--- license: apache-2.0 language: - ar - en task_categories: - text-retrieval - sentence-similarity pretty_name: Arabic Dialect-to-MSA Embedding Training Data --- # Arabic Dialect-to-MSA Embedding Dataset ## Overview This dataset is designed to fine-tune embedding models for retrieving Modern Standard Arabic (MSA) passages from queries written in MSA or Arabic dialects. Each original MSA query has five dialect variants: Saudi, Syrian, Egyptian, Sudanese, and Tunisian. The dataset uses one positive MSA passage and nine hard negative passages for contrastive training. ## Dataset Structure | Split | Number of rows | |---|---:| | Train | 9,901 | | Validation | 1,238 | | **Total** | **11,139** | Each split contains the following columns: - `query`: Original MSA query. - `positive`: Relevant MSA passage. - `negative1` to `negative9`: Nine hard negative MSA passages. - `saudi_variant`: Saudi Arabic query. - `syrian_variant`: Syrian Arabic query. - `egyptian_variant`: Egyptian Arabic query. - `sudanese_variant`: Sudanese Arabic query. - `tunisian_variant`: Tunisian Arabic query. ## Data Generation The original queries are written in MSA. Dialect variants are generated using a two-stage translation pipeline: 1. MSA to English using `facebook/nllb-200-1.3B`. 2. English to Arabic dialect using `oddadmix/Jisr-MT-50M-AllDialects`. ## Intended Use This dataset supports contrastive fine-tuning of embedding models for Arabic semantic search, cross-dialect retrieval, and MSA document retrieval. ## Limitations Dialect variants are machine-translated and may not cover all natural dialect expressions. Only five Arabic dialects are included. ## Related Resources - [Fine-tuned model](https://huggingface.co/OussamaBenSlama/e5-small-dialect2MSA) - [Test dataset](https://huggingface.co/datasets/OussamaBenSlama/e5-dialect2MSA-test-embedding-data) - [Base model](https://huggingface.co/intfloat/multilingual-e5-small) - [Arabic Triplet With Multi-Negatives](https://huggingface.co/datasets/NAMAA-Space/Arabic-Triplet-With-Multi-Negatives) - [Mr. TyDi](https://huggingface.co/datasets/castorini/mr-tydi) ## License Apache-2.0.