Large language models remain predominantly English-centric, which limits their utility for underrepresented languages. We help bridge this gap for Hindi with Llama-3-Nanda-10B-Chat (aka Nanda-10B) and Llama-3.1-Nanda-87B-Chat (aka Nanda-87B), forming the Nanda family of open-weight bilingual models (https://github.com/MBZUAI-IFM/Nanda-Family). Our approach integrates: (i) a tokenizer extending Llama’s vocabulary with 20% Hindi-specific tokens, thus halving Hindi tokenization fertility while preserving English efficiency, (ii) Hindi-first parameter-efficient continual pretraining using Llama Pro on a 65B-token corpus spanning Devanagari script, code-mixed, and Romanized Hindi, and (iii) bilingual instruction and safety alignment on a large culturally grounded dataset. The resulting Nanda models outperform open-weight LLMs of comparable size: Nanda-87B yields high generative quality, and Nanda-10B shows competitive general-purpose performance. Nanda-87B demonstrates state-of-the-art performance on summarization, translation, transliteration, and instruction following. Moreover, both models achieve state-of-the-art performance in safety and in cultural knowledge. Our results demonstrate that careful tokenizer design, data curation, and continual pretraining can yield capable and safe LLMs for resource-poor languages without compromising English performance.

Do large language models (LLMs) model linguistic variation? We investigate this question through Hindi-English (Hinglish) verb code-mixing, where speakers can use either a Hindi verb or an English verb with the light verb karna (’do’). Both forms are grammatical, but speakers show unexplained variation in language choice for the verb. We compare human preferences on controlled code-mixed minimal pairs to LLM perplexities spanning families, sizes, and training language compositions. We find that current LLMs do not reliably classify verb language preferences to match native speaker judgments. We also see that with specific supervision, some models do predict human preference to an extent. We release native speaker acceptability judgments on 30 verb pairs, perplexity ratios for 4,279 verb pairs across 7 models, and experimental materials.

We present CoPara, the first publicly available paragraph-level (n-way aligned) multilingual parallel corpora for Dravidian languages. The collection contains 2856 paragraph/passage pairs between English and four Dravidian languages. We source the parallel paragraphs from the New India Samachar magazine and align them with English as a pivot language. We do human and artificial evaluations to validate the high-quality alignment and richness of the parallel paragraphs of a range of lengths. To show one of the many ways this dataset can be wielded, we finetuned IndicBART, a seq2seq NMT model on all XX-En pairs of languages in CoPara which perform better than existing sentence-level models on standard benchmarks (like BLEU) on sentence level translations and longer text too. We show how this dataset can enrich a model trained for a task like this, with more contextual cues and beyond sentence understanding even in low-resource settings like that of Dravidian languages. Finally, the dataset and models are made available publicly at CoPara to help advance research in Dravidian NLP, parallel multilingual, and beyond sentence-level tasks like NMT, etc.

Data augmentation is an important method for evaluating the robustness of and enhancing the diversity of training data for natural language processing (NLP) models. In this paper, we present NL-Augmenter, a new participatory Python-based natural language (NL) augmentation framework which supports the creation of transformations (modifications to the data) and filters (data splits according to specific features). We describe the framework and an initial set of 117 transformations and 23 filters for a variety of NL tasks annotated with noisy descriptive tags. The transformations incorporate noise, intentional and accidental human mistakes, socio-linguistic variation, semantically-valid style, syntax changes, as well as artificial constructs that are unambiguous to humans. We demonstrate the efficacy of NL-Augmenter by using its transformations to analyze the robustness of popular language models. We find different models to be differently challenged on different tasks, with quasi-systematic score decreases. The infrastructure, datacards, and robustness evaluation results are publicly available on GitHub for the benefit of researchers working on paraphrase generation, robustness analysis, and low-resource NLP.