Explore high-quality datasets for your AI and machine learning projects.
We provide a Hindi‑Telugu parallel corpus across various technical domains (natural sciences, computer science, law, healthcare, and general domain). The corpus contains 700 K parallel sentences, of which 535 K were created through extraction, alignment, manual translation, iterative back‑translation with post‑editing, and 165 K were collected from the public domain. We report comparative evaluations of the corpus’s representativeness and diversity. The corpus is pre‑processed for machine translation; we trained a neural MT system and reported state‑of‑the‑art baselines on several domains and benchmarks. This defines a new task for domain‑specific MT for low‑resource language pairs such as Hindi‑Telugu. The 535 K curated corpus is freely available for non‑commercial research and is, to our knowledge, the largest, carefully curated, publicly available Hindi‑Telugu domain parallel corpus.
