datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
CodeMixBench
ℹ️Dataset Card for CodeMixBench
[EMNLP'25] CodeMixBench: Evaluating Code-Mixing Capabilities of LLMs Across 18 Languages
Code-mixing is a linguistic phenomenon where multilingual speakers switch or mix two or more languages within a single utterance or conversation.
To evaluate LLMs’ comprehension of multilingual code-mixed texts, we introduce CodeMixBench, a benchmark comprising eight tasks across 18 languages.… See the full description on the dataset page: https://huggingface.co/datasets/CodeMixBench/CodeMixBench.hindi-english-code-mixed-tweets-sentimentCodeMixBench
ℹ️Dataset Card for CodeMixBench
[EMNLP'25] CodeMixBench: Evaluating Code-Mixing Capabilities of LLMs Across 18 Languages
Code-mixing is a linguistic phenomenon where multilingual speakers switch or mix two or more languages within a single utterance or conversation.
To evaluate LLMs’ comprehension of multilingual code-mixed texts, we introduce CodeMixBench, a benchmark comprising eight tasks across 18 languages.… See the full description on the dataset page: https://huggingface.co/datasets/Tanushreeeeee/CodeMixBench.Code-Mixed-Sentiment-Analysis-Dataset
Dataset Generation:
Initially, we select the Amazon Review Dataset as our base data, referenced from Ni et al. (2019)[^1]. We randomly extract 100,000 instances from this dataset. The original labels in this dataset are ratings, scaled from 1 to 5. For our specific task, we categorize them into Positive (rating > 3), Neutral (rating = 3), and Negative (rating < 3), ensuring a balanced number of instances for each label. To generate the synthetic Code-mixed dataset, we apply two… See the full description on the dataset page: https://huggingface.co/datasets/md-nishat-008/Code-Mixed-Sentiment-Analysis-Dataset.Code-Mixed-Offensive-Language-Detection-Dataset
Code-Mixed-Offensive-Language-Identification
This is a dataset for the offensive language detection task. It contains 100k code mixed data. The languages are Bangla-English-Hindi.
Dataset Generation:
Initially, the labelling schema of OLID[^1] and SOLID[^2] serves as the seed data, from which we randomly select 100,000 data instances. The labels in this dataset are categorized as Non-Offensive and Offensive for the purpose of our task. We meticulously ensure an equal… See the full description on the dataset page: https://huggingface.co/datasets/md-nishat-008/Code-Mixed-Offensive-Language-Detection-Dataset.codemixed-ind-classificationEn-Bn-Code-Mixed-Two-Class-Sentiment-Dataset📊 En-Bn-Code-Mixed-Two-Class-Sentiment-Dataset
The En-Bn-Code-Mixed-Two-Class-Sentiment-Dataset is a multilingual dataset of 100,000 product review texts designed for code-mixed sentiment analysis involving English, Bengali, and Roman Bengali.
Each record includes:
🆔 Id
🛒 ProductId
💬 Code-Mixed-Text
💡 Sentiment
The dataset captures diverse linguistic styles, authentic code-mixing, and real-world sentiment patterns from multilingual digital communication.
🌐 Text Distribution
The dataset… See the full description on the dataset page: https://huggingface.co/datasets/DaliaBarua/En-Bn-Code-Mixed-Two-Class-Sentiment-Dataset.gujarati-english-codemixed-sentiment
Gujarati-English / Hindi-English Code-Mixed Sentiment Dataset
Dataset Summary
This dataset addresses an empirically confirmed gap in the code-mixed
sentiment analysis literature: a systematic literature review of 48
retrieved records (39 core studies) found zero existing
Gujarati-English code-mixed sentiment datasets, despite Gujarati having
over 55 million native speakers. This dataset provides the first
Gujarati-English resource of this kind, alongside… See the full description on the dataset page: https://huggingface.co/datasets/ShrutiPatel3011/gujarati-english-codemixed-sentiment.Code_Mixed_video_ComplaintCodemix_tamil_englishArzEn-CodeMixed
ArzEn-CodeMixed: English-Arabic Intra-Sentential Code-Mixed Translation Dataset
This dataset contains aligned English, Arabic, and English-Arabic as well as Arabic-English intra-sentential code-mixed sentences, created for evaluating large LLMs on translating code-mixed text. It is constructed from the ArzEn-MultiGenre parallel corpus and enhanced with carefully prompted and human-validated code-mixed variants.
Dataset Details
Instances: Each entry contains:… See the full description on the dataset page: https://huggingface.co/datasets/taha-alnasser/ArzEn-CodeMixed.Advanced_CodeMix_Normalization_Dataset_India
Evaluation & Benchmarking
To validate dataset usefulness, normalization accuracy can be evaluated using:
Exact Match Accuracy
BLEU Score for text similarity
Human evaluation for real-world correctness
This dataset is designed to improve performance of multilingual NLP systems in handling noisy, code-mixed Indian queries.
Data Transformation Approach
The dataset was created by transforming real-world code-mixed queries into structured English. Variations include:… See the full description on the dataset page: https://huggingface.co/datasets/karanverma19/Advanced_CodeMix_Normalization_Dataset_India.code-mix-thai-engCodeMix_Query_Normalization_India
CodeMix Query Normalization (India)
Overview
This dataset contains code-mixed user queries from Indian contexts, primarily in Hinglish and Punjabi, normalized into clean English. It reflects how users naturally communicate in real-world scenarios by mixing local languages with English.
Features
100 high-quality samples
Code-mixed queries (Hinglish, Punjabi)
Clean normalized English outputs
Real-world, informal user language patterns
Covers domains such as… See the full description on the dataset page: https://huggingface.co/datasets/karanverma19/CodeMix_Query_Normalization_India.English_French_safety_code-mixing_datasetUsing HarmBench promtps as the English baselines
codemixed-synthetic-sarc-11kcode-mixing_safety