UPDESH: Synthesizing Grounded Instruction Tuning Data for 13 Indic Languages
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866918357846982656 |
|---|---|
| author | Chitale, Pranjal A. Gumma, Varun Ahuja, Sanchit Kodali, Prashant Uppadhyay, Manan Sudharsan, Deepthi Sitaram, Sunayana |
| author_facet | Chitale, Pranjal A. Gumma, Varun Ahuja, Sanchit Kodali, Prashant Uppadhyay, Manan Sudharsan, Deepthi Sitaram, Sunayana |
| contents | Developing culturally grounded multilingual AI systems remains challenging, particularly for low-resource languages. While synthetic data offers promise, its effectiveness in multilingual and multicultural contexts is underexplored. We investigate bottom-up synthetic data generation using large open-source LLMs (>= 235B parameters) grounded in language-specific Wikipedia content, complementing dominant top-down translation-based approaches from English. We introduce Updesh, a high-quality large-scale synthetic instruction-following dataset comprising 9.5M data points across 13 Indian languages and English, encompassing diverse reasoning and generative tasks. Comprehensive evaluation using automated metrics and 10K human assessments confirms high data quality. Downstream evaluations performed by fine-tuning models on various datasets and assessing performance across 13 diverse multilingual datasets and model comparative evaluations, demonstrate that models trained on Updesh consistently obtain significant improvements on NLU, NLG evaluations. Finally, through ablation studies and cultural evaluations, we show that context-aware, culturally grounded data generation is essential for effective multilingual AI development. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_21294 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | UPDESH: Synthesizing Grounded Instruction Tuning Data for 13 Indic Languages Chitale, Pranjal A. Gumma, Varun Ahuja, Sanchit Kodali, Prashant Uppadhyay, Manan Sudharsan, Deepthi Sitaram, Sunayana Computation and Language Developing culturally grounded multilingual AI systems remains challenging, particularly for low-resource languages. While synthetic data offers promise, its effectiveness in multilingual and multicultural contexts is underexplored. We investigate bottom-up synthetic data generation using large open-source LLMs (>= 235B parameters) grounded in language-specific Wikipedia content, complementing dominant top-down translation-based approaches from English. We introduce Updesh, a high-quality large-scale synthetic instruction-following dataset comprising 9.5M data points across 13 Indian languages and English, encompassing diverse reasoning and generative tasks. Comprehensive evaluation using automated metrics and 10K human assessments confirms high data quality. Downstream evaluations performed by fine-tuning models on various datasets and assessing performance across 13 diverse multilingual datasets and model comparative evaluations, demonstrate that models trained on Updesh consistently obtain significant improvements on NLU, NLG evaluations. Finally, through ablation studies and cultural evaluations, we show that context-aware, culturally grounded data generation is essential for effective multilingual AI development. |
| title | UPDESH: Synthesizing Grounded Instruction Tuning Data for 13 Indic Languages |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2509.21294 |