system survey — 📄 arXiv 2209.08119
- SUBESCO (PLOS One 2021) — 📄 PLOS
Workshops
- BLP-2023 @ EMNLP (Singapore) — 🌐 blp-workshop.github.io · 📚 ACL Anthology
- BLP-2025 @ AACL-IJCNLP (Mumbai) — 🌐 blp-workshop.github.io · shared tasks: Bangla code generation, sentiment, etc.
🇧🇩 Bangladesh Government Resources
বাংলাদেশ সরকারের তথ্যপ্রযুক্তি ও বাংলা ভাষাগত সম্পদ:
| Resource | বাংলা নাম | Link |
|---|---|---|
| Bangladesh National Corpus (BdNC) | বাংলাদেশ জাতীয় কর্পাস (৪০ GB / ৩B+ শব্দ) | 🌐 corpus.bangla.gov.bd |
| Govt. Bangla OCR | সরকারি বাংলা OCR সেবা | 🌐 ocr.bangla.gov.bd |
| Accessible Dictionary | প্রতিবন্ধী-বান্ধব বাংলা অভিধান | 🌐 accessibledictionary.gov.bd |
| EBLICT Project | তথ্যপ্রযুক্তিতে বাংলা ভাষা সমৃদ্ধকরণ প্রকল্প | 🌐 eblict.portal.gov.bd |
| IPA — Information Processing Authority | বাংলা ভাষা প্রক্রিয়াকরণ কর্তৃপক্ষ | 🌐 ipa.bangla.gov.bd |
| Bangladesh Computer Council (BCC) | বাংলাদেশ কম্পিউটার কাউন্সিল | 🌐 bcc.gov.bd |
| Bangla Academy | বাংলা একাডেমি | 🌐 banglaacademy.gov.bd |
🔗 Curated Aggregator Lists (start here)
| Resource | Maintainer | Link |
|---|---|---|
| bnlp-resources | banglanlp | 🔗 GitHub · 🌐 site |
| awesome-bangla | banglakit | 🔗 GitHub |
| bangla-corpus | sagorbrur | 🔗 GitHub |
| Awesome_Bangla_Datasets | sabbirhossainujjal | 🔗 GitHub |
| AI4Bharat Indic NLP catalog | AI4Bharat | 🔗 GitHub |
| Mahadih534 — Bangla NLP HF collection | Mahadih534 | 🤗 HF |
| Mahadih534 — Bangla TTS collection | Mahadih534 | 🤗 HF |
| Mahadih534 — Bangla LLM finetuning | Mahadih534 | 🤗 HF |
| csebuetnlp organization | BUET CSE NLP | 🤗 HF |
| Sudipta Kar Bangla NLP resources | personal | 🌐 site |
| হাতেকলমে Bangla NLP (Rakibul Hassan) | educational notebooks | 🔗 GitHub · 🌐 book |
| GitHub topic: bangla-nlp | community-tagged | 🌐 topic |
| GitHub topic: bengali-nlp | community-tagged | 🌐 topic |
| Awesome Public Datasets (general NLP) | community | 🔗 GitHub |
| NLP Datasets Collection | community | 🔗 GitHub |
💡 Motivation & Contribution
Bangla is the 6th-most-spoken language in the world (~270 million native speakers) but remains classified as low-resource in NLP. This repository exists to make every Bangla NLP resource one click away — accurately documented and free of fabricated links.
How to Get Started
- For Pre-trained Models — visit the HuggingFace links above and load directly with
transformers. - For Tools —
pip install bnlp_toolkitorpip install bnltk. - For Datasets — follow individual links; honor each dataset's license (most are CC BY-NC-SA 4.0).
- For Research — start with the BLP workshop proceedings and the latest 2024–2026 release table.
Contributing
- 📝 Submit new datasets via pull request — include a working URL and one-line description.
- 🐛 Report broken / fabricated links by opening an issue.
- 🔬 Tag your own paper to share with the community.
- ✅ Every PR is welcome — please verify URLs return 200 before submitting.
⭐ If this repository helps your work, please give it a star! ⭐
🤝 Contributions, corrections, and additions are very welcome.
🌟 Thanks to every researcher and developer pushing Bangla NLP forward.
☕ Support This Project
If this resource has been helpful, you can support its maintenance:
bengali-dataset
dataset
nlp
question-answering
sentiment-analysis
text-classification
text-summarization
Bangla NLP dataset. Bangla NER,POStag, text summarization, stopword, translate, sentiment analysis, wiki articles, root word, dataset etc.
72
15 commits
updated May 3, 2026
README
🇧🇩 Bangla NLP Dataset
A comprehensive, URL-validated collection of Bangla / Bengali NLP datasets, models, tools, and corpora — for researchers, students, and developers.
বাংলা ভাষার এনএলপি গবেষণা, শিক্ষা এবং প্রায়োগিক কাজের জন্য একটি যাচাইকৃত সম্পদ-সংগ্রহ।
🔄 Our sbnltk dataset is in LFS mode — clone the repository to download data.
🚀 All deep-learning-era datasets are linked below; we'll keep adding new releases.
📑 Table of Contents
- 📖 About
- 🎯 sbnltk Dataset List
- 🤖 Pre-trained Language Models
- 📚 Latest 2024–2026 Datasets
- 📊 Benchmarking and Evaluation
- 📰 News, Corpora & Pretraining
- 🔄 Machine Translation & Paraphrase
- 🎤 Speech (ASR / TTS / Emotion)
- 😊 Sentiment / Emotion / Sarcasm
- 🛡️ Hate Speech / Toxic / Cyberbullying
- 🕵️ Fake News & Misinformation
- 🏷️ NER / POS / Parsing
- ❓ Question Answering
- 📝 Text Summarization
- 🖊️ OCR, Handwriting & Document Layout
- ✋ Sign Language & Multimodal Vision
- 🗺️ Regional Dialects
- 🧪 NLI / Bias / Misc
- 🔧 NLP Tools & Libraries
- 📄 Research Papers
- 🇧🇩 Bangladesh Government Resources
- 🔗 Curated Aggregator Lists
- 💡 Motivation & Contribution
📖 About
This repository contains the sbnltk datasets used in the Bangla NLP toolkit sbnltk, and serves as a comprehensive, URL-validated catalogue of publicly available Bangla NLP resources contributed by the worldwide Bangla research community.
Validation note: Every link in this document was tested between 2025–2026. Resources that previously appeared here under fabricated GitHub paths (e.g.
github.com/poetry-bangla/corpus,github.com/medical-bangla/medical-translation, etc.) have been removed. If you find a dead link, please open an issue.
🎯 sbnltk Dataset List (DUMP & HUMAN Evaluated)
| Dataset | Description | Link |
|---|---|---|
| Number List | Bangla number list | 📥 Download |
| Root Word List | Bangla root word list | 📥 Download |
| Word List | Bangla word list (highest → lowest occurrence) | 📥 Download |
| Wiki Dump | Bangla wiki dump words | 📥 Download |
| POS Tag Static | Bangla POS-tag static dataset (single word) | 📥 Download |
| NER Static | Bangla NER static dataset (single word) | 📥 Download |
| Stop Words | Bangla stop-word list | 📥 Download |
| Dump POS Tag | Bangla dump POS-tag | 📥 Download |
| Question Classification | Bangla dump question classification dataset | 📥 Download |
| Sentiment Analysis | Bangla dump sentiment analysis | 📥 Download |
| Translation Dataset | Google translation dataset | 📥 Download |
| NER Enhanced | Existing NER dataset (modified + Date entity) | 📥 Download |
| News Articles | News article dataset | 📥 Download |
| POS Converted | POS-tag converted data | 📥 Download |
| POS Human Evaluated | POS-tag human-evaluated data | 📥 Download |
| NER Dump (Both) | Dump NER (active + passive) | 📥 Download |
| NER Dump (Active) | Dump NER (active only) | 📥 Download |
| Extractive Summarization | Extractive text summarization | 🔗 GitHub |
| Abstractive Summarization | Abstractive summarization (newspaper) | 📥 Drive · 📊 Kaggle |
| Text Classification | News article classification | 📥 Drive · 📊 Kaggle |
| Keywords Classification | Topic-keyword classification | 📥 Drive · 📊 Kaggle |
🤖 Pre-trained Language Models
BERT-style Encoders
| Model | Description | Params | Link |
|---|---|---|---|
| BanglaBERT | ELECTRA discriminator, SOTA Bangla NLU (BUET CSE NLP) | 110M | 🤗 HF · 🔗 GitHub |
| BanglaBERT (Small) | Lightweight variant | 13M | 🤗 HF |
| BanglaBERT (Large) | Large variant, top scores on BLUB | 335M | 🤗 HF |
| BanglishBERT | Bilingual (Bangla + English) | 110M | 🤗 HF |
| Bangla BERT Base (sagorsarker) | Popular community BERT | 110M | 🤗 HF |
| mBERT-Bengali-NER | Multilingual BERT fine-tuned for NER | — | 🤗 HF |
| mBERT-Bengali-TyDiQA-QA | mBERT fine-tuned for QA | — | 🤗 HF |
| sahajBERT | ALBERT-based collaborative training | 18M | 🤗 HF |
| MuRIL | Google multilingual (17 Indian) | 236M | 🤗 HF |
| IndicBERT | AI4Bharat (12 Indian) | — | 🤗 HF |
Generative / Seq2Seq Models
| Model | Description | Params | Link |
|---|---|---|---|
| BanglaT5 | T5-style seq2seq (BUET) | 247M | 🤗 HF |
| BanglaT5 (small) | Small T5 variant | 60M | 🤗 HF |
| BanglaT5 NMT bn↔en | Translation seq2seq | — | 🤗 bn→en · 🤗 en→bn |
| BanglaT5-Paraphrase | Paraphrase seq2seq | — | 🤗 HF |
| BanglaByT5 | Byte-level T5 | small | 📄 arXiv 2505.17102 |
| GPT-2 Bengali | Flax-community GPT-2 | 117M | 🤗 HF |
Bangla LLMs (2025)
| Model | Description | Params | Link |
|---|---|---|---|
| TigerLLM-1B-it | Bangla instruction-tuned LLM | 1B | 🤗 HF |
| TigerLLM-9B-it | Larger variant, beats GPT-3.5 on Bangla | 9B | 🤗 HF |
| TituLLMs (1B / 3B) | Family of Bangla LLMs with benchmarks | 1B / 3B | 📄 arXiv 2502.11187 |
| TigerLLM Paper | ACL 2025 short paper | — | 📄 arXiv 2503.10995 · 📄 ACL 2025 |
| BanglaLLaMA-3-8B-BnWiki-Instruct | Llama-3 fine-tuned on Bn Wiki | 8B | 🤗 HF |
| Bangla LLaMA (saiful9379) | LoRA-tuned LLaMA | — | 🔗 GitHub |
Speech Models
| Model | Description | Performance | Link |
|---|---|---|---|
| Wav2Vec2-Bengali (300M) | Self-supervised ASR | 17.8 % WER | 🤗 HF |
| Wav2Vec2-XLSR Bengali | XLSR fine-tune | — | 🤗 HF |
| BanglaConformer | Conformer ASR by Bengali.AI | — | 🤗 HF |
| BanglaASR | Whisper fine-tuned for Bengali | 14.73 % WER | 🤗 HF · 🔗 GitHub |
| Whisper (multilingual) | OpenAI base model — Bn supported | various sizes | 🤗 HF |
Word & Sentence Embeddings
| Resource | Description | Link |
|---|---|---|
| Bangla FastText (sagorsarker) | 20 M-token wiki-trained skipgram + CBOW | 🤗 HF |
| Bangla Word2Vec (sagorsarker) | 100-d Wikipedia embeddings | 🤗 HF |
| fastText 157-language Bengali | Facebook 300-d Wiki + CC | 🌐 fastText |
Spark NLP bengali_cc_300d | Production embedding | 🔗 Spark NLP |
| BanglaEmbed | Cross-lingual distilled sentence embeddings | 📄 arXiv 2411.15270 |
📚 Latest 2024–2026 Datasets
These are the most relevant new releases — cite the original authors when used.
| Dataset | Task | Size / Notes | Link |
|---|---|---|---|
| Bangla-Instruct | Instruction-tuning | 342 K instruction–response pairs | 🤗 HF |
| Bangla-TextBook | LM pretraining | 9.9 M tokens, 163 NCTB textbooks | 🤗 HF |
| BanglaSTEM | Technical-domain MT | 5 K Bn-En STEM sentence pairs | 📄 arXiv 2511.03498 |
| NCTB-QA | Educational QA | 87 805 QA pairs (grade 1–10) | 📄 arXiv 2603.05462 |
| BanglaQuAD | Open-domain QA | 30 808 QA pairs | 📄 arXiv 2410.10229 |
| ANCHOLIK-NER | Regional-dialect NER | 17 405 sentences, 5 regions | 📄 arXiv 2502.11198 |
| BanNERD (NAACL 2025) | NER, 10 classes / 29 domains | 85 K sentences, 991 K tokens | 🔗 GitHub |
| ONUBAD | Dialect→Standard MT | Chittagong / Sylhet / Barisal | 🔗 ScienceDirect |
| BanglaDial | Dialect text corpus | 60 729 entries × 11 dialects | 🔗 PMC |
| BIDWESH | Regional hate speech | Multi-region | 📄 arXiv 2507.16183 |
| BanglaTLit | Romanized→Bn back-transliteration | 42.7 K + 245.7 K pretrain | 📄 ACL 2024 |
| BanglishRev | E-commerce code-mix reviews | 1.74 M Daraz reviews | 📄 arXiv 2412.13161 |
| BengaliSent140 | Hate vs non-hate fusion | 140 K speeches | 📄 arXiv 2601.20129 · 🔗 IEEE DataPort |
| BLUCK | LLM cultural-knowledge benchmark | 2 366 MCQs / 23 categories | 📄 arXiv 2505.21092 |
| BNLI (refined) | NLI | Curated entail/contra/neutral | 📄 arXiv 2511.08813 |
| MultiBanAbs | Multi-domain abstractive sum. | Multi-corpus | 📄 arXiv 2511.19317 |
| MultiBanFakeDetect | Multimodal fake news | Text + image | 🔗 ScienceDirect |
| BanFakeNews-2.0 | Fake news (2024) | 47 K real + 13 K fake | 📊 Mendeley |
| BanglaHealth | Health-domain paraphrase | 200 K sentences | 🔗 ScienceDirect |
| BanglaCHQ-Summ | Consumer-health-question summary | 2 350 pairs (BLP-2023) | 🔗 GitHub |
| Bangla-MedER | Medical NER | 2 980 texts, 6 entity types | 📊 Mendeley |
| BanglaSarc3 | Sarcasm (ternary) | 12 089 FB comments | 🔗 ScienceDirect |
| VACASPATI | Bangla literature corpus | 11 M sentences / 115 M words | 📄 arXiv 2307.05083 |
| MixSarc | Code-mix sarcasm/humor/offence | Bn-En transliterated | 📄 arXiv 2602.21608 |
| EmoMix-3L | Code-mix emotion | 1 071 Bn-Hi-En instances | 🔗 GitHub |
| Bangla-ToCo | Context-aware toxic | 1 004 FB news comments | 🔗 ScienceDirect |
| BanglaDocAtlas | Document-layout, 8 classes | annotated complex docs | 🔗 IEEE |
| FoodBD | BD cuisine images, 67 categories | 3 523 polygon-annotated meals | 🔗 Springer 2025 |
| DeshiFoodBD | BD traditional food images | 5 425 images / 19 dishes | 🔗 Springer |
📊 Benchmarking and Evaluation
BLUB — Bangla Language Understanding Benchmark
The first comprehensive Bangla NLU benchmark, introduced with BanglaBERT (NAACL 2022).
| Task | Dataset | Metric | Best Model | Score |
|---|---|---|---|---|
| Sentiment Classification | SentNoB | Macro-F1 | BanglaBERT | 72.89 |
| Natural Language Inference | XNLI-bn / BNLI | Accuracy | BanglaBERT (Large) | 83.41 |
| Named Entity Recognition | MultiCoNER | Micro-F1 | BanglaBERT (Large) | 79.20 |
| Question Answering | SQuAD-bn / TyDiQA | EM / F1 | BanglaBERT (Large) | 76.10 / 81.50 |
📄 BLUB code & leaderboard: github.com/csebuetnlp/banglabert
BLUCK — Bangla LLM Cultural & Linguistic Benchmark (2025)
2 366 multiple-choice questions across 23 categories covering Bangladesh culture, history, and Bangla linguistics — designed to probe LLM cultural knowledge. 📄 arXiv 2505.21092
Recent Benchmark Datasets
| Dataset | Task | Size | Link |
|---|---|---|---|
| BanglaBook | Sentiment | 158 065 reviews | 🔗 GitHub |
| SentMix-3L / OffMix-3L | Code-mix sentiment / offence | ~1 K each | 📄 ACL |
| MultiCoNER (Bangla) | Multilingual complex NER | task | 🔗 multiconer.github.io |
📰 News, Corpora & Pretraining Data
| Dataset | Size | Link |
|---|---|---|
| Bangla2B+ (BanglaBERT pretraining corpus) | 27.5 GB / 110 sites | 🔗 GitHub |
| BanglaLM (data-mining corpus) | 14 GB | 📄 IEEE |
| BdNC – Bangladesh National Corpus ✅ | 40 GB / 3 B+ words | 🔗 corpus.bangla.gov.bd |
| VACASPATI literary corpus | 11 M sentences | 📄 arXiv 2307.05083 |
| CC-100 Bangla | 8.3 GB | 🔗 StatMT |
| OSCAR Bangla | 12 GB+ | 🔗 OSCAR |
| AI4Bharat IndicCorp | 9 B tokens incl. Bangla | 🔗 site |
| AI4Bharat IndicNLP corpus + catalog | meta-resource | 🔗 corpus · 🔗 catalog |
| Bangla Wikipedia Corpus (Kaggle) | wiki-text | 📊 Kaggle |
| Wikipedia bnwiki dumps | latest dumps | 🔗 dumps.wikimedia.org |
| Leipzig Bengali corpora (2021) | 1.65 M sentences | 🔗 corpora.uni-leipzig.de |
| Wiki Articles (Kaggle) | wiki snapshot | 📊 Kaggle |
| 40k News Articles | 40 K | 📊 Kaggle |
| Largest Bangla Newspaper | large multi-paper | 📊 Kaggle |
| bdNews24 corpus | bdnews24 articles | 📊 Kaggle |
| Bangladesh Protidin | Bangladesh Protidin news | 📊 Kaggle |
| csebuetnlp/xlsum | XL-Sum (Bangla subset) | 🤗 HF |
| csebuetnlp/dailydialogue_bn | Daily-dialogue translated | 🤗 HF |
| goru001/nlp-for-bengali | ULMFiT model + Wiki / news data | 🔗 GitHub |
| masiur/Bangla-Corpus | Open community corpus | 🔗 GitHub |
🔄 Machine Translation & Paraphrase
| Dataset | Description | Link |
|---|---|---|
| csebuetnlp/BanglaNMT | 2.38 M Bn-En pairs (133 MB) | 🤗 HF · 🔗 GitHub |
| AI4Bharat Samanantar | 49.6 M sentence pairs across Indic languages | 🤗 HF · 🔗 site |
| SUPara0.8M | Balanced En-Bn corpus | 🔗 IEEE DataPort |
| BanglaSTEM | 5 K STEM Bn-En pairs | 📄 arXiv 2511.03498 |
| WMT24 Bangla seed dataset | High-quality manual translation | 📄 ACL 2024 |
| BanglaParaphrase | 466 K paraphrase pairs (AACL 2022) | 🤗 HF · 🔗 GitHub |
| OPUS Collections | Multi-source parallel corpora | 🔗 OPUS |
| Bengali Visual Genome 1.0 | 29 K image-caption multimodal | 🔗 LINDAT |
| Google Dakshina | 12 South-Asian transliteration / parallel | 🔗 GitHub |
| BanglaTLit | Romanized → Bangla | 📄 ACL 2024 |
| bntranslit | Transliteration toolkit | 🔗 GitHub |
| Bengali Dictionary (Minhas Kamal) | Dictionary | 🔗 GitHub |
| TED2020 (Bangla) | TED multilingual | 📥 TSV |
🎤 Speech (ASR / TTS / Emotion)
ASR / Recognition
| Dataset | Description / Size | Link |
|---|---|---|
| OpenSLR-53 | Large Bengali ASR — 196 K utterances / 14.6 GB (Google) | 🔗 OpenSLR |
| OpenSLR (HF mirror) | All OpenSLR languages | 🤗 HF |
| OpenSLR-104 | Multilingual code-switching | 🔗 OpenSLR |
| Bengali Common Voice (Mozilla) | 399 h+ / 19 817 contributors (v9.0) | 🔗 Mozilla |
| Bengali.AI OOD-Speech | 1 177 h / 22 645 speakers — largest Bn ASR | 🔗 Bengali.AI |
| FLEURS Bangla | Cross-lingual 12 h | 🤗 HF |
| BanglaASR Dataset | Fine-tuned ASR | 🔗 GitHub |
| SUST-CSE-Speech / banspeech | TTS / ASR corpus | 🤗 HF |
| SKNahin / open-large-bengali-asr-data | Large open ASR | 🤗 HF |
| Bangla-Speech-Corpora (Bangla-Language-Processing) | Cleaned TTS-ready speech | 🔗 GitHub |
TTS
| Dataset | Description | Link |
|---|---|---|
| OpenSLR-37 | High-quality Google Bengali TTS | 🔗 OpenSLR |
| Bengali.AI TTS dataset | Studio-quality TTS | 🔗 site |
| bangla-tts (zabir-nabil) | Real-time multilingual synthesis | 🔗 GitHub |
Speech Emotion
| Dataset | Description | Link |
|---|---|---|
| SUBESCO | 7 K utterances / 7 emotions, gender-balanced | 📄 PLOS One |
| BanglaSER | 1 467 utterances / 34 speakers / 5 emotions | 🔗 ScienceDirect |
| KBES | Realistic speech-emotion w/ intensity | 🔗 ScienceDirect |
| BANSpEmo | Bangla emotional speech | 📄 arXiv 2312.14020 |
Speech-to-Text Toolkits
| Tool | Notes | Link |
|---|---|---|
| BanglaSpeech2Text | Whisper-FT offline ASR (mp3/mp4/wav) | 🔗 GitHub |
😊 Sentiment / Emotion / Sarcasm
| Dataset | Size / Notes | Link |
|---|---|---|
| BanglaBook | 158 K book reviews | 🔗 GitHub |
| SentNoB | Noisy social-media sentiment | 📊 Kaggle |
| EmoNoBa | 22 698 comments / 6 emotions (AACL 2022) | 📊 Kaggle · 📄 ACL |
| BanglaEmotion (shaoncsecu) | Emotion benchmark | 🔗 GitHub |
| MONOVAB | Multi-label emotion | 📄 paper |
| BAN-ABSA / BANGLA-ABSA | 9 009 aspect-level comments | 📄 arXiv 2012.00288 · 📊 Mendeley |
| BanglaSenti (lexicon) | 61 582 polarity words | 🔗 GitHub |
| banglanlp/bangla-sentiment-classification | Compiled benchmarking sets | 🔗 GitHub |
| Ayubur sentiment datasets | Multiple legacy datasets | 🔗 GitHub |
| BanglaSarc | 5 112 sarcasm samples | 📊 Kaggle |
| BanglaSarc3 | 12 089 ternary-class sarcasm | 🔗 ScienceDirect |
| BnSentMix | 20 K Bn-En code-mix sentiment | 📄 paper |
| SentMix-3L / OffMix-3L | Bn-En-Hi code-mix | 📄 ACL |
| Drama Review | Bengali drama reviews | 📊 Figshare |
| YouTube Sentiment / Emotion | YT comments | 📊 Kaggle |
| News Comments Sentiment | Bn news comments | 📊 Kaggle |
| News Headline Categories | Headline classification | 📊 Kaggle |
| Big News Classification | Large news classifier | 📊 Kaggle |
| News Article Classification (IndicNLP) | Indic news classifier | 📊 Kaggle |
| Bengali-Banglish Emotion | Mixed-script | 📊 Mendeley |
| E-commerce Sentiment + Emotion | 78 130 Daraz / Pickaboo reviews | 🔗 ScienceDirect |
| BanglishRev | 1.74 M Daraz code-mix reviews | 📄 arXiv 2412.13161 |
🛡️ Hate Speech / Toxic / Cyberbullying
| Dataset | Size / Notes | Link |
|---|---|---|
| Bengali-Hate-Speech (rezacsedu) | 6 418 / 5 categories | 🔗 GitHub · 📊 UCI |
| Bengali Hate Speech (naurosromim) | Annotated hate dataset | 📊 Kaggle |
| BIDWESH | Regional-based hate speech | 📄 arXiv 2507.16183 |
| BengaliSent140 | 140 K hate vs non-hate | 🔗 IEEE DataPort |
| ToxLex_bn | 1 959 bigrams from 2.2 M FB comments | 📊 Mendeley |
| Multi-Labeled Bengali Toxic | 16 073 / 7 labels | 🔗 GitHub |
| Bangla-ToCo | 1 004 context-aware toxic | 🔗 ScienceDirect |
| Bangla Multilabel Cyberbully | 12 557 (5 classes) | 📊 Mendeley |
| Bangla Social-Media Cyberbullying | YT/FB/IG/TikTok | 🔗 IEEE DataPort |
| Code-mixed Chaos (Banglish toxic) | 10 234 multi-label | 📊 Mendeley |
| BanglaMedia | 7 725 YT comments — 10 topics, 4 sentiments | 📊 Mendeley |
| VITD (BLP-2023 violence) | Violence-inciting text — 3 classes | 🔗 BLP |
🕵️ Fake News & Misinformation
| Dataset | Size | Link |
|---|---|---|
| BanFakeNews (LREC 2020) | 50 K news | 📊 Kaggle · 📄 arXiv 2004.08789 |
| BanFakeNews-2.0 (2024) | 47 K real + 13 K fake | 📊 Mendeley |
| MultiBanFakeDetect | Multimodal text + image | 🔗 ScienceDirect |
| Rowan1224/FakeNews | Code & data | 🔗 GitHub |
| DataCOVID19 | 14 571 COVID misinformation | 🔗 Springer |
🏷️ NER / POS / Parsing
| Dataset | Size / Notes | Link |
|---|---|---|
| B-NER (IEEE Access) | 22 144 sentences | 🔗 IEEE |
| BanNERD (NAACL 2025) | 85 K sentences / 991 K tokens / 10 classes / 29 domains | 🔗 GitHub |
| NER-Bangla-Dataset (MISabic) | 70 K sentences / 5 types | 🔗 GitHub |
| bnlp-resources NER | Train/dev/test splits | 🌐 banglanlp.github.io |
| ANCHOLIK-NER | Regional NER, 5 regions | 📄 arXiv 2502.11198 |
| celloscope_bangla_ner_dataset | 319 K NER | 🤗 HF |
| celloscope_bangla_ner_dataset (small) | 6.57 K | 🤗 HF |
| Bangla-MedER | Medical NER 2 980 / 6 types | 📊 Mendeley |
| Bangla NER (towhidahmedfoysal) | 400 K word-level | 📊 Kaggle |
| POS — 3 K sentences | abhishekgupta92 | 🔗 GitHub |
| POS — 100 K+ words (towhidahmedfoysal) | Word-level POS | 📊 Kaggle |
| UD_Bengali-BRU treebank | 14 UPOS tags, UD v2.9+ | 🌐 universaldependencies.org |
❓ Question Answering
| Dataset | Size | Link |
|---|---|---|
| csebuetnlp/squad_bn | 118 K train QA | 🤗 HF |
| BanglaRQA | 14 889 QA / 3 K passages (EMNLP 2022) | 🔗 GitHub |
| Bengali QA (Mayeesha) | SQuAD 2.0–style Bn QA | 📊 Kaggle |
| BanglaQuAD | 30 808 QA pairs | 📄 arXiv 2410.10229 |
| NCTB-QA | 87 805 educational QA | 📄 arXiv 2603.05462 |
| doctor_qa_bangla | 5.14 K medical QA | 🤗 HF |
| TyDiQA (Bengali subset) | Cross-lingual QA | 🤗 HF |
📝 Text Summarization
| Dataset | Size / Notes | Link |
|---|---|---|
| csebuetnlp/xlsum | XL-Sum (Bangla subset, BBC) | 🤗 HF |
| BanglaCHQ-Summ | 2 350 health-question summary pairs | 🔗 GitHub |
| MultiBanAbs | Multi-domain abstractive | 📄 arXiv 2511.19317 |
| BNLPC + NCTB | EACL 2021 unsupervised abstractive | 🔗 GitHub |
| BANSData (Prithwiraj) | News abstractive | 📊 Kaggle |
| Bengali Text Summarization (Hasan Moni) | Extractive + abstractive | 📊 Kaggle |
| BUSUM-BNLP | Multi-document update summarization | 📊 Kaggle |
| bnSum_gemma7b-it | Gemma-7B inst news summary system | 🔗 GitHub |
| Bangla-Text-summarization-Dataset (Abid) | Extractive | 🔗 GitHub |
| 3 Human-Evaluated articles (BNLPC) | Reference summaries | 🌐 BNLPC |
🖊️ OCR, Handwriting & Document Layout
| Dataset | Size / Notes | Link |
|---|---|---|
| NumtaDB | 85 K handwritten Bn-digit images | 📊 Kaggle |
| Bengali.AI Handwritten Grapheme Classification | Kaggle competition | 📊 Kaggle |
| Ekush | Handwritten Bn characters (largest) | 🌐 rabby.dev/ekush · 📊 Kaggle |
| BN-HTRd | 788 pages / 150 writers / 108 K words (HTR) | 📊 Mendeley |
| BanglaWriting | Multi-purpose offline HTR | 📄 paper |
| Bayanno | Multi-purpose handwriting | 📊 Mendeley |
| Bongabdo | Bn handwritten script | 📄 arXiv 2101.00204 |
| CMATERdb (1 / 2.1.2) | 5 K word imgs / 18 K Bn city names | 🔗 site |
| BaDLAD | 33 695 layout samples / 6 domains | 📄 paper |
| BanglaDocAtlas | 8-class complex Bn document layout | 🔗 IEEE |
| DL Sprint 2.0 (BUET CSE Fest 2023) | Layout segmentation | 📊 Kaggle |
| Bangla License Plate 2.5 K | 2 519 plate images | 🔗 Zenodo |
| BD-ALPDR | 725 high-res LP images | 🌐 site |
| Govt. Bangla OCR service | Free Bangla OCR | 🌐 ocr.bangla.gov.bd |
✋ Sign Language & Multimodal Vision
| Dataset | Size / Notes | Link |
|---|---|---|
| BdSLW60 | 60 sign words / 9 307 video trials | 📄 arXiv 2402.08635 |
| KU-BdSL | 30 classes / 38 consonants | 📊 Mendeley |
| BAUST Lipi | 18 K imgs / 36 alphabets | 📄 arXiv 2408.10518 |
| BDSL 49 | 29 490 imgs / 49 labels | 📄 arXiv 2208.06827 |
| BdSL36 | 4 M+ imgs / 36 cats | 📄 paper |
| Ego-SLD | Egocentric Bn sign-language video | 🔗 IEEE DataPort |
| Bengali Visual Genome 1.0 | Multimodal MT + captioning | 🔗 LINDAT |
| BAN-Cap | English-Bangla image-description | 📄 arXiv 2205.14462 |
| BNATURE | Bn image captioning | 📊 Kaggle |
| BanglaView | 31 783 imgs / 158 K captions | 📊 Mendeley |
| csebuetnlp/illusionVQA | Bn-aware VQA | 🤗 HF |
| DeshiFoodBD | 5 425 BD-cuisine images / 19 dishes | 🔗 Springer |
| FoodBD | 3 523 polygon-annotated meals (2025) | 🔗 Springer |
| BnLiT — Bangla Image-to-Text | Natural-language image-text | 📊 Kaggle |
🗺️ Regional Dialects
| Dataset | Coverage | Link |
|---|---|---|
| ANCHOLIK-NER | Barishal, Chittagong, Mymensingh, Noakhali, Sylhet | 📄 arXiv 2502.11198 |
| ONUBAD | Chittagong / Sylhet / Barisal → Standard Bn | 🔗 ScienceDirect |
| BanglaDial | 11 dialects / 60 729 entries | 🔗 PMC |
| BIDWESH | Regional hate speech | 📄 arXiv 2507.16183 |
| BanglaCHQ-Summ + dialect benchmarks | Sylheti / Chittagonian | 📄 ACL 2025 |
| Sylheti → Standard NMT | Sylheti corpus | 🔗 ScienceDirect |
🧪 NLI / Bias / Misc
| Dataset | Description | Link |
|---|---|---|
| csebuetnlp/xnli_bn | Bangla NLI translated from XNLI | 🤗 HF |
| BNLI (refined) | Curated entail/contra/neutral | 📄 arXiv 2511.08813 |
| csebuetnlp/BanglaSocialBias | Social-bias evaluation | 🤗 HF |
| csebuetnlp/BanglaContextualBias | Contextual-bias evaluation | 🤗 HF |
| csebuetnlp/CrossSum | Cross-lingual summarization | 🤗 HF |
| stopwords-iso/stopwords-bn | Stopword list | 🔗 GitHub |
| Bangla Plagiarism Dataset | 59.9 K | 🤗 HF |
| Banking 14-intents (en + bn + banglish) | 16.5 K intent samples | 🤗 HF |
| Massive intent (bn-BD) | 16.5 K | 🤗 HF |
| BanglaMusicStylo | 2 824 lyrics / 211 lyricists | 🔗 IEEE DataPort |
| Bangla Song Lyrics (genres + artists) | Bn song-lyric corpus | 📊 Kaggle |
| Bn Numbers w/ Words | Number-name dataset | 📊 Kaggle |
| likhonsheikh/BanglaNLP | 120 K parallel news pairs | 🤗 HF |
🔧 NLP Tools & Libraries
Python Libraries
| Library | Description | Link |
|---|---|---|
| BNLP (sagorbrur) | Comprehensive Bengali NLP toolkit | 🔗 GitHub · pip install bnlp_toolkit |
| BNLTK | Bangla NLP toolkit (tokenize / stem / POS) | 🔗 GitHub |
| sbnltk | This repo's toolkit (sentiment / NER / POS / sum) | 🔗 GitHub |
| bangla-stemmer | Lightweight Bn stemmer | 🔗 PyPI |
| bnunicode | Bijoy → Unicode normalization | 🔗 GitHub |
| Indic NLP Library | Multi-Indic processing / transliteration | 🔗 GitHub |
| bntranslit | Bengali transliteration | 🔗 GitHub |
| BanglaSpeech2Text | Bangla offline ASR | 🔗 GitHub |
| bangla-tts | Bangla TTS library | 🔗 GitHub |
| BanglaKit organisation | Tools, datasets, resources | 🔗 GitHub |
OCR / Vision
| Tool | Notes | Link |
|---|---|---|
| EasyOCR | Built-in Bangla support | 🔗 GitHub |
| Tesseract OCR | ben traineddata available | 🔗 GitHub |
| ocr.bangla.gov.bd | Govt-hosted Bangla OCR service | 🌐 site |
📄 Research Papers
Foundational
- BanglaBERT (NAACL Findings 2022) — 📄 ACL · 📄 arXiv 2101.00204
- BanglaNLG / BanglaT5 (EACL 2023) — 📄 arXiv 2205.11081
- BanglaParaphrase (AACL 2022) — 📄 arXiv 2210.05109
- Not Low-Resource Anymore: BanglaNMT (EMNLP 2020) — 📄 ACL
- MuRIL (Indic multilingual) — 📄 arXiv 2103.10730
- Bangla NLP Comprehensive Analysis — 📄 arXiv 2105.14875
LLMs & Generation (2024–2025)
- TigerLLM (ACL 2025) — 📄 arXiv 2503.10995
- TituLLMs — 📄 arXiv 2502.11187
- BanglaByT5 — 📄 arXiv 2505.17102
- BanglaLlama — 📄 arXiv 2410.21200
- WMT24 Bangla seed — 📄 ACL
QA / Reading Comprehension
- BanglaRQA (EMNLP 2022 Findings) — 📄 ACL
- BanglaQuAD — 📄 arXiv 2410.10229
- NCTB-QA — 📄 arXiv 2603.05462
NER
- B-NER (IEEE Access) — 🔗 IEEE
- ANCHOLIK-NER — 📄 arXiv 2502.11198
- BanNERD (NAACL 2025) — 🔗 GitHub
Summarization
- BanglaCHQ-Summ (BLP 2023) — 📄 ACL
- Bangla extractive (IEEE) — 🔗 IEEE
- MultiBanAbs — 📄 arXiv 2511.19317
Speech
- BanglaDialecto (2024) — 📄 arXiv 2411.10879
- Bengali Common Voice ASR — 📄 arXiv 2206.14053
- Bengali ASR system survey — 📄 arXiv 2209.08119
- SUBESCO (PLOS One 2021) — 📄 PLOS
Workshops
- BLP-2023 @ EMNLP (Singapore) — 🌐 blp-workshop.github.io · 📚 ACL Anthology
- BLP-2025 @ AACL-IJCNLP (Mumbai) — 🌐 blp-workshop.github.io · shared tasks: Bangla code generation, sentiment, etc.
🇧🇩 Bangladesh Government Resources
বাংলাদেশ সরকারের তথ্যপ্রযুক্তি ও বাংলা ভাষাগত সম্পদ:
| Resource | বাংলা নাম | Link |
|---|---|---|
| Bangladesh National Corpus (BdNC) | বাংলাদেশ জাতীয় কর্পাস (৪০ GB / ৩B+ শব্দ) | 🌐 corpus.bangla.gov.bd |
| Govt. Bangla OCR | সরকারি বাংলা OCR সেবা | 🌐 ocr.bangla.gov.bd |
| Accessible Dictionary | প্রতিবন্ধী-বান্ধব বাংলা অভিধান | 🌐 accessibledictionary.gov.bd |
| EBLICT Project | তথ্যপ্রযুক্তিতে বাংলা ভাষা সমৃদ্ধকরণ প্রকল্প | 🌐 eblict.portal.gov.bd |
| IPA — Information Processing Authority | বাংলা ভাষা প্রক্রিয়াকরণ কর্তৃপক্ষ | 🌐 ipa.bangla.gov.bd |
| Bangladesh Computer Council (BCC) | বাংলাদেশ কম্পিউটার কাউন্সিল | 🌐 bcc.gov.bd |
| Bangla Academy | বাংলা একাডেমি | 🌐 banglaacademy.gov.bd |
🔗 Curated Aggregator Lists (start here)
| Resource | Maintainer | Link |
|---|---|---|
| bnlp-resources | banglanlp | 🔗 GitHub · 🌐 site |
| awesome-bangla | banglakit | 🔗 GitHub |
| bangla-corpus | sagorbrur | 🔗 GitHub |
| Awesome_Bangla_Datasets | sabbirhossainujjal | 🔗 GitHub |
| AI4Bharat Indic NLP catalog | AI4Bharat | 🔗 GitHub |
| Mahadih534 — Bangla NLP HF collection | Mahadih534 | 🤗 HF |
| Mahadih534 — Bangla TTS collection | Mahadih534 | 🤗 HF |
| Mahadih534 — Bangla LLM finetuning | Mahadih534 | 🤗 HF |
| csebuetnlp organization | BUET CSE NLP | 🤗 HF |
| Sudipta Kar Bangla NLP resources | personal | 🌐 site |
| হাতেকলমে Bangla NLP (Rakibul Hassan) | educational notebooks | 🔗 GitHub · 🌐 book |
| GitHub topic: bangla-nlp | community-tagged | 🌐 topic |
| GitHub topic: bengali-nlp | community-tagged | 🌐 topic |
| Awesome Public Datasets (general NLP) | community | 🔗 GitHub |
| NLP Datasets Collection | community | 🔗 GitHub |
💡 Motivation & Contribution
Bangla is the 6th-most-spoken language in the world (~270 million native speakers) but remains classified as low-resource in NLP. This repository exists to make every Bangla NLP resource one click away — accurately documented and free of fabricated links.
How to Get Started
- For Pre-trained Models — visit the HuggingFace links above and load directly with
transformers. - For Tools —
pip install bnlp_toolkitorpip install bnltk. - For Datasets — follow individual links; honor each dataset's license (most are CC BY-NC-SA 4.0).
- For Research — start with the BLP workshop proceedings and the latest 2024–2026 release table.
Contributing
- 📝 Submit new datasets via pull request — include a working URL and one-line description.
- 🐛 Report broken / fabricated links by opening an issue.
- 🔬 Tag your own paper to share with the community.
- ✅ Every PR is welcome — please verify URLs return 200 before submitting.
⭐ If this repository helps your work, please give it a star! ⭐
🤝 Contributions, corrections, and additions are very welcome.
🌟 Thanks to every researcher and developer pushing Bangla NLP forward.
☕ Support This Project
If this resource has been helpful, you can support its maintenance:
bengali-dataset
dataset
nlp
question-answering
sentiment-analysis
text-classification
text-summarization